RailwayHow to get your startup cited by ChatGPT, Claude and Perplexity
A practical GEO checklist for founders: llms.txt, markdown twins, schema.org, crawler policy, MCP — and the one thing that matters more than all of them.
"Generative engine optimisation" sounds like a new discipline. Most of it is old SEO plus three new files. Here is what actually changes whether an assistant names your product when someone asks "what's a good tool for X?".
1. Answer-first pages
Assistants quote pages that answer a question in the first paragraph. Rewrite your homepage hero and each feature page so the first 60 words say what it is, for whom, and the key benefit, in plain words, without a CTA. This is the same tagline discipline we enforce when we generate profiles: 6–12 words, no "Get started today".
Then add a real FAQ section (5–8 questions people actually type) and mark it up as FAQPage with our schema generator. Assistants lift Q&A pairs almost verbatim.
2. llms.txt — the agent's table of contents
/llms.txt is a plain Markdown file that tells an AI what your site is and which pages matter. It is cheap and increasingly checked first by agents and crawlers. Ours is at exitfounder.io/llms.txt; generate yours in a minute with the llms.txt generator.
Keep it short: one blockquote summary, one paragraph of detail, a "Docs" list with 3–8 links and a one-line note per link. Put the pricing page in it — "how much does X cost" is the most common follow-up prompt.
3. Markdown twins
HTML is expensive for models; Markdown is not. Serve a .md version of your key pages (/pricing.md, /docs/getting-started.md) and list them in llms.txt. Every ExitFounder profile has one (/startup/{slug}.md) with YAML frontmatter, facts and FAQ — which is exactly the shape assistants like to cite.
4. Let the right crawlers in
Blocking GPTBot stops OpenAI training, but blocking OAI-SearchBot, Claude-SearchBot or PerplexityBot removes you from their answers. Decide per bot with the robots.txt generator; the default there allows answer/search bots and blocks pure training crawlers.
5. Structured data that describes a product
SoftwareApplication + Organization JSON-LD with offers (price), applicationCategory and sameAs links gives assistants (and Google's AI overviews) unambiguous facts. Vague or missing pricing is the most common reason an assistant answers "pricing is not disclosed" about a product that clearly has a pricing page.
6. Off-page consensus
Assistants weigh agreement across sources. Being listed with the same one-liner and category on 5–10 reputable directories (see our DR-ranked list) does more than one long guest post. Consistency matters: same name, same tagline, same category everywhere.
7. Give agents an API, not just a page
The next step is being usable by agents, not just quotable. TinyLaunch lets an assistant launch a product over its API; ScrollLaunch and TrustMRR expose MCP servers; ExitFounder has a public API and an MCP server (docs). When an agent can call search_startups and get a structured answer with provenance, it will prefer that over scraping a competitor's HTML.
The one thing that matters more
None of this outranks a page that already answers the question well and is linked from sites the model trusts. GEO is a multiplier on good content, not a substitute. Do 1–5 in an afternoon, then spend the next month on 6.
Checklist
- First 60 words of every key page answer "what / for whom / benefit"
- FAQ section with
FAQPageschema -
/llms.txtwith pricing and docs links -
.mdtwins for pricing and docs - robots.txt allows answer bots, blocks training bots (your choice)
-
SoftwareApplicationJSON-LD with real prices - Listed on 5–10 directories with identical name/tagline/category
- Profile claimed on ExitFounder so assistants get verified, provenance-labeled facts