ProgrammableMarketing/arena.ai (opens in a new tab)
PDF

Views

  • Plan
  • Roadmap
  • Research

On this page

  • The brief
  • Start here
  • Idea portfolio
  • Effort vs impact
  • This week · 6
  • This quarter · 11
  • Big bets · 3
  • Wildcards · 2
  • How we can help
Generated Oct 23, 2026
  1. Programmable
  2. Playbooks
  3. Arena
Arena logo

growth playbook · Oct 23, 2026

arena.ai (opens in a new tab)

Arena growth playbook

Diagnosis

Arena has the audience (10M+ monthly visitors, 82M+ votes), the revenue ($100M run rate within eight months) and the capital ($200M Series B at $3.1B), and it is now hiring senior growth marketers, a marketing engineer, product GMs for growth and enterprise, and a GTM strategy lead.

The website has not caught up with the business: there is no page for the enterprise offering, missing pages answer with status 200, old links break, the About page shows last round's numbers, and real visitors fail Core Web Vitals.

This playbook gives the enterprise side a front door first, fixes the site basics and adds leaderboard alerts, then builds model pages, AEO, activation and personal leaderboards, a self-serve evaluation tier, co-launch pages and occupational reports, with private arenas, a Max API and embeddable leaderboards as big bets.

Growth Score

8/10

Biggest Opportunity

Turn the public audience into owned channels and a visible path to paid evaluations: an Evaluations section, model pages and leaderboard alerts.

Biggest Risk

Arena's value depends on being seen as neutral; commercial growth with labs must stay visibly separate from the rankings, while rivals such as Artificial Analysis and Scale publish their own leaderboards.

01

Start here

The three moves to make first. Each one is written up in full below.

  1. 1

    Give the $100M enterprise business a front door: arena.ai has no page for labs and enterprises to buy

    Draft the three page outlines with the solutions engineers and agree the form fields with the GTM lead.

    High impact First result in 1 month
  2. 2

    Get real visitors under the Core Web Vitals lines: LCP is 4.2 seconds and INP 376 ms

    Profile the first prompt and the vote click on a mid-range phone.

    High impact First result in 1 month
  3. 3

    Make missing pages return 404 and redirect the old LMArena-era links

    List every blog slug and add root-level redirects for them.

    Medium impact First result in 2 weeks

02

Your idea portfolio

22 ideas, grouped by when to run them. Open any idea for how to run it: the steps, who to approach, copy you can use, and how to measure it.

  • 6 ideas are marked Unexpected: plays you are unlikely to have tried.

Every idea placed by the effort it takes against the impact it could have. Quick wins sit top left; select a number to open that idea. The filters below narrow the list by status, timing, category and what Programmable can build.

Effort vs impactEach number is an idea. Select one to open it.
Impact →
Quick wins · 0Big bets · 14Fill-ins · 5Reconsider · 3
Effort →

Showing 22 of 22 ideas

This week 6

Quick wins with a clear first action.

  • 01

    Arena passed $100M in annualized revenue within eight months of launching its enterprise offering, yet the site those customers visit has no page for it. The sitemap has 83 leaderboards and no page for AI Evaluations, the API, pricing or contact; /enterprise, /pricing and /contact show the not-found page. The evaluations product's only call to action is an email address in a 2025 blog post, and the CEO's revenue post asks would-be customers to message him on X. The new growth hires are asked to reach "AI labs, and enterprise customers" among Arena's audiences, and the marketing engineer to build "product, features" pages.

    High impactMedium effort First result in 1 month Member of Marketing Staff Gro…

    First action

    Draft the three page outlines with the solutions engineers and agree the form fields with the GTM lead.

    How to measure it

    Inbound evaluation inquiries

    Now
    –
    Target
    60 a month
    By
    90 days

    Steps

    1. 1

      Agree the offer

      What labs, enterprises and API users can buy and how to start.

      Member of Product Staff 3 days

    2. 2

      Write the pages

      Labs, enterprises, API; proof from the Alignment Index and research.

      Member of Marketing Staff - Growth 1 week

    3. 3

      Build them

      In the Sanity CMS and design system.

      Software Engineer - Marketing & Web 1 week

    4. 4

      Add the form

      Company type, use case, timeline; routed to the solutions team.

      GTM Strategy & Ops Lead 2 days

    5. 5

      Link everywhere

      Navigation, footer and an "Evaluate your model" link on leaderboards.

      Software Engineer - Marketing & Web 1 day

    6. 6

      Redirect guesses

      /enterprise, /pricing, /contact to the new section.

      Software Engineer - Marketing & Web 1 hour

    7. 7

      Track

      Inquiries, meetings and pipeline by source.

      GTM Strategy & Ops Lead weekly

    First result
    1 month
    Cost
    Staff time only
    Owner
    Member of Marketing Staff Growth

    Smallest test

    Publish a single "Work with Arena" page with a form and link it from the footer this week.

    Why it works

    Publish an Evaluations section: one page for model labs (pre-release testing, SLAs, co-launches, data deliveries), one for enterprises choosing models (occupational and expert categories, private evaluations), and one for the API, each with a short qualifying form routed to the solutions team. Link it from the main navigation, the footer and every leaderboard ("Evaluate your model"), and redirect /enterprise, /pricing and /contact to it.

    Inbound demand already exists: tens of millions of visitors include the researchers and buyers who become customers. A visible path with a form turns DMs and cold emails into a tracked pipeline the GTM lead can forecast, and it tells new segments that Arena sells to them too.

    Ready to use

    ChecklistEvaluations section launch list

    [ ] /evaluations overview with the neutrality commitments
    [ ] Page for model labs (pre-release, SLAs, co-launches)
    [ ] Page for enterprises (choosing models, private evals)
    [ ] API page
    [ ] Qualifying form routed to solutions
    [ ] Navigation + footer links
    [ ] "Evaluate your model" link on leaderboards
    [ ] Redirect /enterprise, /pricing, /contact
    [ ] Pipeline tracking by source

    Risks and how to handle them

    • Inquiries from unqualified visitors

      Ask company type and use case on the form and route accordingly.

    • Concern about neutrality

      Repeat the AI Evaluations commitments (same methodology for all models) on every page.

    Evidence

    • Your websiteBlog, "Arena Reaches $100M in 8 Months" (June 29, 2026): "$100M annualized revenue run rate within eight months of launching our enterprise offering"; "10M+ monthly visitors, 700M+ total conversations, and 82M+ total votes"; Agent Mode "5M+ turns per month and growing 10% week over week". The CEO closes: "If you want to be part of our growth as a customer, partner, investor, or anything else, please reach out directly to me on X".Source
    • SitemapSitemaps (checked October 22, 2026): 99 site URLs, 83 of them leaderboards (text, WebDev, vision, image, video, agent, search, document; languages from Chinese to Spanish; eight occupational categories), plus 80 blog URLs. There is no page for the enterprise offering, AI Evaluations, the API, pricing or contact, and no page per model. The footer's Company column lists About Us, How It Works, Careers, FAQ, Help Center, two changelogs and Blog.Source
    • Site checkChecked twice on October 22, 2026: /enterprise, /pricing and /contact, and the links /ai-evaluations/, /arena-expert/ and /code-arena/ inside the Series A post (written before posts moved under /blog/), all show the site's "could not be found" page but return HTTP status 200 rather than 404.Source
    • Your websiteBlog, "New Product: AI Evaluations" (September 16, 2025): a commercial service for "enterprises, model labs, and developers" with "Service-level agreements (SLA)" and "flexible pricing" for open-weight models and nonprofits; its only call to action is an email address.Source
    • HiringMember of Marketing Staff - Growth (posted September 28, 2026) and Member of Growth Staff (posted October 6, 2026), both a "senior growth marketer" to "define and drive how we grow our products and lines of business" across "acquisition and activation through engagement, retention, conversion, and monetization", using "SEO/AEO, paid acquisition, email, web, programmatic growth, PLG, launches, and emerging AI-native distribution", and to reach "Arena's distinct audiences, including consumers, developers, researchers, AI labs, and enterprise customers".Source
    • HiringMember of Product Staff (posted September 25, 2026), GM of one pillar: "Enterprise & API" (own "packaging and monetization so that every new product ships instrumented, billable"), "Evaluation" (the leaderboard and methodology), or "Growth" ("How tens of millions of people discover, use, and come back to Arena").Source
    • HiringGTM Strategy & Ops Lead (posted August 12, 2026): "Set pricing and packaging strategy", "Develop the go-to-market approach for new segments", run the deal desk and "quote-to-cash"; "our customers are the labs and enterprises building at the frontier"; "most of this is still being built".Source
  • 02

    Real Chrome visitors to arena.ai wait 4.2 seconds for the main content (rated poor) and see 376 ms delays when they interact, in a product built on typing a prompt and voting. The lab test shows about 730 KB of unused JavaScript, worth 6 seconds on mobile.

    High impactMedium effort First result in 1 month Software Engineer Marketing &…

    First action

    Profile the first prompt and the vote click on a mid-range phone.

    How to measure it

    75th-percentile LCP and INP

    Now
    –
    Target
    under 2.5 s and 200 ms
    By
    90 days

    Steps

    1. 1

      Profile

      Landing, first prompt, vote.

      Software Engineer - Marketing & Web 2 days

    2. 2

      Split bundles

      Load chat features on demand.

      Software Engineer - Marketing & Web 2 weeks

    3. 3

      Server-render

      Hero and prompt box.

      Software Engineer - Marketing & Web 1 week

    4. 4

      Defer scripts

      reCAPTCHA and tags after first input.

      Software Engineer - Marketing & Web 2 days

    5. 5

      Fix long tasks

      Vote and model switch.

      Software Engineer - Marketing & Web 1 week

    6. 6

      Measure

      Votes per visit before and after.

      Member of Product Staff 4 weeks

    First result
    1 month
    Cost
    Engineering time
    Owner
    Software Engineer Marketing & Web

    Smallest test

    Defer one large script and compare LCP in the lab.

    Why it works

    Split the chat app bundle so the landing view loads only what the first prompt needs, server-render the hero and prompt box, defer the reCAPTCHA and analytics scripts until after first input, and break long tasks on vote and model switch. Re-test until real-visitor LCP is under 2.5 s and INP under 200 ms.

    Faster first prompts and snappier votes mean more battles per visit, which is Arena's core data, and better search visibility for the leaderboards.

    Risks and how to handle them

    • Bot protection weakens

      Load reCAPTCHA at submit, not at page load.

    Evidence

    • Google PageSpeedReal Chrome visitors to arena.ai do not pass Core Web Vitals: LCP 4.2 s at the 75th percentile (over 4 s is rated poor) and INP 376 ms (over 200 ms needs improvement); CLS 0.06 is good. The mobile lab test measured LCP 19.1 s, with "Reduce unused JavaScript" worth about 6 s and 730 KB; the LCP element is the "For the latest in AI news" link.Source
    • HiringSoftware Engineer - Marketing & Web (posted August 27, 2026): owns "the blog, marketing pages, and the leaderboard UI that millions of people use"; builds "marketing pages (product, features, resources, careers, campaign pages)"; must "Keep these surfaces fast, accessible, and indexable"; bonus for "SEO, Core Web Vitals, structured data / JSON-LD"; stack Next.js, Tailwind, Sanity CMS.Source
    • Your websiteThe FAQ and How It Works pages have no canonical tag; the blog index has no H1 and the meta description "Blog page for Arena.ai". How It Works: "In battle mode, you'll be served 2 anonymous models", you vote, and "the model identities are revealed". The FAQ explains the Bradley-Terry rating behind the rankings.Source
  • 03

    Pages that do not exist, such as /pricing, and older links inside Arena's own Series A post (/ai-evaluations/, /arena-expert/, /code-arena/, written before posts moved under /blog/) show the not-found page but answer with HTTP status 200. Search engines then treat empty pages as real, and readers following Arena's own links hit a dead end.

    Medium impactLow effort First result in 2 weeks Software Engineer Marketing &…

    First action

    List every blog slug and add root-level redirects for them.

    How to measure it

    Soft 404s reported

    Now
    –
    Target
    0
    By
    60 days

    Steps

    1. 1

      Fix the status

      Not-found route returns 404.

      Software Engineer - Marketing & Web 2 hours

    2. 2

      Add redirects

      Root slugs to /blog/<slug>/.

      Software Engineer - Marketing & Web 1 day

    3. 3

      Fix old posts

      Update internal links in the CMS.

      Software Engineer - Marketing & Web 1 day

    4. 4

      Improve the page

      Helpful links on not-found.

      Member of Design Staff - Product 1 day

    5. 5

      Check Search Console

      Soft 404 and coverage reports.

      Software Engineer - Marketing & Web 2 weeks

    6. 6

      Monitor

      Monthly crawl of internal links.

      Software Engineer - Marketing & Web monthly

    First result
    2 weeks
    Cost
    $0
    Owner
    Software Engineer Marketing & Web

    Smallest test

    Redirect the three broken Series A links today.

    Why it works

    Return a true 404 status from the not-found route, add redirects from old blog slugs at the root to /blog/<slug>/, fix the links inside older posts, and give the not-found page links to the leaderboard, blog and the new Evaluations section.

    Clean status codes keep crawl budget on real pages and pass old links' value to the right posts.

    Risks and how to handle them

    • Breaking app routes

      Test model and chat URLs before deploying.

    Evidence

    • Site checkChecked twice on October 22, 2026: /enterprise, /pricing and /contact, and the links /ai-evaluations/, /arena-expert/ and /code-arena/ inside the Series A post (written before posts moved under /blog/), all show the site's "could not be found" page but return HTTP status 200 rather than 404.Source
    • HiringSoftware Engineer - Marketing & Web (posted August 27, 2026): owns "the blog, marketing pages, and the leaderboard UI that millions of people use"; builds "marketing pages (product, features, resources, careers, campaign pages)"; must "Keep these surfaces fast, accessible, and indexable"; bonus for "SEO, Core Web Vitals, structured data / JSON-LD"; stack Next.js, Tailwind, Sanity CMS.Source
    • SitemapSitemaps (checked October 22, 2026): 99 site URLs, 83 of them leaderboards (text, WebDev, vision, image, video, agent, search, document; languages from Chinese to Spanish; eight occupational categories), plus 80 blog URLs. There is no page for the enterprise offering, AI Evaluations, the API, pricing or contact, and no page per model. The footer's Company column lists About Us, How It Works, Careers, FAQ, Help Center, two changelogs and Blog.Source
  • 04

    After the $200 million Series B at $3.1 billion, the About page still says "$250 million in venture funding" and leads its coverage with the $1.7 billion headlines. Audience numbers also differ: "tens of millions" in job posts, "10M+ monthly visitors" in the revenue post.

    Medium impactLow effort First result in 1 week Member of Marketing Staff Gro…

    First action

    Agree the current audience numbers with the data team.

    How to measure it

    Pages using the agreed figures

    Now
    –
    Target
    100%
    By
    30 days

    Steps

    1. 1

      Agree numbers

      Visitors, conversations, votes.

      Member of Marketing Staff - Growth 1 day

    2. 2

      Update About

      Funding, investors, coverage.

      Member of Marketing Staff - Growth 2 hours

    3. 3

      Update careers

      Same figures.

      Member of Marketing Staff - Growth 1 hour

    4. 4

      Press kit

      One fact sheet.

      Member of Marketing Staff - Growth 1 day

    5. 5

      Schedule reviews

      After each milestone.

      Member of Marketing Staff - Growth quarterly

    6. 6

      Brief the team

      Use the fact sheet.

      Member of Marketing Staff - Growth 1 hour

    First result
    1 week
    Cost
    $0
    Owner
    Member of Marketing Staff Growth

    Smallest test

    Change the funding line today.

    Why it works

    Refresh the About page with the current funding, valuation and investors, the latest coverage, and one set of audience numbers (monthly visitors, conversations, votes) used everywhere: About, careers, press kit and enterprise pages.

    Journalists, candidates and enterprise buyers check the About page; current numbers make the momentum visible.

    Risks and how to handle them

    • Figures go stale again

      Tie updates to each announcement checklist.

    Evidence

    • Your websiteAbout page: "With $250 million in venture funding"; featured coverage still headed "AI Startup LMArena Triples Its Valuation to $1.7 Billion in Latest Fundraise" and "LMArena Lands $1.7B Valuation". The careers posts say "Tens of millions of people use Arena each month"; the revenue post says "10M+ monthly visitors".Source
    • Your websiteBlog, Series B (October 8, 2026): "$200 million Series B at a $3.1 billion valuation", with the Arena Alignment Index "comparing 27 models across 90,000 real-world agent sessions".Source
    • Your websiteBlog, "Arena Reaches $100M in 8 Months" (June 29, 2026): "$100M annualized revenue run rate within eight months of launching our enterprise offering"; "10M+ monthly visitors, 700M+ total conversations, and 82M+ total votes"; Agent Mode "5M+ turns per month and growing 10% week over week". The CEO closes: "If you want to be part of our growth as a customer, partner, investor, or anything else, please reach out directly to me on X".Source
  • 05

    The FAQ and How It Works pages carry no canonical tag, and the blog index has no H1 and the description "Blog page for Arena.ai". These are the pages that explain how the rankings are made, which is what people and AI assistants ask about.

    Medium impactLow effort First result in 2 weeks Software Engineer Marketing &…

    First action

    Add canonical tags to the FAQ and How It Works pages.

    How to measure it

    Search impressions for "how Arena ranks"

    Now
    –
    Target
    +30%
    By
    90 days

    Steps

    1. 1

      Audit tags

      All marketing pages.

      Software Engineer - Marketing & Web 1 day

    2. 2

      Add canonicals

      Self-referencing.

      Software Engineer - Marketing & Web 2 hours

    3. 3

      Fix blog index

      H1 and description.

      Member of Marketing Staff - Growth 1 hour

    4. 4

      FAQ schema

      Structured data.

      Software Engineer - Marketing & Web 2 hours

    5. 5

      Build check

      Fail on missing tags.

      Software Engineer - Marketing & Web 1 day

    6. 6

      Measure

      Search Console.

      Member of Marketing Staff - Growth monthly

    First result
    2 weeks
    Cost
    $0
    Owner
    Software Engineer Marketing & Web

    Smallest test

    Fix the blog index description today.

    Why it works

    Add self-canonicals across marketing pages, an H1 and a descriptive meta for the blog ("Research and news on how AI models perform in the real world"), FAQ structured data on the FAQ page, and an automated check in the build for missing tags.

    Clear signals help the explanation pages rank and get quoted for "how does Arena rank models".

    Risks and how to handle them

    • Duplicate leaderboard URLs

      Canonicalize filtered views to the main category.

    Evidence

    • Your websiteThe FAQ and How It Works pages have no canonical tag; the blog index has no H1 and the meta description "Blog page for Arena.ai". How It Works: "In battle mode, you'll be served 2 anonymous models", you vote, and "the model identities are revealed". The FAQ explains the Bradley-Terry rating behind the rankings.Source
    • HiringSoftware Engineer - Marketing & Web (posted August 27, 2026): owns "the blog, marketing pages, and the leaderboard UI that millions of people use"; builds "marketing pages (product, features, resources, careers, campaign pages)"; must "Keep these surfaces fast, accessible, and indexable"; bonus for "SEO, Core Web Vitals, structured data / JSON-LD"; stack Next.js, Tailwind, Sanity CMS.Source
  • 06

    The only email capture is a "Subscribe to Arena news and research" box on the blog. People visit leaderboards to see who leads coding, image or video, and many come back only when a new model lands. Arena knows the moment a ranking changes.

    High impactMedium effort First result in 1 month Member of Growth Staff

    First action

    Add the alert box to the WebDev leaderboard.

    How to measure it

    Leaderboard alert subscribers

    Now
    –
    Target
    100,000
    By
    2 quarters

    Steps

    1. 1

      Design the alert

      Category, frequency.

      Member of Design Staff - Product 1 week

    2. 2

      Build signup

      On leaderboards.

      Software Engineer - Marketing & Web 1 week

    3. 3

      Trigger emails

      Top-five changes.

      Software Engineer - Marketing & Web 2 weeks

    4. 4

      Write templates

      What changed, try it.

      Member of Growth Staff 1 week

    5. 5

      Launch three categories

      WebDev, Text, Text-to-Image.

      Member of Growth Staff 1 week

    6. 6

      Measure

      Opens, battles from email.

      Member of Growth Staff weekly

    First result
    1 month
    Cost
    Email platform costs
    Owner
    Member of Growth Staff

    Smallest test

    Run one manual email to blog subscribers when the next leader changes.

    Why it works

    Add "Alert me when this changes" to every leaderboard: a weekly or instant email when the top five shift in that category, with a link to try the new leader in battle mode. Start with WebDev, Text and Text-to-Image.

    It turns one-time visitors into an owned audience, brings them back at the exact moment there is news, and feeds more votes.

    Risks and how to handle them

    • Too many emails during busy weeks

      Default to a weekly digest.

    Evidence

    • Your websiteThe blog has a "Subscribe to Arena news and research" email box; the community is on Discord; datasets are released on Hugging Face (for example "a dataset of 5k expert conversations"). Recent research: Max, an "agentic router" with "37% cost savings compared to top models"; "LLM judges prefer their own responses 70% more often than humans do"; HarnessTax across "21 coding agent model-harness combinations"; agent leaderboard categories with task cost.Source
    • SitemapSitemaps (checked October 22, 2026): 99 site URLs, 83 of them leaderboards (text, WebDev, vision, image, video, agent, search, document; languages from Chinese to Spanish; eight occupational categories), plus 80 blog URLs. There is no page for the enterprise offering, AI Evaluations, the API, pricing or contact, and no page per model. The footer's Company column lists About Us, How It Works, Careers, FAQ, Help Center, two changelogs and Blog.Source
    • HiringMember of Marketing Staff - Growth (posted September 28, 2026) and Member of Growth Staff (posted October 6, 2026), both a "senior growth marketer" to "define and drive how we grow our products and lines of business" across "acquisition and activation through engagement, retention, conversion, and monetization", using "SEO/AEO, paid acquisition, email, web, programmatic growth, PLG, launches, and emerging AI-native distribution", and to reach "Arena's distinct audiences, including consumers, developers, researchers, AI labs, and enterprise customers".Source

This quarter 11

The core plays for the next three months.

  • 07

    Arena has rankings for hundreds of models but no page per model in its sitemap. People search "<model> vs <model>" and "<model> ranking" every time a model launches, and rivals such as Artificial Analysis and Vellum publish model pages with price and speed.

    High impactHigh effort First result in 1 quarter Member of Marketing Staff Gro…

    First action

    Build the template for the top 20 models.

    How to measure it

    Model-page sessions

    Now
    –
    Target
    1M a month
    By
    2 quarters

    Steps

    1. 1

      Define the template

      Scores, history, battles.

      Member of Product Staff 2 weeks

    2. 2

      Build pages

      Generated from leaderboard data.

      Software Engineer - Marketing & Web 1 month

    3. 3

      Pick pairs

      By search volume.

      Member of Marketing Staff - Growth 1 week

    4. 4

      Add structured data

      Dataset and item lists.

      Software Engineer - Marketing & Web 1 week

    5. 5

      Launch top 20

      Then expand.

      Member of Marketing Staff - Growth 2 weeks

    6. 6

      Measure

      Sessions and battles.

      Member of Growth Staff monthly

    First result
    1 quarter
    Cost
    Engineering time
    Owner
    Member of Marketing Staff Growth

    Smallest test

    Publish one head-to-head page for the next big launch.

    Why it works

    Generate a page for each ranked model (score by category, rank history, confidence interval, sample battles, release date) and head-to-head pages for the most searched pairs, with "try both now" leading into side-by-side mode. Keep them neutral and data-only.

    Programmatic pages capture launch-day search demand at scale and send it straight into battles.

    Risks and how to handle them

    • Thin pages for small models

      Index only models with enough votes.

    Evidence

    • SitemapSitemaps (checked October 22, 2026): 99 site URLs, 83 of them leaderboards (text, WebDev, vision, image, video, agent, search, document; languages from Chinese to Spanish; eight occupational categories), plus 80 blog URLs. There is no page for the enterprise offering, AI Evaluations, the API, pricing or contact, and no page per model. The footer's Company column lists About Us, How It Works, Careers, FAQ, Help Center, two changelogs and Blog.Source
    • Competitor researchRivals' own pages: Artificial Analysis, "Independent benchmarks across key performance metrics including quality, price, output speed & latency"; OpenRouter Rankings, "ranked by tokens processed"; Scale Labs, "expert-driven LLM benchmarks"; Vellum's LLM Leaderboard, "with pricing and speed data".Source
    • HiringMember of Marketing Staff - Growth (posted September 28, 2026) and Member of Growth Staff (posted October 6, 2026), both a "senior growth marketer" to "define and drive how we grow our products and lines of business" across "acquisition and activation through engagement, retention, conversion, and monetization", using "SEO/AEO, paid acquisition, email, web, programmatic growth, PLG, launches, and emerging AI-native distribution", and to reach "Arena's distinct audiences, including consumers, developers, researchers, AI labs, and enterprise customers".Source
    • HiringSoftware Engineer - Marketing & Web (posted August 27, 2026): owns "the blog, marketing pages, and the leaderboard UI that millions of people use"; builds "marketing pages (product, features, resources, careers, campaign pages)"; must "Keep these surfaces fast, accessible, and indexable"; bonus for "SEO, Core Web Vitals, structured data / JSON-LD"; stack Next.js, Tailwind, Sanity CMS.Source
  • 08

    The growth roles name AEO and "emerging AI-native distribution". Assistants are asked "what is the best model for coding" constantly; Arena's leaderboards are the source many would cite, but they need clear dates, structured data and short text summaries to be quoted correctly.

    High impactMedium effort First result in 1 quarter Software Engineer Marketing &…

    First action

    Ask five assistants ten ranking questions and record citations.

    How to measure it

    Questions where Arena is cited

    Now
    –
    Target
    60%
    By
    2 quarters

    Steps

    1. 1

      Baseline

      Thirty questions.

      Member of Marketing Staff - Growth 1 day

    2. 2

      Text summaries

      Per leaderboard.

      Software Engineer - Marketing & Web 2 weeks

    3. 3

      JSON-LD

      Dataset with dates.

      Software Engineer - Marketing & Web 1 week

    4. 4

      Citation links

      Stable URLs.

      Software Engineer - Marketing & Web 1 week

    5. 5

      Re-check

      Monthly.

      Member of Marketing Staff - Growth monthly

    6. 6

      Report

      Share of citations.

      Member of Growth Staff monthly

    First result
    1 quarter
    Cost
    Engineering time
    Owner
    Software Engineer Marketing & Web

    Smallest test

    Add a dated summary line to the Text leaderboard.

    Why it works

    Add a server-rendered text summary to each leaderboard ("As of <date>, the top model for WebDev is ..."), Dataset JSON-LD with the update date and method, and a stable citation link. Track how often assistants cite Arena for 30 standard questions.

    Being the cited answer keeps Arena's brand in front of users who never open a browser.

    Risks and how to handle them

    • Summaries go stale

      Generate them with each leaderboard update.

    Evidence

    • HiringMember of Marketing Staff - Growth (posted September 28, 2026) and Member of Growth Staff (posted October 6, 2026), both a "senior growth marketer" to "define and drive how we grow our products and lines of business" across "acquisition and activation through engagement, retention, conversion, and monetization", using "SEO/AEO, paid acquisition, email, web, programmatic growth, PLG, launches, and emerging AI-native distribution", and to reach "Arena's distinct audiences, including consumers, developers, researchers, AI labs, and enterprise customers".Source
    • HiringSoftware Engineer - Marketing & Web (posted August 27, 2026): owns "the blog, marketing pages, and the leaderboard UI that millions of people use"; builds "marketing pages (product, features, resources, careers, campaign pages)"; must "Keep these surfaces fast, accessible, and indexable"; bonus for "SEO, Core Web Vitals, structured data / JSON-LD"; stack Next.js, Tailwind, Sanity CMS.Source
    • SitemapSitemaps (checked October 22, 2026): 99 site URLs, 83 of them leaderboards (text, WebDev, vision, image, video, agent, search, document; languages from Chinese to Spanish; eight occupational categories), plus 80 blog URLs. There is no page for the enterprise offering, AI Evaluations, the API, pricing or contact, and no page per model. The footer's Company column lists About Us, How It Works, Careers, FAQ, Help Center, two changelogs and Blog.Source
    • Your websiteThe FAQ and How It Works pages have no canonical tag; the blog index has no H1 and the meta description "Blog page for Arena.ai". How It Works: "In battle mode, you'll be served 2 anonymous models", you vote, and "the model identities are revealed". The FAQ explains the Bradley-Terry rating behind the rankings.Source
  • 09

    Arena's data comes from battle-mode votes: two anonymous models, a vote, then the reveal. New visitors arrive from news and search, often without knowing what to type. Every visitor who leaves without voting is lost data.

    High impactMedium effort First result in 1 quarter Member of Product Staff

    First action

    Measure today's share of new visitors who vote.

    How to measure it

    First-visit vote rate

    Now
    –
    Target
    +25%
    By
    1 quarter

    Steps

    1. 1

      Measure

      Funnel for new visitors.

      Member of Product Staff 1 week

    2. 2

      Design starters

      Three prompts per entry page.

      Member of Design Staff - Product 1 week

    3. 3

      Build

      Starter prompts and reveal.

      Software Engineer - Marketing & Web 2 weeks

    4. 4

      Test

      Half of new visitors.

      Member of Product Staff 1 month

    5. 5

      Check data quality

      Starter prompts weighted fairly.

      Research team 2 weeks

    6. 6

      Roll out

      If quality holds.

      Member of Product Staff 1 week

    First result
    1 quarter
    Cost
    Engineering time
    Owner
    Member of Product Staff

    Smallest test

    Test one starter prompt on the WebDev entry page.

    Why it works

    Offer three one-click starter prompts matched to the referring page (coding, writing, images), explain the vote in one line, and celebrate the reveal with "you just helped rank AI". Measure time to first vote and votes per new visitor.

    A faster first vote raises activation and the volume of fresh data that labs pay for.

    Risks and how to handle them

    • Starter prompts bias the data

      Tag them and let research weight or exclude them.

    Evidence

    • Your websiteThe FAQ and How It Works pages have no canonical tag; the blog index has no H1 and the meta description "Blog page for Arena.ai". How It Works: "In battle mode, you'll be served 2 anonymous models", you vote, and "the model identities are revealed". The FAQ explains the Bradley-Terry rating behind the rankings.Source
    • HiringMember of Product Staff (posted September 25, 2026), GM of one pillar: "Enterprise & API" (own "packaging and monetization so that every new product ships instrumented, billable"), "Evaluation" (the leaderboard and methodology), or "Growth" ("How tens of millions of people discover, use, and come back to Arena").Source
    • Your websiteBlog, "Arena Reaches $100M in 8 Months" (June 29, 2026): "$100M annualized revenue run rate within eight months of launching our enterprise offering"; "10M+ monthly visitors, 700M+ total conversations, and 82M+ total votes"; Agent Mode "5M+ turns per month and growing 10% week over week". The CEO closes: "If you want to be part of our growth as a customer, partner, investor, or anything else, please reach out directly to me on X".Source
  • 10

    People vote without seeing anything about themselves. After enough votes, Arena could tell each person which models they prefer by task, a personal leaderboard they would return to and share.

    Medium impactMedium effort First result in 1 quarter Member of Product Staff

    First action

    Build the personal leaderboard for logged-in voters with 20+ votes.

    How to measure it

    Monthly votes per returning voter

    Now
    –
    Target
    +20%
    By
    2 quarters

    Steps

    1. 1

      Define method

      Simple per-user ranking.

      Research team 2 weeks

    2. 2

      Design view

      Private by default.

      Member of Design Staff - Product 1 week

    3. 3

      Build

      Personal leaderboard.

      Software Engineer - Marketing & Web 1 month

    4. 4

      Share card

      Opt-in image.

      Member of Design Staff - Product 1 week

    5. 5

      Launch

      Logged-in voters.

      Member of Product Staff 1 week

    6. 6

      Measure

      Return rate and shares.

      Member of Growth Staff monthly

    First result
    1 quarter
    Cost
    Engineering time
    Owner
    Member of Product Staff

    Smallest test

    Email 500 active voters their own top three models.

    Why it works

    After 20 votes, show a private "your leaderboard" (top models for your coding, writing and image prompts), with an opt-in share card. Update it as votes accumulate.

    Personal results give voters a reason to keep voting, and shared cards bring new voters.

    Risks and how to handle them

    • Gaming votes for a card

      Keep cards private unless shared and exclude flagged votes.

    Evidence

    • Your websiteThe FAQ and How It Works pages have no canonical tag; the blog index has no H1 and the meta description "Blog page for Arena.ai". How It Works: "In battle mode, you'll be served 2 anonymous models", you vote, and "the model identities are revealed". The FAQ explains the Bradley-Terry rating behind the rankings.Source
    • HiringMember of Product Staff (posted September 25, 2026), GM of one pillar: "Enterprise & API" (own "packaging and monetization so that every new product ships instrumented, billable"), "Evaluation" (the leaderboard and methodology), or "Growth" ("How tens of millions of people discover, use, and come back to Arena").Source
    • Your websiteBlog, "Arena Reaches $100M in 8 Months" (June 29, 2026): "$100M annualized revenue run rate within eight months of launching our enterprise offering"; "10M+ monthly visitors, 700M+ total conversations, and 82M+ total votes"; Agent Mode "5M+ turns per month and growing 10% week over week". The CEO closes: "If you want to be part of our growth as a customer, partner, investor, or anything else, please reach out directly to me on X".Source
  • 11

    Agent Mode passed 5M turns a month within a month and grows about 10% week over week. Agents produce things worth showing (apps, sites, analyses), and every shared result is an invitation to try Arena.

    High impactMedium effort First result in 1 quarter Member of Growth Staff

    First action

    Add share links to WebDev results first.

    How to measure it

    Visitors from shares

    Now
    –
    Target
    500,000 a month
    By
    2 quarters

    Steps

    1. 1

      Design share page

      Output, models, CTA.

      Member of Design Staff - Product 1 week

    2. 2

      Build links

      With privacy controls.

      Software Engineer - Marketing & Web 2 weeks

    3. 3

      Review safety

      No private data in public links.

      Member of Security Staff, Privacy & Trust Ops Lead 1 week

    4. 4

      Launch

      WebDev, then Agent Mode.

      Member of Growth Staff 1 week

    5. 5

      Seed

      Share the best results on X.

      Member of Marketing Staff - Growth weekly

    6. 6

      Measure

      Visitors and battles from shares.

      Member of Growth Staff weekly

    First result
    1 quarter
    Cost
    Engineering time
    Owner
    Member of Growth Staff

    Smallest test

    Hand-pick ten great results and post them with links.

    Why it works

    Add a one-click public share link for finished agent and WebDev results, showing the output, both models and a "run your own" button. Track signups and battles from shared links.

    User-made outputs are the most persuasive ads for a product built on comparing models.

    Risks and how to handle them

    • Users share sensitive prompts

      Make sharing opt-in with a preview.

    Evidence

    • Your websiteBlog, "Arena Reaches $100M in 8 Months" (June 29, 2026): "$100M annualized revenue run rate within eight months of launching our enterprise offering"; "10M+ monthly visitors, 700M+ total conversations, and 82M+ total votes"; Agent Mode "5M+ turns per month and growing 10% week over week". The CEO closes: "If you want to be part of our growth as a customer, partner, investor, or anything else, please reach out directly to me on X".Source
    • HiringMember of Marketing Staff - Growth (posted September 28, 2026) and Member of Growth Staff (posted October 6, 2026), both a "senior growth marketer" to "define and drive how we grow our products and lines of business" across "acquisition and activation through engagement, retention, conversion, and monetization", using "SEO/AEO, paid acquisition, email, web, programmatic growth, PLG, launches, and emerging AI-native distribution", and to reach "Arena's distinct audiences, including consumers, developers, researchers, AI labs, and enterprise customers".Source
    • News coverageArena results travel: a Hacker News post in October 2026 was titled "Only 1 of the Top 5 AI Coding Models on WebDev Arena Isn't Chinese".Source
  • 12

    The GTM lead will "Develop the go-to-market approach for new segments" and set packaging, and Arena promised "flexible pricing" for open-weight models and nonprofits. Today every path into evaluations is a conversation.

    High impactMedium effort First result in 1 quarter GTM Strategy & Ops Lead

    First action

    Define the smallest evaluation the team can deliver without custom work.

    How to measure it

    Self-serve evaluations

    Now
    –
    Target
    50 a quarter
    By
    2 quarters

    Steps

    1. 1

      Define scope

      Battles, categories, report.

      Member of Product Staff 2 weeks

    2. 2

      Price it

      Rate card and discounts.

      GTM Strategy & Ops Lead 2 weeks

    3. 3

      Build checkout

      And intake form.

      Software Engineer - Marketing & Web 1 month

    4. 4

      Automate report

      Template.

      Applied AI Engineer 1 month

    5. 5

      Publish

      On the Evaluations section.

      Member of Marketing Staff - Growth 1 week

    6. 6

      Measure

      Sales and upgrades.

      GTM Strategy & Ops Lead monthly

    First result
    1 quarter
    Cost
    Engineering time
    Owner
    GTM Strategy & Ops Lead

    Smallest test

    Sell the package by hand to five open-weight teams.

    Why it works

    Package a small, fixed-scope evaluation (a set number of battles in chosen categories, a report, a sample of feedback) with a published price and card checkout, and route larger needs to the solutions team.

    A self-serve tier opens the long tail of model builders and creates leads for full evaluations.

    Risks and how to handle them

    • Delivery load on solutions engineers

      Cap monthly slots at first.

    Builds on

    Evidence

    • HiringGTM Strategy & Ops Lead (posted August 12, 2026): "Set pricing and packaging strategy", "Develop the go-to-market approach for new segments", run the deal desk and "quote-to-cash"; "our customers are the labs and enterprises building at the frontier"; "most of this is still being built".Source
    • Your websiteBlog, "New Product: AI Evaluations" (September 16, 2025): a commercial service for "enterprises, model labs, and developers" with "Service-level agreements (SLA)" and "flexible pricing" for open-weight models and nonprofits; its only call to action is an email address.Source
    • HiringMember of Product Staff (posted September 25, 2026), GM of one pillar: "Enterprise & API" (own "packaging and monetization so that every new product ships instrumented, billable"), "Evaluation" (the leaderboard and methodology), or "Growth" ("How tens of millions of people discover, use, and come back to Arena").Source
    • HiringApplied AI Engineer (posted September 10, 2026) and Product Solutions Engineer: each owns "5-15 active model lab relationships", delivers evals, and drives "onboarding, data + product releases, co-launches"; "Help expand strategic accounts".Source
  • 13

    Solutions engineers drive "co-launches" with labs, and new model results travel fast (a WebDev Arena ranking reached the Hacker News front page). Launch days bring Arena huge attention, usually only on X.

    High impactMedium effort First result in 1 quarter Member of Marketing Staff Gro…

    First action

    Write the kit and agree it with one lab.

    How to measure it

    Launch-day sessions

    Now
    –
    Target
    +50%
    By
    2 quarters

    Steps

    1. 1

      Write the kit

      Page, video, thread.

      Member of Marketing Staff - Growth 1 week

    2. 2

      Agree with labs

      Terms and timing.

      Applied AI Engineer per launch

    3. 3

      Build template

      Results page.

      Software Engineer - Marketing & Web 2 weeks

    4. 4

      Run launches

      Every major model.

      Member of Marketing Staff - Growth per launch

    5. 5

      Track links

      From lab announcements.

      Member of Growth Staff per launch

    6. 6

      Review

      After each launch.

      Member of Marketing Staff - Growth per launch

    First result
    1 quarter
    Cost
    Staff time only
    Owner
    Member of Marketing Staff Growth

    Smallest test

    Publish a results page for the next big release.

    Why it works

    Agree a standard co-launch kit with labs: a results page on arena.ai, a short video, a thread, and the model live in battle mode the same hour, with the lab linking back. Publish the results page for every major launch.

    Launch-day pages capture search and press traffic, and labs see Arena as a launch partner, not just a vendor.

    Risks and how to handle them

    • Perceived favoritism

      Offer the same kit to every lab.

    Evidence

    • HiringApplied AI Engineer (posted September 10, 2026) and Product Solutions Engineer: each owns "5-15 active model lab relationships", delivers evals, and drives "onboarding, data + product releases, co-launches"; "Help expand strategic accounts".Source
    • News coverageArena results travel: a Hacker News post in October 2026 was titled "Only 1 of the Top 5 AI Coding Models on WebDev Arena Isn't Chinese".Source
    • Your websiteBlog, "Arena Reaches $100M in 8 Months" (June 29, 2026): "$100M annualized revenue run rate within eight months of launching our enterprise offering"; "10M+ monthly visitors, 700M+ total conversations, and 82M+ total votes"; Agent Mode "5M+ turns per month and growing 10% week over week". The CEO closes: "If you want to be part of our growth as a customer, partner, investor, or anything else, please reach out directly to me on X".Source
  • 14

    Arena's YouTube channel has about 35,000 subscribers and 187 videos. Leaderboard movements every week are a natural show: who rose, who fell, and what voters liked.

    Medium impactLow effort First result in 1 quarter Member of Marketing Staff Gro…

    First action

    Record the first episode with this week's moves.

    How to measure it

    YouTube subscribers

    Now
    –
    Target
    60,000
    By
    2 quarters

    Steps

    1. 1

      Format

      Five minutes, fixed segments.

      Member of Marketing Staff - Growth 1 week

    2. 2

      Record weekly

      Leaderboard moves.

      Member of Marketing Staff - Growth weekly

    3. 3

      Cut shorts

      Three per episode.

      Member of Marketing Staff - Growth weekly

    4. 4

      Link alerts

      In descriptions.

      Member of Growth Staff weekly

    5. 5

      Review numbers

      Views and votes.

      Member of Growth Staff monthly

    6. 6

      Adjust

      After eight episodes.

      Member of Marketing Staff - Growth 2 months

    First result
    1 quarter
    Cost
    Production time
    Owner
    Member of Marketing Staff Growth

    Smallest test

    Publish one episode.

    Why it works

    Publish a five-minute weekly episode with the biggest moves, two example battles and a research highlight, cut into shorts for X and LinkedIn, and link to the leaderboard alerts.

    A regular show builds habit and brings people back to vote on the models in the news.

    Risks and how to handle them

    • Neutrality questions

      Report data, no opinions on labs.

    Evidence

    • YouTubeThe @ArenaAIOfficial YouTube channel has about 35,000 subscribers, 187 videos and 1.3 million total views.Source
    • SitemapSitemaps (checked October 22, 2026): 99 site URLs, 83 of them leaderboards (text, WebDev, vision, image, video, agent, search, document; languages from Chinese to Spanish; eight occupational categories), plus 80 blog URLs. There is no page for the enterprise offering, AI Evaluations, the API, pricing or contact, and no page per model. The footer's Company column lists About Us, How It Works, Careers, FAQ, Help Center, two changelogs and Blog.Source
    • Your websiteThe blog has a "Subscribe to Arena news and research" email box; the community is on Discord; datasets are released on Hugging Face (for example "a dataset of 5k expert conversations"). Recent research: Max, an "agentic router" with "37% cost savings compared to top models"; "LLM judges prefer their own responses 70% more often than humans do"; HarnessTax across "21 coding agent model-harness combinations"; agent leaderboard categories with task cost.Source
  • 15

    Arena's research is newsworthy: LLM judges prefer their own answers 70% more often than humans, coding harnesses change cost, Max routes for 37% savings, an Alignment Index over 27 models. Most of it is published as long blog posts.

    High impactMedium effort First result in 1 quarter Member of Marketing Staff Gro…

    First action

    Package the self-preference study this week.

    How to measure it

    Signups per research post

    Now
    –
    Target
    2,000
    By
    2 quarters

    Steps

    1. 1

      Template the kit

      Chart, thread, video, press.

      Member of Marketing Staff - Growth 1 week

    2. 2

      Package the last three posts

      Back catalogue.

      Member of Marketing Staff - Growth 2 weeks

    3. 3

      Press list

      AI reporters.

      Member of Marketing Staff - Growth 1 week

    4. 4

      Add signup

      On research posts.

      Software Engineer - Marketing & Web 2 days

    5. 5

      Publish datasets

      Hugging Face.

      Research team per post

    6. 6

      Measure

      Citations and signups.

      Member of Growth Staff monthly

    First result
    1 quarter
    Cost
    Staff time only
    Owner
    Member of Marketing Staff Growth

    Smallest test

    Make a one-chart summary of the self-preference result.

    Why it works

    For each post, ship a one-chart summary, a thread, a two-minute video, the dataset on Hugging Face and a press note to AI reporters; add an email signup on the post. Track citations and signups per post.

    Research is Arena's strongest brand asset; packaging it for each channel multiplies reach without new studies.

    Risks and how to handle them

    • Oversimplified findings

      Researchers approve each summary.

    Evidence

    • Your websiteThe blog has a "Subscribe to Arena news and research" email box; the community is on Discord; datasets are released on Hugging Face (for example "a dataset of 5k expert conversations"). Recent research: Max, an "agentic router" with "37% cost savings compared to top models"; "LLM judges prefer their own responses 70% more often than humans do"; HarnessTax across "21 coding agent model-harness combinations"; agent leaderboard categories with task cost.Source
    • Your websiteBlog, Series B (October 8, 2026): "$200 million Series B at a $3.1 billion valuation", with the Arena Alignment Index "comparing 27 models across 90,000 real-world agent sessions".Source
    • HiringMember of Marketing Staff - Growth (posted September 28, 2026) and Member of Growth Staff (posted October 6, 2026), both a "senior growth marketer" to "define and drive how we grow our products and lines of business" across "acquisition and activation through engagement, retention, conversion, and monetization", using "SEO/AEO, paid acquisition, email, web, programmatic growth, PLG, launches, and emerging AI-native distribution", and to reach "Arena's distinct audiences, including consumers, developers, researchers, AI labs, and enterprise customers".Source
  • 16

    Arena already ranks models for medicine, legal, finance, software and other occupations, and the GTM lead is asked to size new segments. Enterprises choosing a model for a function need exactly this, beyond the public table.

    High impactMedium effort First result in 2 quarters GTM Strategy & Ops Lead

    First action

    Draft the software and legal reports first.

    How to measure it

    Report-sourced enterprise leads

    Now
    –
    Target
    100 a quarter
    By
    2 quarters

    Steps

    1. 1

      Pick occupations

      Two to start.

      GTM Strategy & Ops Lead 1 week

    2. 2

      Analyze

      Rankings and examples.

      Research team 1 month

    3. 3

      Write reports

      Summary and full.

      Member of Marketing Staff - Growth 2 weeks

    4. 4

      Gate

      Form on the Evaluations section.

      Software Engineer - Marketing & Web 1 week

    5. 5

      Promote

      Industry newsletters.

      Member of Growth Staff 1 month

    6. 6

      Follow up

      Solutions team.

      Applied AI Engineer ongoing

    First result
    2 quarters
    Cost
    Analyst time
    Owner
    GTM Strategy & Ops Lead

    Smallest test

    Publish one occupation summary.

    Why it works

    Publish a free summary per occupation (top models, strengths, cost notes) with a gated full report including prompt examples and failure cases, and offer a private evaluation as the next step. Promote through industry newsletters.

    It opens enterprise buyers who are not labs, using data Arena already has.

    Risks and how to handle them

    • Report seen as endorsement

      Show methodology and dates prominently.

    Builds on

    Evidence

    • SitemapSitemaps (checked October 22, 2026): 99 site URLs, 83 of them leaderboards (text, WebDev, vision, image, video, agent, search, document; languages from Chinese to Spanish; eight occupational categories), plus 80 blog URLs. There is no page for the enterprise offering, AI Evaluations, the API, pricing or contact, and no page per model. The footer's Company column lists About Us, How It Works, Careers, FAQ, Help Center, two changelogs and Blog.Source
    • HiringGTM Strategy & Ops Lead (posted August 12, 2026): "Set pricing and packaging strategy", "Develop the go-to-market approach for new segments", run the deal desk and "quote-to-cash"; "our customers are the labs and enterprises building at the frontier"; "most of this is still being built".Source
    • HiringMember of Product Staff (posted September 25, 2026), GM of one pillar: "Enterprise & API" (own "packaging and monetization so that every new product ships instrumented, billable"), "Evaluation" (the leaderboard and methodology), or "Growth" ("How tens of millions of people discover, use, and come back to Arena").Source
    • Your websiteBlog, "New Product: AI Evaluations" (September 16, 2025): a commercial service for "enterprises, model labs, and developers" with "Service-level agreements (SLA)" and "flexible pricing" for open-weight models and nonprofits; its only call to action is an email address.Source
  • 17

    Arena ranks models in Chinese, French, German, Japanese, Korean, Polish, Russian and Spanish, but the site, blog and help are in English. Speakers of those languages search for the best model in their own language.

    Medium impactMedium effort First result in 2 quarters Member of Growth Staff

    First action

    Translate one landing page for Japanese.

    How to measure it

    Non-English votes

    Now
    –
    Target
    +30%
    By
    2 quarters

    Steps

    1. 1

      Pick languages

      By traffic and votes.

      Member of Growth Staff 1 week

    2. 2

      Translate

      Landing and how-to.

      Member of Growth Staff 1 month

    3. 3

      Build pages

      With hreflang.

      Software Engineer - Marketing & Web 2 weeks

    4. 4

      Starter prompts

      Native speakers.

      Research team 2 weeks

    5. 5

      Promote

      Local communities.

      Member of Growth Staff 1 month

    6. 6

      Measure

      Votes by language.

      Member of Product Staff monthly

    First result
    2 quarters
    Cost
    Translation costs
    Owner
    Member of Growth Staff

    Smallest test

    Launch Japanese first.

    Why it works

    Create a landing page in each language explaining the leaderboard and how to vote, with the language leaderboard embedded and starter prompts in that language; promote in local AI communities.

    More non-English votes strengthen the language leaderboards, and the pages rank for local searches.

    Risks and how to handle them

    • Poor translations

      Native-speaker review.

    Evidence

    • SitemapSitemaps (checked October 22, 2026): 99 site URLs, 83 of them leaderboards (text, WebDev, vision, image, video, agent, search, document; languages from Chinese to Spanish; eight occupational categories), plus 80 blog URLs. There is no page for the enterprise offering, AI Evaluations, the API, pricing or contact, and no page per model. The footer's Company column lists About Us, How It Works, Careers, FAQ, Help Center, two changelogs and Blog.Source
    • HiringMember of Marketing Staff - Growth (posted September 28, 2026) and Member of Growth Staff (posted October 6, 2026), both a "senior growth marketer" to "define and drive how we grow our products and lines of business" across "acquisition and activation through engagement, retention, conversion, and monetization", using "SEO/AEO, paid acquisition, email, web, programmatic growth, PLG, launches, and emerging AI-native distribution", and to reach "Arena's distinct audiences, including consumers, developers, researchers, AI labs, and enterprise customers".Source
    • Your websiteThe FAQ and How It Works pages have no canonical tag; the blog index has no H1 and the meta description "Blog page for Arena.ai". How It Works: "In battle mode, you'll be served 2 anonymous models", you vote, and "the model identities are revealed". The FAQ explains the Bradley-Terry rating behind the rankings.Source

Big bets 3

Larger plays that can change the trajectory.

  • 18

    The Enterprise & API pillar owns packaging and monetization, and enterprises need to know which model works for their work, not the public average. Arena's method (blind pairwise votes, Bradley-Terry ratings) works for any group of raters.

    High impactHigh effort First result in 3 quarters Member of Product Staff

    First action

    Pilot a private arena with two existing customers.

    How to measure it

    Paying private arenas

    Now
    –
    Target
    30
    By
    3 quarters

    Steps

    1. 1

      Define product

      Prompts, raters, reports.

      Member of Product Staff 1 month

    2. 2

      Security review

      Data isolation.

      Member of Security Staff, Privacy & Trust Ops Lead 1 month

    3. 3

      Build MVP

      Private workspace.

      Software Engineer - Marketing & Web 1 quarter

    4. 4

      Pilot

      Two customers.

      Applied AI Engineer 1 quarter

    5. 5

      Price

      Seats or evaluations.

      GTM Strategy & Ops Lead 1 month

    6. 6

      Launch

      Evaluations section.

      Member of Marketing Staff - Growth 1 month

    First result
    3 quarters
    Cost
    Engineering time
    Owner
    Member of Product Staff

    Smallest test

    Run one private arena by hand for a customer.

    Why it works

    Offer private arenas: a company uploads prompts or invites employees, they vote blind on models of the company's choice, and the company gets a private leaderboard and report. Price per seat or per evaluation.

    It is a product only Arena can credibly offer and a large new revenue line beyond labs.

    Risks and how to handle them

    • Enterprise data concerns

      Isolated storage and no training use.

    Builds on

    Evidence

    • HiringMember of Product Staff (posted September 25, 2026), GM of one pillar: "Enterprise & API" (own "packaging and monetization so that every new product ships instrumented, billable"), "Evaluation" (the leaderboard and methodology), or "Growth" ("How tens of millions of people discover, use, and come back to Arena").Source
    • Your websiteThe FAQ and How It Works pages have no canonical tag; the blog index has no H1 and the meta description "Blog page for Arena.ai". How It Works: "In battle mode, you'll be served 2 anonymous models", you vote, and "the model identities are revealed". The FAQ explains the Bradley-Terry rating behind the rankings.Source
    • HiringGTM Strategy & Ops Lead (posted August 12, 2026): "Set pricing and packaging strategy", "Develop the go-to-market approach for new segments", run the deal desk and "quote-to-cash"; "our customers are the labs and enterprises building at the frontier"; "most of this is still being built".Source
    • Your websiteBlog, "New Product: AI Evaluations" (September 16, 2025): a commercial service for "enterprises, model labs, and developers" with "Service-level agreements (SLA)" and "flexible pricing" for open-weight models and nonprofits; its only call to action is an email address.Source
  • 19

    Max reached frontier performance in Agent Arena with 37% cost savings compared with top models. Developers pay for exactly that trade-off, and the routing comes from Arena's preference data.

    High impactHigh effort First result in 3 quarters Co Founder and CEO

    First action

    Estimate unit costs of serving Max at scale.

    How to measure it

    Max API revenue

    Now
    –
    Target
    $1M ARR
    By
    4 quarters

    Steps

    1. 1

      Unit economics

      Cost per task.

      GTM Strategy & Ops Lead 1 month

    2. 2

      API design

      Endpoints and limits.

      Member of Product Staff 1 month

    3. 3

      Billing

      Metering and invoices.

      GTM Strategy & Ops Lead 1 quarter

    4. 4

      Docs

      Quickstart.

      Software Engineer - Marketing & Web 1 month

    5. 5

      Beta

      Fifty developers.

      Member of Product Staff 1 quarter

    6. 6

      Launch

      Pricing page.

      Member of Marketing Staff - Growth 1 month

    First result
    3 quarters
    Cost
    Engineering and compute costs
    Owner
    Co Founder and CEO

    Smallest test

    Offer a waitlist on the Max post.

    Why it works

    Offer Max as a metered API with a free allowance, a public price per task, and dashboards showing savings, built on the quote-to-cash work the GTM lead owns.

    It turns research into recurring developer revenue and puts Arena inside production apps.

    Risks and how to handle them

    • Conflict with labs as customers

      Position as routing between their models, sending them traffic.

    Evidence

    • Your websiteThe blog has a "Subscribe to Arena news and research" email box; the community is on Discord; datasets are released on Hugging Face (for example "a dataset of 5k expert conversations"). Recent research: Max, an "agentic router" with "37% cost savings compared to top models"; "LLM judges prefer their own responses 70% more often than humans do"; HarnessTax across "21 coding agent model-harness combinations"; agent leaderboard categories with task cost.Source
    • HiringGTM Strategy & Ops Lead (posted August 12, 2026): "Set pricing and packaging strategy", "Develop the go-to-market approach for new segments", run the deal desk and "quote-to-cash"; "our customers are the labs and enterprises building at the frontier"; "most of this is still being built".Source
    • HiringMember of Product Staff (posted September 25, 2026), GM of one pillar: "Enterprise & API" (own "packaging and monetization so that every new product ships instrumented, billable"), "Evaluation" (the leaderboard and methodology), or "Growth" ("How tens of millions of people discover, use, and come back to Arena").Source
  • 20

    Arena rankings are quoted widely, often as screenshots that go stale. An embed keeps them current and links back.

    High impactMedium effort First result in 2 quarters Software Engineer Marketing &…

    First action

    Build an embed for the WebDev top five.

    How to measure it

    Active embeds

    Now
    –
    Target
    1,000
    By
    2 quarters

    Steps

    1. 1

      Design widget

      Light and dark.

      Member of Design Staff - Product 2 weeks

    2. 2

      Build

      Fast and cached.

      Software Engineer - Marketing & Web 1 month

    3. 3

      Embed code

      On each leaderboard.

      Software Engineer - Marketing & Web 1 week

    4. 4

      Press page

      How to embed.

      Member of Marketing Staff - Growth 1 week

    5. 5

      Outreach

      AI newsletters and reporters.

      Member of Marketing Staff - Growth 1 month

    6. 6

      Measure

      Embeds and visits.

      Member of Growth Staff monthly

    First result
    2 quarters
    Cost
    Engineering time
    Owner
    Software Engineer Marketing & Web

    Smallest test

    Send the embed to five AI newsletters.

    Why it works

    Offer an embeddable widget for any category (top 5 or 10, last updated date, link to vote), with a simple embed code on each leaderboard and a press page.

    Every embed is a permanent backlink and a source of new voters.

    Risks and how to handle them

    • Load on servers

      Cache with a CDN.

    Evidence

    • News coverageArena results travel: a Hacker News post in October 2026 was titled "Only 1 of the Top 5 AI Coding Models on WebDev Arena Isn't Chinese".Source
    • SitemapSitemaps (checked October 22, 2026): 99 site URLs, 83 of them leaderboards (text, WebDev, vision, image, video, agent, search, document; languages from Chinese to Spanish; eight occupational categories), plus 80 blog URLs. There is no page for the enterprise offering, AI Evaluations, the API, pricing or contact, and no page per model. The footer's Company column lists About Us, How It Works, Careers, FAQ, Help Center, two changelogs and Blog.Source
    • HiringSoftware Engineer - Marketing & Web (posted August 27, 2026): owns "the blog, marketing pages, and the leaderboard UI that millions of people use"; builds "marketing pages (product, features, resources, careers, campaign pages)"; must "Keep these surfaces fast, accessible, and indexable"; bonus for "SEO, Core Web Vitals, structured data / JSON-LD"; stack Next.js, Tailwind, Sanity CMS.Source

Wildcards 2

Unexpected experiments, each with a cheap way to test it.

  • 21

    Arena started at UC Berkeley and runs an Academic Partnerships Program with calls every spring and fall. Students are heavy AI users and the best source of expert prompts in their fields.

    Medium impactLow effort First result in 2 quarters Member of Growth Staff

    First action

    Invite the first cohort's universities to nominate ambassadors.

    How to measure it

    Campus expert prompts

    Now
    –
    Target
    50,000
    By
    2 quarters

    Steps

    1. 1

      Design program

      Roles and rewards.

      Member of Growth Staff 2 weeks

    2. 2

      Recruit

      Ten universities.

      Member of Growth Staff 1 month

    3. 3

      Kit

      Event guide.

      Member of Marketing Staff - Growth 2 weeks

    4. 4

      Run events

      Voting nights.

      Member of Growth Staff monthly

    5. 5

      Tag data

      Expert prompts.

      Research team ongoing

    6. 6

      Measure

      Prompts and users.

      Member of Product Staff monthly

    First result
    2 quarters
    Cost
    Program budget
    Owner
    Member of Growth Staff

    Smallest test

    Run one voting night at one campus.

    Why it works

    Recruit ambassadors at partner universities to run voting nights, expert-prompt drives for their discipline, and course integrations, with credits and recognition.

    It grows expert-level data and the next generation of users at low cost.

    Risks and how to handle them

    • Low-quality votes

      Use existing quality filters.

    Evidence

    • Your websiteArena's Academic Partnerships Program opens research proposals "twice a year, every Spring and Fall", and announced its first cohort on September 1, 2026. LMArena became Arena at arena.ai on January 28, 2026.Source
    • HiringMember of Marketing Staff - Growth (posted September 28, 2026) and Member of Growth Staff (posted October 6, 2026), both a "senior growth marketer" to "define and drive how we grow our products and lines of business" across "acquisition and activation through engagement, retention, conversion, and monetization", using "SEO/AEO, paid acquisition, email, web, programmatic growth, PLG, launches, and emerging AI-native distribution", and to reach "Arena's distinct audiences, including consumers, developers, researchers, AI labs, and enterprise customers".Source
    • Your websiteThe blog has a "Subscribe to Arena news and research" email box; the community is on Discord; datasets are released on Hugging Face (for example "a dataset of 5k expert conversations"). Recent research: Max, an "agentic router" with "37% cost savings compared to top models"; "LLM judges prefer their own responses 70% more often than humans do"; HarnessTax across "21 coding agent model-harness combinations"; agent leaderboard categories with task cost.Source
  • 22

    Launch days bring the biggest crowds to Arena. A live stream showing votes come in and the first ranking appear would turn a static table into an event.

    Medium impactMedium effort First result in 1 quarter Member of Marketing Staff Gro…

    First action

    Plan a stream for the next major release.

    How to measure it

    Votes per stream

    Now
    –
    Target
    100,000
    By
    2 quarters

    Steps

    1. 1

      Plan

      Format and roles.

      Member of Marketing Staff - Growth 1 week

    2. 2

      Build overlay

      Live counts.

      Software Engineer - Marketing & Web 2 weeks

    3. 3

      Rehearse

      Researcher segment.

      Research team 1 week

    4. 4

      Stream

      Next launch.

      Member of Marketing Staff - Growth per launch

    5. 5

      Clip

      Highlights.

      Member of Marketing Staff - Growth per launch

    6. 6

      Review

      Votes and views.

      Member of Growth Staff per launch

    First result
    1 quarter
    Cost
    Production time
    Owner
    Member of Marketing Staff Growth

    Smallest test

    Stream the next launch for 30 minutes.

    Why it works

    Stream on YouTube and X when a major model enters the arena: live vote counts, early results with confidence intervals, a researcher explaining the method. Invite viewers to vote during the stream.

    Live events attract press and voters at once and show the rigor behind the numbers.

    Risks and how to handle them

    • Early results mislead

      Show confidence intervals and wait thresholds.

    Evidence

    • YouTubeThe @ArenaAIOfficial YouTube channel has about 35,000 subscribers, 187 videos and 1.3 million total views.Source
    • Your websiteThe FAQ and How It Works pages have no canonical tag; the blog index has no H1 and the meta description "Blog page for Arena.ai". How It Works: "In battle mode, you'll be served 2 anonymous models", you vote, and "the model identities are revealed". The FAQ explains the Bradley-Terry rating behind the rankings.Source
    • News coverageArena results travel: a Hacker News post in October 2026 was titled "Only 1 of the Top 5 AI Coding Models on WebDev Arena Isn't Chinese".Source

03

How Programmable can help

Where an idea in this playbook is something Programmable builds or runs, it's listed here with what we'd do. Everything else is yours to run.

ProgrammableDigital

Show it, don't just describe it

These ideas need something to show: a demo, a video, an animation, graphics or a deck. Digital turns screenshots or the site into all of them, in the company's own brand.

Make a demo (opens programmable.digital in a new tab)
More example playbooks and reviews

More growth playbooks

Other companies we have researched, each read from what they are hiring for.

  • Aemon growth strategyAn autonomous AI R&D engineer: given a problem and an eval, Aemon studies the research frontier, runs experiments inside the customer's…23 growth ideas
  • Outlook Amusements (California Psychics) growth strategyRuns California Psychics, an online marketplace for psychic, tarot and astrology readings by phone and chat with vetted advisors, plus…20 growth ideas
  • Comfrt growth strategyA fast-growing direct-to-consumer comfort brand: weighted and signature hoodies, sweatpants, loungewear, Dreamer blankets, kids and pet…21 growth ideas
  • Theraplex growth strategyA dermatologist-owned skincare brand making simple moisturizers and skin protectants for sensitive skin, eczema, psoriasis, ichthyosis and…20 growth ideas
  • saas.group growth strategyAcquires and grows bootstrapped, product-led B2B SaaS companies, with central marketing, finance, platform and talent teams supporting…20 growth ideas
  • Welltech growth strategyA wellness app company behind Muscle Booster, Yoga-Go and WalkFit, sold as subscriptions through quiz funnels and performance marketing…21 growth ideas
  • Cognition growth strategyAn applied AI lab that makes Devin, an autonomous AI software engineer (cloud agents, Devin Desktop, formerly Windsurf, and the CLI), used…20 growth ideas
  • Checkout.com growth strategyA global digital payments platform: acquiring, payment processing, fraud detection, authentication, Flow checkout, payouts, issuing…23 growth ideas

See every playbook and website review

Independent research. Programmable wrote this playbook from public sources, as they stood on October 23, 2026, and cites them throughout. Arena did not commission, review or endorse it, and Programmable is not affiliated with Arena. Company and product names are trademarks of their owners, used only to identify them. The ideas are suggestions and opinion, not professional, legal or financial advice, and not statements about how Arena runs its business.

Something wrong, or do you represent Arena? Email [email protected] and we will correct it or take it down.

30/60/90 roadmap

What to focus on in each month, and the ideas that carry it.

  1. Days 1 to 306 ideas

    Evaluations front door, Core Web Vitals, real 404s and redirects, About page figures, canonicals and leaderboard alerts.

  2. Days 31 to 6013 ideas

    Model pages, AEO, first-vote activation, personal leaderboards, share links, self-serve tier, co-launch pages, YouTube, research kits, occupational reports and language pages; campus ambassadors and launch streams.

  3. Days 61 to 903 ideas

    Commit to private arenas, a Max API and embeddable leaderboards.

The research behind the plan

3 research areas, each summarised in a few lines. Open any of them for the full detail.

Customers

What customers say

Customers value blind, side-by-side comparisons by real people. The most cited problem: static benchmarks that models learn to game.

  • Free access to frontier models
  • Choosing a model for a specific job
  • In their words: “leaderboard”

Market

Competitive landscape

4 competitors compared: Artificial Analysis, OpenRouter Rankings, Scale Labs leaderboards and Vellum LLM Leaderboard.

  • Gap to exploit: Model choice by occupation for enterprises
  • Artificial Analysis: Not human-preference based
  • OpenRouter Rankings: Popularity, not quality

Hiring signals

Arena is building its first real growth and GTM functions.

  • Member of Marketing Staff - Growth: Full-funnel growth across labs, enterprises and consumers. Ideas 1, 7, 8, 14, 15.
  • Member of Growth Staff: SEO/AEO, email, PLG and AI-native distribution. Ideas 6, 11, 17.
  • Software Engineer - Marketing & Web: Fast, indexable marketing pages and leaderboard UI. Ideas 2, 3, 5, 20.

From the team that wrote this playbook

All products
  • ProgrammableAgencyWorking AI systems in days, not quarters (opens programmable.agency in a new tab)
  • ProgrammableInfoAn agent that finds, researches and asks first (opens programmable.info in a new tab)
  • ProgrammableDigitalScreenshots in, an animated product demo out (opens programmable.digital in a new tab)
  • ProgrammableWatchKnow the day a supplier falls out of compliance (opens programmable.watch in a new tab)
  • ProgrammableTalkGet registrants for your events, and know who’s coming (opens programmable.talk in a new tab)
  • ProgrammablePageText what you need, your customer gets a page that works (opens programmable.page in a new tab)
  • ProgrammableCommunityForums with labelled AI members, never empty (opens programmable.community in a new tab)
ProgrammableMarketing
ExamplesLeaderboardsProductsTermsPrivacyCookiesDPALegalContact

© 2026 AI-Powered Growth Intelligence