Katman gets SaaS products recommended by ChatGPT, Claude, Gemini and Perplexity. We find the questions your buyers and users ask assistants, then write and ship the pages they cite, or hand your AI coding tool the exact fixes with Katman Pro. B2B or B2C; no SEO background needed.
An assistant answers in one paragraph and names a few products. These are the questions we write for, one page each, with the answer in the first sentences.
One page per job your product does, for the people who do it: who it fits, what it costs, how it compares.
H1 in the buyer’s words. First paragraph: the answer, naming the product, team size and price. Then: the retainer workflow in five steps, a plain-text price table, the integrations that matter, an honest “not for you if” list, FAQ, JSON-LD. Served as HTML, readable without JavaScript.
The user describes a situation, not a category. The page written for that situation is the one an assistant picks.
H1 is the situation, not the product. First paragraph: what to use, free or paid, on which device. Then: before and after with a real note, three steps, the limits of the free plan, an FAQ in students’ words, JSON-LD.
Fair pages with dated facts from each competitor’s own site, so the assistant can trust and quote them.
A table of five tools with price, seat limits and the one thing each is best at, each fact dated and linked to the vendor’s own page. Your product listed where it honestly fits, not always first. Updated every 90 days.
One page per integration that exists: what it syncs, how to set it up, what it does not do.
H1 names both products. First paragraph: yes or no, and on which plan. Then: what syncs in which direction, a five-step setup, known limits, FAQ. One page per integration, never one page listing forty.
Plans, limits, security and data answers in plain HTML, the same numbers everywhere assistants look.
Every plan, limit and price in text, not in a JavaScript widget or an image. The same numbers in the page, the meta description, llms.txt and the JSON-LD, so no assistant reads two prices.
Developers ask their coding assistant. It needs docs it can read, quote and link.
A docs page that answers the question in its first lines, a runnable example, a short comparison with the two options developers weigh, and a markdown twin of the page for coding assistants.
How it works
01 · Done for you
We research the questions your buyers ask, write the pages that answer them, fix what keeps the crawlers out, and ask the assistants again every 30 days.
Growth: $1,490/month
Apply for Growth02 · One-time
How assistants read your product today, plus the first 30 pages, at one fixed price. On time or your money back, and credited in full to Growth within 30 days.
Launch Pack: $990 once
Start with the Launch Pack03 · Do it yourself
An MCP server for Cursor, Claude Code, VS Code and four more. It plans the pages, hands your agent tasks with acceptance criteria, and audits the result: 3 sites, 30 audits a month.
Katman Pro: $99/month
Get Katman ProRequests your homepage as 12 crawlers and a browser, and shows which ones get turned away.
Drafts an llms.txt for your site from its own pages, ready to edit and publish.
The questions your buyers and users type into assistants, grouped by situation, comparison and price.
One page per question, with the sentence you want the assistant to repeat. You approve it first.
robots.txt, CDN rules, server-rendered HTML, JSON-LD, sitemap and IndexNow, each with a pass or fail check.
The same questions asked again in ChatGPT, Claude, Gemini and Perplexity: who was named, which pages were cited.
Never ask a coding tool to “rank higher”. Give it a measurable task (“a unique title on every page”) and an acceptance criterion (“no two pages share a title”). Every task Katman writes, for us or for your agent, is written this way.
The method
Search ranks ten links. An assistant writes one paragraph and names a few products. To be one of them, a page has to be read, quoted and matched to the question.
| Gate | What fails | What we check |
|---|---|---|
| 1. Read | If a bot cannot reach the page, nothing else matters: robots.txt, a CDN that turns bots away, or text that only appears after JavaScript runs. | robots.txt · CDN and firewall access per bot · server HTML vs JavaScript · real 404s · sitemap |
| 2. Quote | An assistant lifts one sentence from your page. It has to be in the first lines: short, concrete and clear on its own, and the same everywhere. | first paragraph · meta description · llms.txt · the “not X” note · JSON-LD |
| 3. Fit | For “best X” questions assistants name the big brands. When someone describes their situation, the page written for that situation wins. | situation pages · an honest comparison guide · a pricing page · FAQ |
Both Google and the assistants are mostly won by the same work: the text is in the HTML, the bots are let in, and each page answers one question.
Why now · our own SaaS, 21 Sept to 17 Oct 2026
of monthly subscription revenue in the first 26 days came from people ChatGPT sent, about 1.9× what Google search brought.
Source: the site’s own analytics
the sign-up rate: 31% of visitors from ChatGPT signed up, against 9% of visitors from Google.
Source: the site’s own analytics
Launch day. A new AI-powered consumer web app goes live, from zero.
About 1.5 days later, the first visitor arrives from ChatGPT, before Google has sent anyone.
Week one: 333 real ChatGPT visitors, 63% of all human traffic. 59% land on two “situation pages”.
416 people from ChatGPT in ten days, bots removed; up to 90 a day.
Day 26: ~1,800 people from ChatGPT, 44% of sign-ups and 46% of subscription revenue.
One sample site, its first 26 days, our own product: an example of what the method did, not a forecast or a guarantee. Counted by the channel of each account’s first visit; ChatGPT visits without a referrer count as direct, so every ChatGPT share here is a floor.
What you get · on every plan
Delivered as files, a pull request or straight into your CMS. They live in your code, not in our tool.
You see the questions and approve the page plan before a single page is written.
Every technical fix comes with an acceptance criterion your developer can check.
The same questions asked again, with dates: who was named, which pages were cited, what changed.
Apply in two minutes. A person reads it and answers, not an autoresponder.
You pay only after you agree to a plan. Prices in US dollars; no VAT is added.
If the Launch Pack is not delivered within 14 days of receiving what we need, you get your money back.
We don’t post on Reddit for you or buy mentions. Nothing a buyer would be embarrassed to find.
Nobody can guarantee a place in an AI answer. We show you dated evidence of what changed instead.
At a glance
Katman gets SaaS products recommended by ChatGPT, Claude, Gemini and Perplexity. We find the questions your buyers and users ask assistants, then write and ship the pages they cite, or hand your AI coding tool the exact fixes with Katman Pro. B2B or B2C; no SEO background needed.
Katman gets SaaS products, B2B and B2C, read, quoted and recommended by ChatGPT, Claude, Gemini and Perplexity. With Growth we do the work for you: research the questions your buyers ask, write 30 pages a month that answer them, list the technical fixes and re-ask the same questions every 30 days. With Katman Pro, an MCP server, your AI coding tool does the work and Katman checks it. Katman is not a rank tracker, and it doesn't buy links or post on Reddit for you.
Yes. Growth is our done-for-you plan: $1,490 a month, first term 3 months, for 30 pages a month written and delivered as files, a pull request or straight into your CMS, plus the technical fixes and a dated assistant report. You can also start with the one-time Launch Pack ($990, 30 pages in 14 days). Apply on the done-for-you page; we reply within 2 working days.
SaaS teams. B2B tools whose buyers ask an assistant which product to shortlist, and B2C apps whose users ask it what to sign up for. The done-for-you plans (Growth and the Launch Pack) are only for software and AI products. Katman Pro, the do-it-yourself MCP, works on any public site you build with an AI coding tool.
No. Katman turns the method from our book into tasks, each with an acceptance criterion, and your AI coding tool does the work. Before any code is written, the agent summarises the research and asks you to approve the page plan.
Any MCP client that can run a local server: Cursor, Claude Code, VS Code with Copilot, Windsurf, Codex CLI, Gemini CLI and Claude Desktop. If you build in Lovable, Bolt or v0, open the project’s code in one of these tools. The live-site audit also runs from the command line against any public URL.
Katman has two plans and one one-time offer. Growth, done for you: $1,490 a month, first term 3 months, for 30 pages a month written and delivered, technical fixes and a dated assistant report every 30 days. Katman Pro, do it yourself: $99 a month for 3 sites (fix prompts and page briefs with no cap, 30 full audits a month), every MCP tool, the course and the community; cancel any time. The Launch Pack: $990 once for a diagnostic and 30 pages in 14 days, credited in full to Growth within 30 days. The book and the 30-day workbook (the PDF course) are sold separately: $119 each or $199 together. The AI crawler checker, the llms.txt generator and the public guides are free. With Katman Pro, visibility runs use your own OpenRouter key and cost about $2.4 for the default panel (range $1.2–3.7), paid by you to OpenRouter.
No. Katman runs locally over stdio. It reads files in your project but never .env files, keys or certificates, and it writes only to a .katman/ folder unless you ask it to write a file. Your OpenRouter key is sent only to openrouter.ai. Katman checks your KATMAN_KEY with katman.pro at most once a day, sending only the key and the MCP version; once per session it loads Katman’s method (prompts, page briefs, checklists) once per session before the first tool your plan includes, sending the key and the MCP version; it stays in memory, never on disk; before each katman_audit_site it also sends the domain you are auditing, so the audit can be counted against your plan; and katman_sync sends aggregate numbers only when you call it. Your code and file contents never leave your machine.
No. Nobody can guarantee a place in an AI answer, and the same question can get a different answer in another session. Katman records what the assistants said, with dates, and asks the same questions again 14 days later so you can see what changed.
On our own sample site, an AI-powered consumer web app launched from zero on 21 September 2026, the first ChatGPT visitor arrived about 1.5 days after launch and 333 real ChatGPT visitors came in the first 7 days. In the first 26 days (to 17 October) ChatGPT sent about 1,800 people, 30% of human visitors, and became the biggest revenue channel: 44% of sign-ups and 46% of monthly subscription revenue, about 1.9 times what Google search brought. That is one site, and it is our own product, not a client.