Manicule growth playbook

Ship /llms.txt this week, listing the canonical pages (/agentrel, /agent-visibility, /services, both case studies, /manifesto, the blog and the research) with one-line descriptions. Add /llms-full.txt and a .md version of each marketing page, reusing the Markdown export Margin already has. Add one plain "Facts" block that agents can quote: plans and prices, assistants tracked today versus "soon", trial terms, what services include, and how long a docs rebuild takes, worded the same way on every page. Then track 20 prompts about Manicule and AgentRel in Manicule's own app before and after, and publish the result against the noise band as Manicule's first case study of itself.

It is a same-day fix that removes a credibility gap in the sales conversation, and it turns the site into a live demo: prospects can see Manicule do on its own site exactly what it sells, with before-and-after numbers from its own tool.

[ ] /llms.txt with canonical pages and one-line descriptions
[ ] /llms-full.txt
[ ] .md version of every marketing page
[ ] One facts block: plans, prices, assistants tracked now and soon, trial, docs rebuild time
[ ] Same claims on homepage, /agent-visibility and /services
[ ] lastmod on every sitemap URL
[ ] 20 Manicule prompts tracked before and after
[ ] Before/after published as a case study

• Assistants are slow to pick up the change

  Report movement only outside the noise band and keep the 60-day window.

• The facts block drifts out of date as plans change

  Generate it from the same source as the plan cards.

• Your websiteAgent-readable files (checked twice on November 9, 2026): manicule.com/llms.txt and manicule.com/llms-full.txt both return 404 Not Found; robots.txt allows all crawlers and points to the sitemap.Source
• Your website"What is AgentRel?" page: "AgentRel (Agent Relations) is the work of making your product easy for AI agents to discover, recommend, and use." Its "Found" tactics include publishing docs as plain HTML or markdown with an llms.txt; "Chosen" includes getting listed on comparison pages and publishing original benchmarks; "Used" includes offering an MCP server or CLI. It has a table comparing SEO, AEO/GEO, DevRel and AgentRel and a six-question FAQ, but no author or date.Source
• Your websiteMargin: "The open-source docs framework"; "Bring your Mintlify docs as they are."; every page as Markdown, with llms.txt and an MCP server on by default, plus /llms-full.txt and /AGENTS.md; hosted free on Manicule with agent analytics, or self-hosted under Apache 2.0; Margin has no web editor and "For writing, there's Manicule". The homepage says "Docs on Margin report to Manicule."Source
• Your websiteSitemap (16 URLs, November 9, 2026): home, /services, /agentrel, /agent-visibility, /manifesto, two case studies (both Supermemory), /blog with one post, /careers and six legal pages; only the blog post carries a lastmod date; there is no /pricing page (it returns 404), no comparison pages and no research page, and the research PDF is not listed.Source
• Your websiteAgent visibility page: headline "Your AI visibility tool is lying to you."; Starter $79/mo (75 tracked prompts, 3 countries, "14-day free trial", "No card required"), Pro $199/mo (200 prompts, "1 more assistant soon") and Max $449/mo (600 prompts, "3 more assistants soon"); every plan lists ChatGPT, Gemini and AI Overviews and none lists Claude; only Starter shows a free trial; yearly billing "saves up to $600".Source

Pick one truth and use it everywhere: either name Claude on the plan cards with a date, or change the homepage to the three surfaces tracked today and say which assistants are coming. Then run the claim checker against manicule.com's own pages each month and fix anything it flags before a prospect does.

Technical buyers check claims, and a company selling honest measurement loses trust fastest on a visible inconsistency. Fixing it also removes a likely question from every sales call.

• Prospects read "no Claude" as a gap

  Pair the change with a dated plan for the next assistant.

• Your websiteHomepage (checked November 9, 2026): "Make your product the one AI agents discover. recommend. use."; a services offer ("We write, ship, and measure.") and a software card that says "Track how ChatGPT, Claude and Gemini describe your brand"; the "Used" pillar promises "We rebuild your docs in 14 days, check every snippet against your live API, then send coding agents through real tasks."; logos for Greptile, Supermemory, Rootly, PromptLayer, Skyvern, Surge, xpander.ai and Context.dev; no prices on the page.Source
• Your websiteAgent visibility page: headline "Your AI visibility tool is lying to you."; Starter $79/mo (75 tracked prompts, 3 countries, "14-day free trial", "No card required"), Pro $199/mo (200 prompts, "1 more assistant soon") and Max $449/mo (600 prompts, "3 more assistants soon"); every plan lists ChatGPT, Gemini and AI Overviews and none lists Claude; only Starter shows a free trial; yearly billing "saves up to $600".Source
• Your websiteAgent visibility features: a noise band so a drop alert fires only on a fall that chance alone would rarely produce; cited pages that mention competitors but not you, with draft outreach emails; claim checking against your facts; betas for agent experience (whether Claude Code and Codex can install and use your product), competitor watch and docs check; consulting sessions on every plan; "Switching from Profound? We import your prompts and history automatically, in under 30 seconds."; the page names Profound, Peec AI, Scrunch, Otterly.AI, AthenaHQ, Evertune, Bluefish, ZipTie, Rankscale, Ahrefs, Semrush and SE Visible; logos for Mastra, AgentMail, TinyFish, Skyvern and Greptile.Source

Create /pricing with the three software plans, the Starter trial, the yearly saving, a clear line on what services cost to start (a range or a fixed-scope offer, see idea 6), and a short FAQ: which assistants, how many runs per question, what a consulting session is. Link it in the top navigation and include it in llms.txt.

Visible pricing qualifies self-serve buyers and gives the new enterprise seller a reference point. It also gives agents one page to quote, so they stop guessing.

• A services price scares larger buyers

  Show "from" pricing and route enterprise to a call.

• Your websiteSitemap (16 URLs, November 9, 2026): home, /services, /agentrel, /agent-visibility, /manifesto, two case studies (both Supermemory), /blog with one post, /careers and six legal pages; only the blog post carries a lastmod date; there is no /pricing page (it returns 404), no comparison pages and no research page, and the research PDF is not listed.Source
• Your websiteAgent visibility page: headline "Your AI visibility tool is lying to you."; Starter $79/mo (75 tracked prompts, 3 countries, "14-day free trial", "No card required"), Pro $199/mo (200 prompts, "1 more assistant soon") and Max $449/mo (600 prompts, "3 more assistants soon"); every plan lists ChatGPT, Gemini and AI Overviews and none lists Claude; only Starter shows a free trial; yearly billing "saves up to $600".Source
• Your websiteServices page: content strategy, technical content, docs, original research and distribution across search, AI discovery, Reddit and developer communities; it cites the Supermemory docs rebuild as completed in 23 days; no prices, packages or engagement terms; the next step is to book a call.Source

Start with Manicule vs Profound and Manicule vs Peec AI: a fair table (assistants tracked, runs per question, noise band, agent-experience testing, devtools focus, price), how the import works, and when the rival is the better fit. Then add a "best AI visibility tools for devtools" roundup that includes Manicule honestly. Back each with one switching customer story when you have it.

Comparison searches are high intent, and assistants lean on comparison pages when asked to pick a tool. It is the method that already worked for Supermemory, applied to Manicule.

• Out-of-date or unfair claims

  Date every page, cite rivals' own pages and invite corrections.

• Your websiteAgent visibility features: a noise band so a drop alert fires only on a fall that chance alone would rarely produce; cited pages that mention competitors but not you, with draft outreach emails; claim checking against your facts; betas for agent experience (whether Claude Code and Codex can install and use your product), competitor watch and docs check; consulting sessions on every plan; "Switching from Profound? We import your prompts and history automatically, in under 30 seconds."; the page names Profound, Peec AI, Scrunch, Otterly.AI, AthenaHQ, Evertune, Bluefish, ZipTie, Rankscale, Ahrefs, Semrush and SE Visible; logos for Mastra, AgentMail, TinyFish, Skyvern and Greptile.Source
• Your websiteSupermemory (DevRel) case study, three months: +90% traffic, 1M+ SEO impressions, 200K+ Reddit views from one benchmark campaign, and #1 and #2 on Google for "mem0 vs supermemory", built around a customer who switched; the founder says "They wrote content that organically brings us millions of impressions on SEO."Source
• Competitor researchProfound's homepage: "The AI marketing platform to win in" ChatGPT, Perplexity, Claude, Gemini, Microsoft Copilot, DeepSeek and Google AI Overviews; serves marketing teams, agencies and partners; says a third of the Fortune 100 uses it; demo-led, with no prices on the homepage.See the research Source

Publish /research as HTML: the headline findings, the method, a chart per finding and a downloadable dataset sample. Write a short answer under each of the four questions on /agent-visibility, linking to the hub. Cut each finding into a standalone post for X, LinkedIn and Reddit, and relaunch on Hacker News with a plain, specific title and the HTML page.

Original data is what assistants and journalists cite, and HTML pages are what they can read. The four questions are the same doubts buyers bring to a sales call.

• Method gets challenged

  Publish the method and data sample and answer publicly.

• Your websiteResearch section "What we learned from a million AI answers": "We asked 173,554 buyer questions across 200 software categories on Claude, ChatGPT and Gemini." One answer misjudged a brand's rate by a median of 14 points, thirteen answers cut that to 7, and half of brand-question pairs needed 18 or more runs to settle. Four question headings ("How many runs before a score means anything?", "Does it matter who's asking?", "What happens when assistants disagree?", "Should questions that name you count?") have no answer text under them. The full study is only a PDF linked from the footer.Source
• Hacker NewsA Hacker News post linking the research PDF ("1,053,690 answers show that the current AEO landscape is...") drew 2 points and 1 comment; the comment's findings: ChatGPT changed its top recommendation 55% of the time when asked the same prompt twice, the three models agreed on the top brand in only 48% of markets, and adding buyer details changed the top pick 63-74% of the time.Source
• Your websiteSitemap (16 URLs, November 9, 2026): home, /services, /agentrel, /agent-visibility, /manifesto, two case studies (both Supermemory), /blog with one post, /careers and six legal pages; only the blog post carries a lastmod date; there is no /pricing page (it returns 404), no comparison pages and no research page, and the research PDF is not listed.Source

Package "Docs rebuild for agents": a fixed scope (audit, information architecture, endpoint docs from the OpenAPI spec, snippet tests against the live API, a coding-agent task run, Margin hosting optional), a fixed timeline that matches the case study, and a starting price. Put it on /services and /pricing with a before-and-after answer-success test included.

A named, priced, time-boxed offer is easier to buy and to forward internally than "book a call", and one consistent timeline protects trust.

• Fixed scope underprices complex docs

  Price by number of products or API surfaces.

• Your websiteHomepage (checked November 9, 2026): "Make your product the one AI agents discover. recommend. use."; a services offer ("We write, ship, and measure.") and a software card that says "Track how ChatGPT, Claude and Gemini describe your brand"; the "Used" pillar promises "We rebuild your docs in 14 days, check every snippet against your live API, then send coding agents through real tasks."; logos for Greptile, Supermemory, Rootly, PromptLayer, Skyvern, Surge, xpander.ai and Context.dev; no prices on the page.Source
• Your websiteSupermemory (Docs) case study: "23 days from kickoff to launch"; answer success rate +30%; "Multiple" enterprise deals closed; an outside reviewer: "Supermemory has some of the best docs I've ever seen."Source
• Your websiteServices page: content strategy, technical content, docs, original research and distribution across search, AI discovery, Reddit and developer communities; it cites the Supermemory docs rebuild as completed in 23 days; no prices, packages or engagement terms; the next step is to book a call.Source

Add the founder as author with a short bio, a published and updated date, a one-sentence definition in a quotable box, FAQ structured data, and a "cite this definition" line. Track "what is AgentRel" and "AgentRel vs DevRel" prompts in the app to see who assistants credit.

The company that defines a category tends to be named whenever it is explained, and it costs an afternoon.

• The term spreads without credit

  Use it consistently in every post and talk.

• Your website"What is AgentRel?" page: "AgentRel (Agent Relations) is the work of making your product easy for AI agents to discover, recommend, and use." Its "Found" tactics include publishing docs as plain HTML or markdown with an llms.txt; "Chosen" includes getting listed on comparison pages and publishing original benchmarks; "Used" includes offering an MCP server or CLI. It has a table comparing SEO, AEO/GEO, DevRel and AgentRel and a six-question FAQ, but no author or date.Source
• HiringEnterprise Sales (San Francisco, $150K-$250K): owns enterprise deals from discovery through close and expansion, selling to founders, CTOs, engineering leaders and developer-marketing teams who may not yet have a budget or vocabulary for the category; in the first 30 days the hire joins every founder-led sales call and sets up the sales system, which does not exist yet.Source

Add a domain box on the homepage and /agent-visibility: Manicule runs 10 generated buyer questions several times each and shows recommendation share with its noise band, the competitors named instead, and three answers to read. Results unlock fully with a free trial signup.

Seeing your own product missing from answers is the strongest reason to start tracking, and the noise band shows Manicule's difference in the first minute.

• Answer costs from casual visitors

  Cache by domain and limit runs per visitor.

• Your websiteResearch section "What we learned from a million AI answers": "We asked 173,554 buyer questions across 200 software categories on Claude, ChatGPT and Gemini." One answer misjudged a brand's rate by a median of 14 points, thirteen answers cut that to 7, and half of brand-question pairs needed 18 or more runs to settle. Four question headings ("How many runs before a score means anything?", "Does it matter who's asking?", "What happens when assistants disagree?", "Should questions that name you count?") have no answer text under them. The full study is only a PDF linked from the footer.Source
• Your websiteManifesto: "AEO tools ask a model a made-up question once, print a number to one decimal place, and call it your visibility."; "We ask your real buyer questions from your data dozens of times each"; "You can open any number and read the answers under it."Source
• Your websiteAgent visibility page: headline "Your AI visibility tool is lying to you."; Starter $79/mo (75 tracked prompts, 3 countries, "14-day free trial", "No card required"), Pro $199/mo (200 prompts, "1 more assistant soon") and Max $449/mo (600 prompts, "3 more assistants soon"); every plan lists ChatGPT, Gemini and AI Overviews and none lists Claude; only Starter shows a free trial; yearly billing "saves up to $600".Source

Make a standard audit the first meeting: for a target devtool, run the docs check, send a coding agent through the quickstart, and run 20 buyer questions with repeated runs. Deliver a two-page report: where agents get stuck, which rivals get named instead, and three fixes. Close into the docs rebuild or software.

A buyer without a category budget responds to their own failure shown live, and the audit doubles as a scoping document for the first deal.

• Audits take too long to scale

  Automate the runs and keep the writing to two pages.

• #6 Sell the docs rebuild as a fixed-scope, priced offer, and pick one number: 14 days or 23

• HiringEnterprise Sales (San Francisco, $150K-$250K): owns enterprise deals from discovery through close and expansion, selling to founders, CTOs, engineering leaders and developer-marketing teams who may not yet have a budget or vocabulary for the category; in the first 30 days the hire joins every founder-led sales call and sets up the sales system, which does not exist yet.Source
• Your websiteAgent visibility features: a noise band so a drop alert fires only on a fall that chance alone would rarely produce; cited pages that mention competitors but not you, with draft outreach emails; claim checking against your facts; betas for agent experience (whether Claude Code and Codex can install and use your product), competitor watch and docs check; consulting sessions on every plan; "Switching from Profound? We import your prompts and history automatically, in under 30 seconds."; the page names Profound, Peec AI, Scrunch, Otterly.AI, AthenaHQ, Evertune, Bluefish, ZipTie, Rankscale, Ahrefs, Semrush and SE Visible; logos for Mastra, AgentMail, TinyFish, Skyvern and Greptile.Source

Inside the Margin dashboard, show agent traffic next to the share of AI answers that cite the docs, with one locked panel: "which rivals agents recommend instead", opened by starting a Manicule trial. Add a one-command migration guide for Mintlify users and a monthly email with each site's agent traffic.

Free hosting brings in exactly the right companies, and their own agent-traffic numbers make the upgrade case for Manicule.

• Open-source users resent upsells

  Keep self-hosting fully free and the upsell to one panel.

• Your websiteMargin: "The open-source docs framework"; "Bring your Mintlify docs as they are."; every page as Markdown, with llms.txt and an MCP server on by default, plus /llms-full.txt and /AGENTS.md; hosted free on Manicule with agent analytics, or self-hosted under Apache 2.0; Margin has no web editor and "For writing, there's Manicule". The homepage says "Docs on Margin report to Manicule."Source
• Your websiteBlog (one post, November 7, 2026): a $1.3M seed round co-led by Y Combinator and Broom Ventures; Manicule has also worked with Greptile, Rootly and PromptLayer; the money goes to the tracking software, internal agents and customer-facing technical staff; developer tools first, other software categories later. It cites Vercel (ChatGPT referred 10% of new signups) and Mintlify (agents and crawlers were 48% of docs traffic). No RSS feed or newsletter signup.Source

Extend the import to Peec AI and other trackers' exports where available, and produce a free report: each reported rise or fall, marked real or within noise. Offer it on the comparison pages and in outreach to teams that mention other trackers publicly.

It turns Manicule's main claim into the buyer's own data, and the import already removes the switching cost.

• Rivals' exports lack run-level data

  Re-run the imported prompts in Manicule to build the band.

• #4 Publish "Manicule vs" pages for the twelve tools you already name, the same play that won "mem0 vs supermemory"

• Your websiteAgent visibility features: a noise band so a drop alert fires only on a fall that chance alone would rarely produce; cited pages that mention competitors but not you, with draft outreach emails; claim checking against your facts; betas for agent experience (whether Claude Code and Codex can install and use your product), competitor watch and docs check; consulting sessions on every plan; "Switching from Profound? We import your prompts and history automatically, in under 30 seconds."; the page names Profound, Peec AI, Scrunch, Otterly.AI, AthenaHQ, Evertune, Bluefish, ZipTie, Rankscale, Ahrefs, Semrush and SE Visible; logos for Mastra, AgentMail, TinyFish, Skyvern and Greptile.Source
• Your websiteResearch section "What we learned from a million AI answers": "We asked 173,554 buyer questions across 200 software categories on Claude, ChatGPT and Gemini." One answer misjudged a brand's rate by a median of 14 points, thirteen answers cut that to 7, and half of brand-question pairs needed 18 or more runs to settle. Four question headings ("How many runs before a score means anything?", "Does it matter who's asking?", "What happens when assistants disagree?", "Should questions that name you count?") have no answer text under them. The full study is only a PDF linked from the footer.Source
• Competitor researchPeec AI's homepage: "AI search analytics for marketing teams"; "Trusted by 4000+ brands and agencies"; separate pricing for brands and agencies, and a free trial.See the research Source

Write three: one software customer (visibility change against the noise band and what they did about it), one docs or agent-experience result for an API or infrastructure company, and one content or Reddit campaign outside AI memory. Use the same format as the Supermemory pages: problem, what was done, timeline, numbers, a quote.

Proof that matches the buyer's niche shortens enterprise sales, and case studies with numbers are what assistants quote when asked who does AgentRel well.

• Customers will not share numbers

  Use percentages and relative changes.

• Your websiteSitemap (16 URLs, November 9, 2026): home, /services, /agentrel, /agent-visibility, /manifesto, two case studies (both Supermemory), /blog with one post, /careers and six legal pages; only the blog post carries a lastmod date; there is no /pricing page (it returns 404), no comparison pages and no research page, and the research PDF is not listed.Source
• Your websiteHomepage (checked November 9, 2026): "Make your product the one AI agents discover. recommend. use."; a services offer ("We write, ship, and measure.") and a software card that says "Track how ChatGPT, Claude and Gemini describe your brand"; the "Used" pillar promises "We rebuild your docs in 14 days, check every snippet against your live API, then send coding agents through real tasks."; logos for Greptile, Supermemory, Rootly, PromptLayer, Skyvern, Surge, xpander.ai and Context.dev; no prices on the page.Source
• Y Combinator profileSpring 2026 batch, founded 2025, team of 8, San Francisco; customers listed include AgentMail, Greptile, Reducto and Mastra.Source

Design the first session around one finding: after prompts generate, show the top cited page that names rivals but not the customer, with the draft outreach email, and book the first consulting session from that screen. Offer the trial on Pro as well, and email day-3 and day-10 summaries of what changed.

Trials convert when users see a specific, actionable gap early, and a booked call with the team is the best moment to sell the next plan.

• Pro trials cost more to run

  Cap trial prompts at the Starter level and unlock on payment.

• Your websiteAgent visibility page: headline "Your AI visibility tool is lying to you."; Starter $79/mo (75 tracked prompts, 3 countries, "14-day free trial", "No card required"), Pro $199/mo (200 prompts, "1 more assistant soon") and Max $449/mo (600 prompts, "3 more assistants soon"); every plan lists ChatGPT, Gemini and AI Overviews and none lists Claude; only Starter shows a free trial; yearly billing "saves up to $600".Source
• Your websiteAgent visibility features: a noise band so a drop alert fires only on a fall that chance alone would rarely produce; cited pages that mention competitors but not you, with draft outreach emails; claim checking against your facts; betas for agent experience (whether Claude Code and Codex can install and use your product), competitor watch and docs check; consulting sessions on every plan; "Switching from Profound? We import your prompts and history automatically, in under 30 seconds."; the page names Profound, Peec AI, Scrunch, Otterly.AI, AthenaHQ, Evertune, Bluefish, ZipTie, Rankscale, Ahrefs, Semrush and SE Visible; logos for Mastra, AgentMail, TinyFish, Skyvern and Greptile.Source

Each week, publish one chart from Manicule's data on a devtools question ("which vector database do assistants recommend this week, and how stable is it?"), adapted per channel: a thread on X, a founder-voice post on LinkedIn, and a methods-first post in the relevant subreddit. Collect the series on the blog with an email signup.

Regular original data builds an audience of exactly the devtools founders and DevRel leads who buy, and gives the new marketers a repeatable format from week one.

• Named companies object to their numbers

  Publish method and invite corrections; lead with categories, not individual brands.

• #5 Turn the million-answer PDF into a /research hub, and answer the four empty questions on /agent-visibility

• HiringContent marketers in Singapore for Reddit, LinkedIn and X, each owning the channel for Manicule and its customers; the Reddit post says "Reddit is not another channel to dump content into." and measures referrals, signups, brand mentions and pipeline; the LinkedIn role measures profile growth, qualified engagement, inbound leads and pipeline.Source
• Your websiteBlog (one post, November 7, 2026): a $1.3M seed round co-led by Y Combinator and Broom Ventures; Manicule has also worked with Greptile, Rootly and PromptLayer; the money goes to the tracking software, internal agents and customer-facing technical staff; developer tools first, other software categories later. It cites Vercel (ChatGPT referred 10% of new signups) and Mintlify (agents and crawlers were 48% of docs traffic). No RSS feed or newsletter signup.Source
• Your websiteResearch section "What we learned from a million AI answers": "We asked 173,554 buyer questions across 200 software categories on Claude, ChatGPT and Gemini." One answer misjudged a brand's rate by a median of 14 points, thirteen answers cut that to 7, and half of brand-question pairs needed 18 or more runs to settle. Four question headings ("How many runs before a score means anything?", "Does it matter who's asking?", "What happens when assistants disagree?", "Should questions that name you count?") have no answer text under them. The full study is only a PDF linked from the footer.Source

Include a Pro or Max seat in every services engagement, set up with the client's prompts on day one, and report all services work through it. When a project ends, offer to keep the software at a reduced rate for the first year.

It turns one-off projects into recurring revenue and makes the services results visible in the client's own dashboard.

• Discounting the software

  Price the seat into the retainer rather than giving it away.

• Your websiteHomepage (checked November 9, 2026): "Make your product the one AI agents discover. recommend. use."; a services offer ("We write, ship, and measure.") and a software card that says "Track how ChatGPT, Claude and Gemini describe your brand"; the "Used" pillar promises "We rebuild your docs in 14 days, check every snippet against your live API, then send coding agents through real tasks."; logos for Greptile, Supermemory, Rootly, PromptLayer, Skyvern, Surge, xpander.ai and Context.dev; no prices on the page.Source
• Your websiteServices page: content strategy, technical content, docs, original research and distribution across search, AI discovery, Reddit and developer communities; it cites the Supermemory docs rebuild as completed in 23 days; no prices, packages or engagement terms; the next step is to book a call.Source
• HiringCareers page: "We went from $0 to $30K MRR in two months."; "We're small, we hire rarely, and people stay."; the Founding Engineer's first 30 days build a context system that pulls from Slack, Linear, GitHub and docs; the Technical Writer & Content Strategist post says "This is not a job for mass-producing AI content."Source

A monthly note per account: changes outside the noise band, the pages or posts shipped that month, new cited pages that name rivals, and one recommended action, with a link to book the included consulting session. Draft it automatically and have a strategist edit it.

Customers renew tools that tell them what to do, and the consulting sessions become a habit rather than an unused perk.

• Notes become generic

  Always include one account-specific action.

• Your websiteAgent visibility features: a noise band so a drop alert fires only on a fall that chance alone would rarely produce; cited pages that mention competitors but not you, with draft outreach emails; claim checking against your facts; betas for agent experience (whether Claude Code and Codex can install and use your product), competitor watch and docs check; consulting sessions on every plan; "Switching from Profound? We import your prompts and history automatically, in under 30 seconds."; the page names Profound, Peec AI, Scrunch, Otterly.AI, AthenaHQ, Evertune, Bluefish, ZipTie, Rankscale, Ahrefs, Semrush and SE Visible; logos for Mastra, AgentMail, TinyFish, Skyvern and Greptile.Source
• Your websiteAgent visibility page: headline "Your AI visibility tool is lying to you."; Starter $79/mo (75 tracked prompts, 3 countries, "14-day free trial", "No card required"), Pro $199/mo (200 prompts, "1 more assistant soon") and Max $449/mo (600 prompts, "3 more assistants soon"); every plan lists ChatGPT, Gemini and AI Overviews and none lists Claude; only Starter shows a free trial; yearly billing "saves up to $600".Source

After each visible win (a case-study number, a ranking, a docs launch), ask the customer's founder for two introductions to peers facing the same problem. Give both sides a free month of software or a free agent-readiness audit. Track referrals by source in the new sales system.

Referrals from respected founders close faster and need no category education, which is exactly what Manicule's buyers lack.

• Asking too early

  Only ask after a measured win.

• HiringCareers page: "We went from $0 to $30K MRR in two months."; "We're small, we hire rarely, and people stay."; the Founding Engineer's first 30 days build a context system that pulls from Slack, Linear, GitHub and docs; the Technical Writer & Content Strategist post says "This is not a job for mass-producing AI content."Source
• Y Combinator profileSpring 2026 batch, founded 2025, team of 8, San Francisco; customers listed include AgentMail, Greptile, Reducto and Mastra.Source
• HiringEnterprise Sales (San Francisco, $150K-$250K): owns enterprise deals from discovery through close and expansion, selling to founders, CTOs, engineering leaders and developer-marketing teams who may not yet have a budget or vocabulary for the category; in the first 30 days the hire joins every founder-led sales call and sets up the sales system, which does not exist yet.Source

Publish /leaderboard: for 30 devtools categories (vector databases, auth, email APIs, observability and so on), the monthly share of answers naming each tool, with noise bands and sample answers. Each tool's row links to a free claim page where the company can see its prompts and start tracking.

A trusted public ranking gets cited by press, developers and assistants themselves, and every company on it becomes a warm lead that wants to know why it ranks where it does.

• Ranked companies dispute results

  Publish method, noise bands and answers; correct openly.

• Answer costs at scale

  Grow categories as trials justify the cost.

• #5 Turn the million-answer PDF into a /research hub, and answer the four empty questions on /agent-visibility

• Your websiteResearch section "What we learned from a million AI answers": "We asked 173,554 buyer questions across 200 software categories on Claude, ChatGPT and Gemini." One answer misjudged a brand's rate by a median of 14 points, thirteen answers cut that to 7, and half of brand-question pairs needed 18 or more runs to settle. Four question headings ("How many runs before a score means anything?", "Does it matter who's asking?", "What happens when assistants disagree?", "Should questions that name you count?") have no answer text under them. The full study is only a PDF linked from the footer.Source
• Hacker NewsA Hacker News post linking the research PDF ("1,053,690 answers show that the current AEO landscape is...") drew 2 points and 1 comment; the comment's findings: ChatGPT changed its top recommendation 55% of the time when asked the same prompt twice, the three models agreed on the top brand in only 48% of markets, and adding buyer details changed the top pick 63-74% of the time.Source
• Your websiteBlog (one post, November 7, 2026): a $1.3M seed round co-led by Y Combinator and Broom Ventures; Manicule has also worked with Greptile, Rootly and PromptLayer; the money goes to the tracking software, internal agents and customer-facing technical staff; developer tools first, other software categories later. It cites Vercel (ChatGPT referred 10% of new signups) and Mintlify (agents and crawlers were 48% of docs traffic). No RSS feed or newsletter signup.Source

Package agent-experience testing as a CI check: on every docs or SDK release, coding agents run a set of real tasks (install, authenticate, first call) and report pass or fail with transcripts. Offer it free for open-source projects and as a paid add-on, and feed failures into the docs check.

It makes the "Used" test continuous and sticky, gives engineering teams (not only marketing) a reason to pay, and is hard for answer-only trackers to copy.

• Flaky agent runs

  Run several times and report with the same noise band approach.

• Your websiteAgent visibility features: a noise band so a drop alert fires only on a fall that chance alone would rarely produce; cited pages that mention competitors but not you, with draft outreach emails; claim checking against your facts; betas for agent experience (whether Claude Code and Codex can install and use your product), competitor watch and docs check; consulting sessions on every plan; "Switching from Profound? We import your prompts and history automatically, in under 30 seconds."; the page names Profound, Peec AI, Scrunch, Otterly.AI, AthenaHQ, Evertune, Bluefish, ZipTie, Rankscale, Ahrefs, Semrush and SE Visible; logos for Mastra, AgentMail, TinyFish, Skyvern and Greptile.Source
• Your website"What is AgentRel?" page: "AgentRel (Agent Relations) is the work of making your product easy for AI agents to discover, recommend, and use." Its "Found" tactics include publishing docs as plain HTML or markdown with an llms.txt; "Chosen" includes getting listed on comparison pages and publishing original benchmarks; "Used" includes offering an MCP server or CLI. It has a table comparing SEO, AEO/GEO, DevRel and AgentRel and a six-question FAQ, but no author or date.Source
• Competitor researchPeec AI's homepage: "AI search analytics for marketing teams"; "Trusted by 4000+ brands and agencies"; separate pricing for brands and agencies, and a free trial.See the research Source

Launch a partner tier: multi-client workspaces, white-label monthly reports, free Margin hosting for partners' clients, an AgentRel training and certification, and a revenue share or partner price. Manicule's own team stays focused on flagship accounts and the method.

Partners multiply reach without hiring, and they carry Manicule's method and vocabulary into many more devtools companies.

• Partners deliver poor work under the AgentRel name

  Certification and review of the first projects.

• Channel conflict with Manicule's own services

  Split by deal size or niche.

• #10 Make Margin the free front door: show every hosted docs site its agent traffic, then offer visibility tracking

• Competitor researchProfound's homepage: "The AI marketing platform to win in" ChatGPT, Perplexity, Claude, Gemini, Microsoft Copilot, DeepSeek and Google AI Overviews; serves marketing teams, agencies and partners; says a third of the Fortune 100 uses it; demo-led, with no prices on the homepage.See the research Source
• Competitor researchPeec AI's homepage: "AI search analytics for marketing teams"; "Trusted by 4000+ brands and agencies"; separate pricing for brands and agencies, and a free trial.See the research Source
• Y Combinator profileSpring 2026 batch, founded 2025, team of 8, San Francisco; customers listed include AgentMail, Greptile, Reducto and Mastra.Source

From the research data, pick the category where assistants' recommendations are least stable and buyers most technical (for example, data infrastructure or security software), run the leaderboard and a snapshot campaign for it, and sign three design partners on software plus a scoped services package.

It tests expansion with data Manicule already holds, and a second category with paying customers makes the company a platform rather than a devtools niche.

• Distraction from the devtools core

  Cap the pilot at one category and two quarters.

• #18 A public AgentRel leaderboard: monthly recommendation share for each devtools category

• Your websiteBlog (one post, November 7, 2026): a $1.3M seed round co-led by Y Combinator and Broom Ventures; Manicule has also worked with Greptile, Rootly and PromptLayer; the money goes to the tracking software, internal agents and customer-facing technical staff; developer tools first, other software categories later. It cites Vercel (ChatGPT referred 10% of new signups) and Mintlify (agents and crawlers were 48% of docs traffic). No RSS feed or newsletter signup.Source
• Your websiteResearch section "What we learned from a million AI answers": "We asked 173,554 buyer questions across 200 software categories on Claude, ChatGPT and Gemini." One answer misjudged a brand's rate by a median of 14 points, thirteen answers cut that to 7, and half of brand-question pairs needed 18 or more runs to settle. Four question headings ("How many runs before a score means anything?", "Does it matter who's asking?", "What happens when assistants disagree?", "Should questions that name you count?") have no answer text under them. The full study is only a PDF linked from the footer.Source
• Competitor researchPeec AI's homepage: "AI search analytics for marketing teams"; "Trusted by 4000+ brands and agencies"; separate pricing for brands and agencies, and a free trial.See the research Source

A fortnightly live session (X, LinkedIn or YouTube Live): founders opt in, a coding agent attempts their quickstart, and the Manicule team narrates what the agent read, where it got stuck and what to fix. Cut clips for each channel and offer participants a free fix plan.

Opt-in teardowns create content, social proof and qualified leads at once, and every participant becomes a candidate for the docs rebuild.

• A participant feels embarrassed

  Opt-in only, with a preview and a constructive tone.

• Your websiteHomepage (checked November 9, 2026): "Make your product the one AI agents discover. recommend. use."; a services offer ("We write, ship, and measure.") and a software card that says "Track how ChatGPT, Claude and Gemini describe your brand"; the "Used" pillar promises "We rebuild your docs in 14 days, check every snippet against your live API, then send coding agents through real tasks."; logos for Greptile, Supermemory, Rootly, PromptLayer, Skyvern, Surge, xpander.ai and Context.dev; no prices on the page.Source
• HiringContent marketers in Singapore for Reddit, LinkedIn and X, each owning the channel for Manicule and its customers; the Reddit post says "Reddit is not another channel to dump content into." and measures referrals, signups, brand mentions and pipeline; the LinkedIn role measures profile growth, qualified engagement, inbound leads and pipeline.Source

While the new designer rebuilds pages, inline critical CSS, defer non-critical scripts and render the marketing pages statically, so the largest text block paints sooner on slow mobile connections. Keep it to one sprint; this is polish, not a problem.

A site selling technical quality to engineers should feel instant, and the fix is cheapest during a redesign.

• Breaking interactive elements

  Test the trial and booking buttons after each change.

• Google PageSpeedMobile score 94 (desktop 92); real visitors pass Core Web Vitals (LCP 2.1 s, INP 95 ms, CLS 0 at the 75th percentile); the lab run flags about 0.65 s of render-blocking requests and 154 KB of unused JavaScript, out of 563 KB of JavaScript on a 761 KB page.Source
• HiringNine open roles on Y Combinator's job board: Founding Engineer, Engineering Intern, Enterprise Sales, Designer / Design Engineer (owns product and web design, including the website, landing pages and product launches), Technical Writer & Content Strategist, Freelance Technical Content Writer, and Reddit, LinkedIn and X Content Marketers.Source