AI search
How to get cited by ChatGPT and Perplexity
The short answer
To get cited by ChatGPT and Perplexity, be fetchable and be quotable. Allow their crawlers in robots.txt, serve the full text in raw HTML without JavaScript, put a 40–60 word direct answer under every heading, ship JSON-LD per page type, and publish a markdown twin at the same URL. Everything else is secondary to those five.
Cite-ability is a build property, not a writing style
Most advice about AI search is about tone. The parts that decide whether you get quoted are structural, and they are decided in the build:
- The crawler is allowed in.
- The text is in the HTML.
- The answer is the first thing under the heading.
- The structured data names the same entities as the visible copy.
- There is a cheap, plain-text version of the page to fetch.
Fix those five and the writing has something to work with. Skip them and no amount of authority in the prose will be retrieved.
1. Let the crawlers in
robots.txt must allow the AI crawlers you want: GPTBot and OAI-SearchBot for OpenAI, ClaudeBot and Claude-SearchBot for Anthropic, PerplexityBot, and Google-Extended if you want Google's generative surfaces to use your content.
Check the layer underneath too. Many shared hosts run a WAF that refuses short or bot-like user agents. If a plain Mozilla/5.0 request gets a 406, assume the crawlers are being refused as well — and that no amount of on-page work matters until it is fixed.
2. Put the text in the HTML
Answer engines fetch and parse; they mostly do not execute JavaScript the way a browser does. A tool page whose explanation and FAQ exist only after hydration is a page with an empty body as far as retrieval is concerned.
Practical rule: the page must be complete and useful with JavaScript disabled. The interactive widget is an enhancement layered on top, not the page itself.
3. Answer first, under every heading
Under each question-shaped heading, the first sentence should be the answer, then the detail. This is the single change with the largest effect, and it is free.
The block under the H1 gets 40–60 words. Longer and the quotable sentence is diluted; shorter and the model has to reconstruct your meaning, which it will do from someone else's page.
4. Structured data per page type
| page | JSON-LD |
|---|---|
| home | Organization + Person with stable @id values |
| guide | Article with headline, dates, author, publisher |
| tool | SoftwareApplication with isAccessibleForFree and a zero-price Offer |
| any Q&A | FAQPage carrying the exact visible question and answer text |
| every page | BreadcrumbList matching the visible breadcrumb |
Two failure modes worth checking for by hand: a mangled @context (which invalidates the whole block silently), and markup that says more than the page does. The second is a manual-action risk, not an optimisation.
5. Serve a markdown twin
Every article and tool page should also exist as plain markdown at a predictable path — <url>.md is the simplest convention. It costs one build step, it strips navigation and styling, and it gives a retrieval system an unambiguous version of your text.
What is folklore
- llms.txt as a ranking lever. Ship it for hygiene if you like. Google has said extra machine-readable files neither help nor harm.
- Keyword density. Answer engines retrieve passages; padding a paragraph with the phrase makes the passage worse.
- Publishing volume. Freshness helps, but a page untouched for a year is a better candidate for a refresh than a site is for another fifty posts.
Measuring it
Pick ten questions you want to own. Ask each engine the same question monthly and record who is cited. Keep the engines separate — Perplexity, ChatGPT and Google's AI surfaces cite different sources, and a single blended number hides the only signal you have.
Traffic is the wrong first metric here. Zero-click results are the norm, so brand recall and citation rate move before sessions do.
Try it on your own site
The AI search readiness check is in development — it will list what a crawler can and cannot read on a URL you enter, in the order above, as a checklist you can run yourself today. Other tools are in the tools section. If you want the changes made rather than listed, book a 20-minute call.
Questions
Does llms.txt improve rankings?
No. Google's position is that extra machine-readable files neither help nor harm. Publish one because it is cheap hygiene for other consumers, never because you expect a ranking lift.
Do AI crawlers run JavaScript?
They generally do not. If a page renders its body in the browser, assume the crawler sees an empty container. The text has to be in the HTML the server sends.
Does FAQ markup still produce rich results in Google?
No, Google retired FAQ rich results in May 2026. The markup is still worth shipping because answer engines parse it, but do not expect it to change your search appearance.
How do I measure whether it is working?
Track citations per engine, not as one number, because the overlap between engines is small. Ask each engine the questions you want to own and record whether your domain, a competitor, or nobody is cited.
Want this scoped for your own workflow?
One call, one workflow, a fixed price in writing before any work begins.