Browser Use growth playbook

Update the Browserbase post to the V2 run, with its date, its site count and a link to /benchmarks/browsers and the GitHub results. Then decide which page answers "Browser Use vs Browserbase": keep the post for that query and /alternatives/browserbase for "Browserbase alternatives", and cross-link the post, the alternatives page and /browserbase-pricing both ways. Do the same for the Kernel and Browserless pairs, which do not link to each other either. Finally, render benchmark numbers on all comparison pages from one shared data file, so the next run updates every page at once.

Comparison searches come from buyers who are close to choosing, and developers check numbers. One consistent, dated figure that links to the raw results is more credible than a bigger gap on one page and a smaller one on the next. A shared data source also stops the problem from coming back with each new benchmark run.

[ ] Browserbase post shows Stealth Bench V2 (Sep 25, 2026): 90.5% vs 78.5%
[ ] Run date, site count and GitHub results link next to every figure
[ ] Post <-> /alternatives/<rival> <-> pricing page links for Browserbase, Kernel, Browserless
[ ] All figures rendered from one shared data file
[ ] "Prices checked on" date on every rival price
[ ] No v1 figures left in /posts
[ ] Titles target distinct queries (vs / alternatives / pricing)

• The V2 gap over Browserbase (12 points) is smaller than the v1 gap (39 points)

  Lead with price and the reproducible results; a smaller, consistent, dated gap is more credible.

• Search engines treat the post and alternatives page as competing

  Give them clearly different queries and titles, and cross-link.

• Your website"Browser Use vs Browserbase" post (published September 21, updated October 26, 2026): stealth success "81% / 84.8%" for Browser Use against "42% / 70.3%" for Browserbase, from Stealth Bench v1 and BrowserBench, with no date given for either stealth run; it links to other "vs" posts but not to /alternatives/browserbase.Source
• Your website/alternatives/browserbase and /browserbase-pricing: Stealth Bench V2 (September 25, 2026) has Browser Use at 90.5% against Browserbase at 78.5%; /browserbase-pricing estimates 50 parallel browsers, 8 hours a day for 22 days, at $929 a month on Browserbase against $176 on Browser Use, before proxy traffic. The "Browser Use vs Kernel" and "Browser Use vs Browserless" posts already use the V2 figures (90.5% against 80.5% and 77.5%).Source
• Your websiteSitemap (652 URLs, checked November 9, 2026): /webmcp 238, /skills 120, /showcase 92, /posts 65, /benchmarks 36, /tools 29, /alternatives 6; Browserbase, Browserless and Kernel each have both an /alternatives/<rival> page and a /posts/browser-use-vs-<rival> post (plus /browserbase-pricing), and the Kernel and Browserless posts do not link to their /alternatives page; other overlaps include /cloud-browsers and /cloud-browser, and /mcp plus four "best MCP servers" posts.Source
• Your websiteHomepage: H1 "Browser agents and cloud browsers."; three products: Web agents ("$0.006 per task, median"), Stealth browsers ("$0.02 per browser hour") and Browser Harness ("Free and open source"); "Everything we sell is benchmarked. In public."; "Start with $1 free. One per person. Never expires."; a strip of 33 company logos (including Apple, Airbnb, Stripe, Uber, Shopify and DHL) with no label, shown twice, and no customer quotes.Source

Add a calculator to /pricing. Inputs: parallel browsers, hours a day, days a month, GB of home-IP traffic, and agent tasks. Outputs: monthly cost, the concurrency tier needed and how much total spend unlocks it, and the same workload priced on Browserbase, Kernel and Anchor using the dated prices you already publish. Give each result a shareable URL, and offer "Talk to sales" when the workload passes 1,000 browsers.

Usage pricing is cheap but hard to forecast, and engineers have to justify a vendor to whoever pays. A number they can paste into a ticket removes that friction, and the side-by-side turns the price lead into something they can see.

• Rival prices change

  Show the date checked and refresh monthly.

• Estimates differ from real bills

  Label it an estimate and link to billing details.

• #1 Your "Browser Use vs Browserbase" post still says 81% vs 42%; give every Browserbase page the same, dated benchmark

• Your websitePricing page (checked November 9, 2026): pay as you go at "$0 per month"; concurrency unlocked by total paid: 10 browsers at $0, 50 at $100, 250 at $1,000, 500 at $5,000, 1,000 at $25,000; a one-time $1 signup credit and credits from $5; home-IP proxy $5/GB (against Anchor $8, Browserbase $10-12, Kernel unmetered from $30/mo), direct or own proxy $0.20/GB; no interactive calculator, only one static example totalling $9.50; no annual plan below Enterprise, which adds "annual credit pools", custom SLAs, HIPAA, zero data retention, EU residency and on-prem.Source
• Your website/alternatives/browserbase and /browserbase-pricing: Stealth Bench V2 (September 25, 2026) has Browser Use at 90.5% against Browserbase at 78.5%; /browserbase-pricing estimates 50 parallel browsers, 8 hours a day for 22 days, at $929 a month on Browserbase against $176 on Browser Use, before proxy traffic. The "Browser Use vs Kernel" and "Browser Use vs Browserless" posts already use the V2 figures (90.5% against 80.5% and 77.5%).Source
• Competitor researchRivals' own homepages: Browserbase claims 10,000+ companies, 2.9M weekly SDK downloads and the open-source Stagehand, plus Search and Fetch APIs; Kernel claims 11,000+ teams, a free plan with $5 a month of credits and Hobbyist at $30 a month, and cites a throughput benchmark with Kernel at 6.43 actions/sec and Browser Use at 3.61; Steel offers "$30 free credits", claims 30,000+ engineers and an open-source browser with 7.9K stars.See the research Source

Give the strip an accurate label for what it represents (for example paying customers, or teams whose engineers use the open-source library). Link every logo that has a case study to it. Show the strip, with the same label and two short quotes, on the enterprise page. Render it once instead of twice.

Enterprise and security reviewers ask what a logo means. An accurate label backed by links turns a decorative strip into proof they can repeat internally, and it avoids the doubt an unlabeled strip of famous brands can create.

• A narrower label means fewer logos

  Fewer logos with a clear meaning beat many without one.

• Your websiteHomepage: H1 "Browser agents and cloud browsers."; three products: Web agents ("$0.006 per task, median"), Stealth browsers ("$0.02 per browser hour") and Browser Harness ("Free and open source"); "Everything we sell is benchmarked. In public."; "Start with $1 free. One per person. Never expires."; a strip of 33 company logos (including Apple, Airbnb, Stripe, Uber, Shopify and DHL) with no label, shown twice, and no customer quotes.Source
• Your websiteEnterprise page: "Browser automation at scale.", 90.5% against 82% for the next provider, 1,000+ concurrent browsers, secret placeholders, a 1Password integration, domain allow and deny lists, retention that can be turned off, SOC 2 Type II, HIPAA, DPA, on-prem and EU residency; no customer logos or quotes; the form asks for runs per month (up to 100,000+), start date ("today" to "no date yet") and compliance needs; no response time is promised.Source
• Your websiteCase studies: five stories (Hermes Agent, Sauna, Parallel, Vesta, Composio); only Hermes publishes numbers (48-66% fewer task tokens, 204 live runs, task accuracy at parity, "Hermes made Browser Use its default"); Sauna, Parallel, Vesta ("Mortgage evidence from websites without APIs") and Composio publish none.Source

For each of the four, add three measured figures from the customer's own usage: pages or tasks per month, success rate on protected sites, and cost or time saved against what they did before. Add one quote. Make "collect a number and a quote" part of every forward-deployed engagement, so each deployment produces a story.

Developers and buyers trust measured results from a named peer more than vendor benchmarks, and the forward-deployed role creates a steady supply of them.

• Customers won't share numbers

  Offer ratios or ranges instead of absolute figures.

• Your websiteCase studies: five stories (Hermes Agent, Sauna, Parallel, Vesta, Composio); only Hermes publishes numbers (48-66% fewer task tokens, 204 live runs, task accuracy at parity, "Hermes made Browser Use its default"); Sauna, Parallel, Vesta ("Mortgage evidence from websites without APIs") and Composio publish none.Source
• HiringCareers page (checked November 9, 2026): six people at an eight-figure annualized run rate, $17M raised, 120k+ GitHub stars, and customer demand growing faster than the team can keep up. Open roles, all in person in San Francisco: Forward-Deployed Engineer ($200K-$300K plus 0.5%-1.2% equity; "Talk to customers. Fix their problems. Ship the product."), Infrastructure Engineer ($200K-$400K plus 0.5%-1.4% equity; "Make one browser boring. Make millions possible."), plus "Browser God" ("Bring your insane browser-hacking skills.") and "Vibe Coder".Source

Route answers by those two fields. "Today" or "next 7 days" with 10,000+ runs goes straight to a founder's phone or Slack with a one-hour target, plus an instant email with the SOC 2 request link and a calendar link. Everyone else gets the self-serve path and a reply within one business day. Show "We reply within one business day; same hour if you're starting today" under the button.

Speed of first reply is one of the strongest drivers of enterprise conversion, and the form already sorts leads for free. A stated response time also lifts submissions.

• Founders get interrupted

  Only the top band triggers an alert; hand over to forward-deployed engineers once hired.

• Your websiteEnterprise page: "Browser automation at scale.", 90.5% against 82% for the next provider, 1,000+ concurrent browsers, secret placeholders, a 1Password integration, domain allow and deny lists, retention that can be turned off, SOC 2 Type II, HIPAA, DPA, on-prem and EU residency; no customer logos or quotes; the form asks for runs per month (up to 100,000+), start date ("today" to "no date yet") and compliance needs; no response time is promised.Source
• HiringCareers page (checked November 9, 2026): six people at an eight-figure annualized run rate, $17M raised, 120k+ GitHub stars, and customer demand growing faster than the team can keep up. Open roles, all in person in San Francisco: Forward-Deployed Engineer ($200K-$300K plus 0.5%-1.2% equity; "Talk to customers. Fix their problems. Ship the product."), Infrastructure Engineer ($200K-$400K plus 0.5%-1.4% equity; "Make one browser boring. Make millions possible."), plus "Browser God" ("Bring your insane browser-hacking skills.") and "Vibe Coder".Source

Cut main-thread work on the homepage: split or defer the 261 KB of unused JavaScript, load Google Tag Manager after interaction or idle, inline critical CSS to remove the 600 ms of render-blocking requests, and render the 33-logo strip once with a CSS loop instead of a duplicated copy. Re-test in PageSpeed and watch the field INP over 28 days.

The homepage is 1 ms from passing, so a modest cut in blocking time should move it into the "good" band. A browser-infrastructure company whose own homepage passes speed checks is a small but visible signal to developers.

• Analytics loses early events

  Queue events until the tag loads.

• PageSpeedHomepage (November 9, 2026): mobile score 68, desktop 74. Real Chrome visitors to the homepage: LCP 2.1 s, INP 201 ms (just over Google's 200 ms line, so the page fails Core Web Vitals on INP), CLS 0; the whole origin's INP is 133 ms. Lab mobile: Total Blocking Time 2,990 ms, 423 KB of JavaScript with 261 KB unused, 600 ms of render-blocking requests, Google Tag Manager 190 KB, 994 DOM elements.Source
• Your websiteHomepage: H1 "Browser agents and cloud browsers."; three products: Web agents ("$0.006 per task, median"), Stealth browsers ("$0.02 per browser hour") and Browser Harness ("Free and open source"); "Everything we sell is benchmarked. In public."; "Start with $1 free. One per person. Never expires."; a strip of 33 company logos (including Apple, Airbnb, Stripe, Uber, Shopify and DHL) with no label, shown twice, and no customer quotes.Source

Add a run card to every showcase page: model, steps, duration, browser minutes, proxy MB and total cost (for example "$0.004"), plus a five-line Python and TypeScript snippet that runs the same task through the API. Record runs for the 42 examples without one, or label them "prompt only".

A visitor who sees that a multi-city flight search cost less than a cent, and can paste the code, has the two things they need to try it. It also makes the low price real for each task instead of a median on the homepage.

• Costs vary run to run

  Show the median of three runs and the date.

• Your websiteShowcase: 81 examples in 9 categories, badge "39 recorded runs · exact prompts"; a detail page (multi-city flight) has the prompt, a Copy button and "Run this task free", which opens the cloud app prefilled, but no code snippet, no model and no cost for the run.Source
• Your websitePricing page (checked November 9, 2026): pay as you go at "$0 per month"; concurrency unlocked by total paid: 10 browsers at $0, 50 at $100, 250 at $1,000, 500 at $5,000, 1,000 at $25,000; a one-time $1 signup credit and credits from $5; home-IP proxy $5/GB (against Anchor $8, Browserbase $10-12, Kernel unmetered from $30/mo), direct or own proxy $0.20/GB; no interactive calculator, only one static example totalling $9.50; no annual plan below Enterprise, which adds "annual credit pools", custom SLAs, HIPAA, zero data retention, EU residency and on-prem.Source
• Your websiteDevelopers page: "Two ways to build with Browser Use."; no numbered quickstart and no time-to-first-run claim; describes how to "Save a successful agent run and replay it without a model call per step"; two onboarding examples in the GitHub repo.Source

Keep the cost the same and change the framing and the path. Show the credit as "enough for about 150 agent tasks or 50 browser hours". Then guide a new account through a first-run checklist: run one showcase task, attach Playwright to a stealth browser, save and replay a run. Give a small top-up (for example $4) when someone finishes the checklist, and, for GitHub sign-ups with a public repo using browser-use, a one-time bonus.

Activation depends on reaching a working result, not on the size of the credit. Rewarding the steps that lead to retention spends a few dollars where they change behaviour, and the reframing answers the rivals' bigger headline numbers.

• Credit farming

  Tie the bonus to the existing one-per-person checks and a real run.

• #7 Show the model, cost and code on every showcase run

• Your websitePricing page (checked November 9, 2026): pay as you go at "$0 per month"; concurrency unlocked by total paid: 10 browsers at $0, 50 at $100, 250 at $1,000, 500 at $5,000, 1,000 at $25,000; a one-time $1 signup credit and credits from $5; home-IP proxy $5/GB (against Anchor $8, Browserbase $10-12, Kernel unmetered from $30/mo), direct or own proxy $0.20/GB; no interactive calculator, only one static example totalling $9.50; no annual plan below Enterprise, which adds "annual credit pools", custom SLAs, HIPAA, zero data retention, EU residency and on-prem.Source
• Open sourceGitHub (browser-use/browser-use): 120.4k stars, 13.5k forks, MIT licence; README: "The open-source browser agent. Run it locally with any LLM, or on Browser Use Cloud."; new signups get a "$1 cloud credit"; the Python library's 1.0 is a release candidate.Source
• Competitor researchRivals' own homepages: Browserbase claims 10,000+ companies, 2.9M weekly SDK downloads and the open-source Stagehand, plus Search and Fetch APIs; Kernel claims 11,000+ teams, a free plan with $5 a month of credits and Hobbyist at $30 a month, and cites a throughput benchmark with Kernel at 6.43 actions/sec and Browser Use at 3.61; Steel offers "$30 free credits", claims 30,000+ engineers and an open-source browser with 7.9K stars.See the research Source

On every directory page, add two or three task prompts generated from the site's own tools (for example "Search products on <site> and return the top 10 as JSON"), each with the showcase-style "Run this task free" button. Where a skill exists for that site, link it. Add the same block to the category pages, and track runs by directory page.

It turns a reference directory into thousands of specific starting points, each matching what the visitor came to do. No rival has an equivalent library, and the button already exists.

• Prompts fail on some sites

  Only show prompts that passed a recent test run.

• Site owners object to automation

  Use only the tools the site itself exposes to agents.

• Your websiteWebMCP directory: "13,352 websites with tools for AI agents", 45,763 distinct tools, scan finished October 19, 2026; each site has its own page; the only calls to action are the site-wide "Start for free" and "Talk to sales".Source
• Your websiteShowcase: 81 examples in 9 categories, badge "39 recorded runs · exact prompts"; a detail page (multi-city flight) has the prompt, a Copy button and "Run this task free", which opens the cloud app prefilled, but no code snippet, no model and no cost for the run.Source
• Your websiteSitemap (652 URLs, checked November 9, 2026): /webmcp 238, /skills 120, /showcase 92, /posts 65, /benchmarks 36, /tools 29, /alternatives 6; Browserbase, Browserless and Kernel each have both an /alternatives/<rival> page and a /posts/browser-use-vs-<rival> post (plus /browserbase-pricing), and the Kernel and Browserless posts do not link to their /alternatives page; other overlaps include /cloud-browsers and /cloud-browser, and /mcp plus four "best MCP servers" posts.Source

Pay a cloud credit for every merged skill (for example $20), show the author on the skill page with a link to their GitHub, and publish a monthly "top skills" list by cloud runs. On every skill page, add a "Run with this skill" button that opens the cloud app with the skill loaded and a sample task.

Small, visible rewards grow open-source contribution, and every new skill is another page for a specific site that can start a run. Contributors who hold credits become cloud users.

• Low-quality submissions

  Pay only on merge after a passing run.

• Your websiteSkills: 134 skills for 119 websites that "tell an AI agent how a site works before it opens it", "Written by browser-harness contributors after real runs", contributed by GitHub pull request, MIT licensed, last updated November 8, 2026.Source
• Open sourceGitHub (browser-use/browser-use): 120.4k stars, 13.5k forks, MIT licence; README: "The open-source browser agent. Run it locally with any LLM, or on Browser Use Cloud."; new signups get a "$1 cloud credit"; the Python library's 1.0 is a release candidate.Source

Under each tool's result, add a short panel: "Run this on 10,000 URLs" with a Python and TypeScript snippet that does the same job through a stealth browser or agent, its estimated cost at $0.02 per browser hour, and a sign-up button. On the Bot Detection Test, show the visitor's result next to a Browser Use stealth browser's result.

A visitor who just used the tool has the problem right now. A matching snippet with a price is the shortest path from a one-off check to a production job.

• The panel feels pushy

  Show it only after the free result appears.

• Your websiteFree tools: 28 browser utilities "free to use without an account" (Website to Markdown, HAR viewer, Bot Detection Test, robots.txt tester, llms.txt checker, Fetch as Googlebot and more); no tool has its own call to action to the API.Source
• Your websitePricing page (checked November 9, 2026): pay as you go at "$0 per month"; concurrency unlocked by total paid: 10 browsers at $0, 50 at $100, 250 at $1,000, 500 at $5,000, 1,000 at $25,000; a one-time $1 signup credit and credits from $5; home-IP proxy $5/GB (against Anchor $8, Browserbase $10-12, Kernel unmetered from $30/mo), direct or own proxy $0.20/GB; no interactive calculator, only one static example totalling $9.50; no annual plan below Enterprise, which adds "annual credit pools", custom SLAs, HIPAA, zero data retention, EU residency and on-prem.Source

When the library detects a block page, CAPTCHA or repeated access denial, print one line in the log: "This site blocked a local browser. Run it on a Browser Use stealth browser with one setting; new accounts get $1 free.", plus the one-line config change. Show it at most once per session, let users turn it off with an environment variable, and announce the change in the release notes.

A message at the moment of failure reaches the right developer at the right time, inside the tool they already trust, and the one-line switch makes trying the cloud nearly free of effort.

• The community sees it as advertising

  Show it only on failure, once, with a clear opt-out, and say so openly.

• Open sourceGitHub (browser-use/browser-use): 120.4k stars, 13.5k forks, MIT licence; README: "The open-source browser agent. Run it locally with any LLM, or on Browser Use Cloud."; new signups get a "$1 cloud credit"; the Python library's 1.0 is a release candidate.Source
• Your websiteBrowser benchmark (September 25, 2026): 100 bot-protected sites, each visited twice; Browser Use 90.5%, Anchor 82%, Kernel 80.5%, Browserbase 78.5%, Browserless 77.5%, Steel 74.5%, Hyperbrowser 73.5%; the sites were chosen where Browser Use had got through before and Chrome with a US proxy had not; the runner and every visit's result are on GitHub.Source

List every query that has two or more pages. For each, keep the strongest page, merge the best content from the others into it, 301-redirect the rest, and update internal links. Keep a simple register of "query to page" that new posts are checked against before publishing.

One strong page usually outranks two weaker ones that split links and clicks, and a register stops new duplicates as publishing continues.

• Temporary ranking drop

  Redirect to the closest page and merge content, not just the URL.

• #1 Your "Browser Use vs Browserbase" post still says 81% vs 42%; give every Browserbase page the same, dated benchmark

• Your websiteSitemap (652 URLs, checked November 9, 2026): /webmcp 238, /skills 120, /showcase 92, /posts 65, /benchmarks 36, /tools 29, /alternatives 6; Browserbase, Browserless and Kernel each have both an /alternatives/<rival> page and a /posts/browser-use-vs-<rival> post (plus /browserbase-pricing), and the Kernel and Browserless posts do not link to their /alternatives page; other overlaps include /cloud-browsers and /cloud-browser, and /mcp plus four "best MCP servers" posts.Source
• Your website"Browser Use vs Browserbase" post (published September 21, updated October 26, 2026): stealth success "81% / 84.8%" for Browser Use against "42% / 70.3%" for Browserbase, from Stealth Bench v1 and BrowserBench, with no date given for either stealth run; it links to other "vs" posts but not to /alternatives/browserbase.Source

Offer self-serve prepaid packs, for example $1,000, $5,000 and $25,000, bought by card or invoice in the dashboard. Each counts immediately toward the matching concurrency tier, adds a bonus (say 10-15% extra credit), never expires, and comes with a receipt finance teams can file. Keep Enterprise for SLAs, compliance and on-prem.

Prepaid packs bring cash forward, lock in usage before a team tries a rival, and give self-serve customers the concurrency jump they would otherwise need sales for, without adding sales work.

• Discount cuts margin

  Keep the bonus small and tied to larger packs.

• #2 A cost calculator on the pricing page, built on the concurrency ladder and the Browserbase math you already publish

• Your websitePricing page (checked November 9, 2026): pay as you go at "$0 per month"; concurrency unlocked by total paid: 10 browsers at $0, 50 at $100, 250 at $1,000, 500 at $5,000, 1,000 at $25,000; a one-time $1 signup credit and credits from $5; home-IP proxy $5/GB (against Anchor $8, Browserbase $10-12, Kernel unmetered from $30/mo), direct or own proxy $0.20/GB; no interactive calculator, only one static example totalling $9.50; no annual plan below Enterprise, which adds "annual credit pools", custom SLAs, HIPAA, zero data retention, EU residency and on-prem.Source
• HiringCareers page (checked November 9, 2026): six people at an eight-figure annualized run rate, $17M raised, 120k+ GitHub stars, and customer demand growing faster than the team can keep up. Open roles, all in person in San Francisco: Forward-Deployed Engineer ($200K-$300K plus 0.5%-1.2% equity; "Talk to customers. Fix their problems. Ship the product."), Infrastructure Engineer ($200K-$400K plus 0.5%-1.4% equity; "Make one browser boring. Make millions possible."), plus "Browser God" ("Bring your insane browser-hacking skills.") and "Vibe Coder".Source

Detect accounts that run near-identical agent tasks repeatedly. Show them in the dashboard and by email: "This task ran 412 times last month at $X; replaying the saved run would cost about $Y.", with a one-click "save and replay" for that task. Add a monthly "you saved" line for accounts that already replay.

Customers who see a vendor cutting their bill stay, and they move more work onto the platform. It also builds a cost-per-workflow story sales can use.

• Revenue per task falls

  Browser hours remain billed, and retention and volume should offset it.

• Your websiteDevelopers page: "Two ways to build with Browser Use."; no numbered quickstart and no time-to-first-run claim; describes how to "Save a successful agent run and replay it without a model call per step"; two onboarding examples in the GitHub repo.Source
• Your websitePricing page (checked November 9, 2026): pay as you go at "$0 per month"; concurrency unlocked by total paid: 10 browsers at $0, 50 at $100, 250 at $1,000, 500 at $5,000, 1,000 at $25,000; a one-time $1 signup credit and credits from $5; home-IP proxy $5/GB (against Anchor $8, Browserbase $10-12, Kernel unmetered from $30/mo), direct or own proxy $0.20/GB; no interactive calculator, only one static example totalling $9.50; no annual plan below Enterprise, which adds "annual credit pools", custom SLAs, HIPAA, zero data retention, EU residency and on-prem.Source

Launch a simple program: a "Runs on Browser Use" badge and docs snippet for partner products, a partner page listing them, a revenue share or credit for customers who sign up through a partner link, and co-written launch posts. Start with the published case-study platforms and the existing integration pages.

Platforms that embed a vendor reach thousands of developers at once. A badge and a referral share turn hidden infrastructure into a visible, trackable channel.

• Partners want to keep their users

  Make the referral optional and focus on badges and launches first.

• #4 Put numbers in four of the five case studies, collected by the forward-deployed engineer

• Your websiteCase studies: five stories (Hermes Agent, Sauna, Parallel, Vesta, Composio); only Hermes publishes numbers (48-66% fewer task tokens, 204 live runs, task accuracy at parity, "Hermes made Browser Use its default"); Sauna, Parallel, Vesta ("Mortgage evidence from websites without APIs") and Composio publish none.Source
• Your websiteIntegration and use-case pages in the sitemap: /n8n, /openclaw, /playwright, /puppeteer, /mcp, /web-scraping, /automated-testing, /form-filling, /voice-agents and /login-automation; the Composio story is titled "Composio Browser Tool runs on Browser Use."Source

Post a short changelog entry every week (new models, speed, pricing changes, fixes), even when small, and send it as an opt-in email to API users with a one-line "what this means for your bill or success rate". Link the release notes for the library from the same page.

Developers who see steady progress trust the vendor more and churn less. A weekly cadence also gives the content team a ready source of posts.

• Weeks with little to report

  Include fixes, benchmarks and new skills.

• Your websiteChangelog: entries dated October 30 ("BU Ultrafast Is Live"), August 24, August 17, August 10, July 20 and July 17, 2026, with nothing between August 24 and October 30, while /posts published more than 20 articles between late September and November 9.Source
• Open sourceGitHub (browser-use/browser-use): 120.4k stars, 13.5k forks, MIT licence; README: "The open-source browser agent. Run it locally with any LLM, or on Browser Use Cloud."; new signups get a "$1 cloud credit"; the Python library's 1.0 is a release candidate.Source

Fund a quarterly benchmark run by an outside party (a university lab or an independent researcher), with a site list drawn at random each quarter from a public pool and frozen before any provider runs. Publish the runner, raw results and every provider's score, including categories where Browser Use loses. Invite rivals to review the method in advance.

A neutral benchmark that Browser Use still wins is far stronger proof than a self-chosen one, and publishing losses makes the wins believable. It can become the reference the category quotes.

• Browser Use's lead shrinks

  Then the product team knows where to work, and the honest result still earns trust.

• #1 Your "Browser Use vs Browserbase" post still says 81% vs 42%; give every Browserbase page the same, dated benchmark

• Your websiteBrowser benchmark (September 25, 2026): 100 bot-protected sites, each visited twice; Browser Use 90.5%, Anchor 82%, Kernel 80.5%, Browserbase 78.5%, Browserless 77.5%, Steel 74.5%, Hyperbrowser 73.5%; the sites were chosen where Browser Use had got through before and Chrome with a US proxy had not; the runner and every visit's result are on GitHub.Source
• Competitor researchRivals' own homepages: Browserbase claims 10,000+ companies, 2.9M weekly SDK downloads and the open-source Stagehand, plus Search and Fetch APIs; Kernel claims 11,000+ teams, a free plan with $5 a month of credits and Hobbyist at $30 a month, and cites a throughput benchmark with Kernel at 6.43 actions/sec and Browser Use at 3.61; Steel offers "$30 free credits", claims 30,000+ engineers and an open-source browser with 7.9K stars.See the research Source
• Your websiteHomepage: H1 "Browser agents and cloud browsers."; three products: Web agents ("$0.006 per task, median"), Stealth browsers ("$0.02 per browser hour") and Browser Harness ("Free and open source"); "Everything we sell is benchmarked. In public."; "Start with $1 free. One per person. Never expires."; a strip of 33 company logos (including Apple, Airbnb, Stripe, Uber, Shopify and DHL) with no label, shown twice, and no customer quotes.Source

Package the first engagement: a two-week paid pilot at a fixed price (credited against the first annual contract). It covers one workflow, built on the customer's real sites by a forward-deployed engineer, with a success target agreed on day one (for example 95% task completion), a written security pack (SOC 2 report, DPA, data-retention settings), and a go/no-go review on day 14. Publish the pilot on the enterprise page with its price and steps.

A fixed scope and price shortens the buying process, filters out tyre-kickers, and gives forward-deployed engineers a repeatable playbook instead of custom work every time. Paid pilots convert better than free ones because the buyer has committed budget.

• Pilots eat engineering capacity

  Cap concurrent pilots at the number of forward-deployed engineers.

• #5 Reply to "start today" enterprise forms within the hour, and say so on the form

• HiringCareers page (checked November 9, 2026): six people at an eight-figure annualized run rate, $17M raised, 120k+ GitHub stars, and customer demand growing faster than the team can keep up. Open roles, all in person in San Francisco: Forward-Deployed Engineer ($200K-$300K plus 0.5%-1.2% equity; "Talk to customers. Fix their problems. Ship the product."), Infrastructure Engineer ($200K-$400K plus 0.5%-1.4% equity; "Make one browser boring. Make millions possible."), plus "Browser God" ("Bring your insane browser-hacking skills.") and "Vibe Coder".Source
• Your websiteEnterprise page: "Browser automation at scale.", 90.5% against 82% for the next provider, 1,000+ concurrent browsers, secret placeholders, a 1Password integration, domain allow and deny lists, retention that can be turned off, SOC 2 Type II, HIPAA, DPA, on-prem and EU residency; no customer logos or quotes; the form asks for runs per month (up to 100,000+), start date ("today" to "no date yet") and compliance needs; no response time is promised.Source
• Your websitePricing page (checked November 9, 2026): pay as you go at "$0 per month"; concurrency unlocked by total paid: 10 browsers at $0, 50 at $100, 250 at $1,000, 500 at $5,000, 1,000 at $25,000; a one-time $1 signup credit and credits from $5; home-IP proxy $5/GB (against Anchor $8, Browserbase $10-12, Kernel unmetered from $30/mo), direct or own proxy $0.20/GB; no interactive calculator, only one static example totalling $9.50; no annual plan below Enterprise, which adds "annual credit pools", custom SLAs, HIPAA, zero data retention, EU residency and on-prem.Source

Package "web evidence": an agent that logs into the portals a lender, insurer or benefits team needs, collects the required pages and documents, and returns time-stamped PDFs and structured JSON with a run recording for the audit trail. Ship templates for the most common evidence types, a compliance pack, and per-file pricing. Build two more stories with Vesta-like customers, then sell into that vertical.

Regulated teams pay for outcomes and audit trails, not browser hours, and per-file pricing captures more of the value. Browser Use already has a published proof point and the compliance features to back it.

• Verticals pull focus from the platform

  Build templates on the public API so the work benefits every customer.

• Portal terms of use

  Run only on portals the customer is authorised to access, with their credentials.

• #4 Put numbers in four of the five case studies, collected by the forward-deployed engineer
• #19 A fixed two-week enterprise pilot run by forward-deployed engineers

• Your websiteCase studies: five stories (Hermes Agent, Sauna, Parallel, Vesta, Composio); only Hermes publishes numbers (48-66% fewer task tokens, 204 live runs, task accuracy at parity, "Hermes made Browser Use its default"); Sauna, Parallel, Vesta ("Mortgage evidence from websites without APIs") and Composio publish none.Source
• Your websiteEnterprise page: "Browser automation at scale.", 90.5% against 82% for the next provider, 1,000+ concurrent browsers, secret placeholders, a 1Password integration, domain allow and deny lists, retention that can be turned off, SOC 2 Type II, HIPAA, DPA, on-prem and EU residency; no customer logos or quotes; the form asks for runs per month (up to 100,000+), start date ("today" to "no date yet") and compliance needs; no response time is promised.Source

Combine them into one free report: enter a domain, get an "agent-ready" score covering WebMCP tools, llms.txt, robots rules for agents, whether a Browser Use agent can complete a common task on the site, and page-to-markdown quality, with fixes. Let owners embed a badge and claim their WebMCP directory entry. Offer agent-builder customers alerts when a site they depend on changes.

Shareable scores spread on their own, every report adds a site to the directory, and the people who run websites at big companies are often the same engineers who will later buy browser infrastructure.

• Low scores annoy site owners

  Only publish scores for sites owners run themselves, except in an aggregate launch post.

• #11 Give each free tool a "do this at scale" API snippet

• Your websiteFree tools: 28 browser utilities "free to use without an account" (Website to Markdown, HAR viewer, Bot Detection Test, robots.txt tester, llms.txt checker, Fetch as Googlebot and more); no tool has its own call to action to the API.Source
• Your websiteWebMCP directory: "13,352 websites with tools for AI agents", 45,763 distinct tools, scan finished October 19, 2026; each site has its own page; the only calls to action are the site-wide "Start for free" and "Talk to sales".Source
• Your websiteSkills: 134 skills for 119 websites that "tell an AI agent how a site works before it opens it", "Written by browser-harness contributors after real runs", contributed by GitHub pull request, MIT licensed, last updated November 8, 2026.Source

Publish the protected sites where stealth browsers currently fail most, taken from the benchmark and from opt-in customer failure data, and pay a bounty to anyone who submits a fix that gets the cloud browser through and passes review. Pay in cash or credits, credit winners publicly, and fast-track top contributors into the hiring process.

It turns the hardest product problem into a public challenge that recruits talent, improves the success rate and creates content, for less than a single hire's recruiting fee.

• Publicising how protections are bypassed

  Keep fixes private; publish only the site, the result and the winner.

• HiringCareers page (checked November 9, 2026): six people at an eight-figure annualized run rate, $17M raised, 120k+ GitHub stars, and customer demand growing faster than the team can keep up. Open roles, all in person in San Francisco: Forward-Deployed Engineer ($200K-$300K plus 0.5%-1.2% equity; "Talk to customers. Fix their problems. Ship the product."), Infrastructure Engineer ($200K-$400K plus 0.5%-1.4% equity; "Make one browser boring. Make millions possible."), plus "Browser God" ("Bring your insane browser-hacking skills.") and "Vibe Coder".Source
• Your websiteBrowser benchmark (September 25, 2026): 100 bot-protected sites, each visited twice; Browser Use 90.5%, Anchor 82%, Kernel 80.5%, Browserbase 78.5%, Browserless 77.5%, Steel 74.5%, Hyperbrowser 73.5%; the sites were chosen where Browser Use had got through before and Chrome with a US proxy had not; the runner and every visit's result are on GitHub.Source