How to get cited in AI answers

To get cited in AI answers, write clear pages that answer the questions buyers ask ChatGPT, Perplexity, Claude, Gemini, Grok and Google AI Overviews. Put the answer in the first lines, say plainly who you are and who you serve, and show up on the public pages those tools already pull from. Then ask the same buyer prompts every week and log who gets named.

Nobody can promise a cite. The answers change from week to week. What you can control is whether a page of yours is easy to find, easy to quote and easy to check.

The short version

• AI answers pull from the public web: search indexes, docs, Reddit, listicles and other pages that already rank or get linked.
• For Google AI Overviews and AI Mode, Google says the usual SEO rules apply. Your page must be indexed and eligible for a snippet.
• ChatGPT searches the web and may cite sources. Perplexity searches in real time and adds numbered citations.
• Write one page per buyer question. Answer in the first two sentences. State what you do and for whom.
• Add an llms.txt file and a clean .md copy of key pages so agents can read you without guessing.
• If you add structured data, make it match the text on the page. Google says AI features need no special markup, so treat schema as normal SEO, not an AI lever.
• Earn honest mentions on Reddit, directories, YouTube and X. Models often quote third-party pages, not only your site.
• Test 20 to 30 real buyer prompts each week. Log who gets cited. Fix the gaps.

Why AI answers matter

When an AI answer names a product, many buyers stop there. Pew Research found people clicked a link inside a Google AI summary in just 1% of visits that showed one. Ahrefs found the top result’s click rate 34.5% lower when an AI Overview appeared, compared with similar informational keywords without one.

Being the page the answer quotes is the new top spot. It is not a ranking you can buy. It is a habit of writing pages that answer one question clearly, early, with facts a model can check.

If you want the full GEO service view, see our GEO page. This post is the how-to you can run yourself. If the terms are new, AEO vs SEO vs GEO explains how they overlap.

How AI answers choose sources

Each tool works a bit differently. The shared pattern is the same: retrieve public pages, then write an answer that points at some of them.

Google AI Overviews and AI Mode

Google’s guide to AI features and your website says AI Overviews and AI Mode need nothing extra beyond normal search. A page must be indexed and eligible to show with a snippet, and the site must be included in Search generative AI features in Search Console, which is the default. Google also says you do not need special AI text files or markup for those features. The same crawling, indexing and people-first content rules still apply. The full checklist is in how to show up in Google AI Overviews.

ChatGPT

OpenAI’s help centre says ChatGPT can search the web for current information and may include citations. It partners with search providers (Microsoft is listed) and rewrites your question into targeted queries. Placement is not guaranteed. To be eligible, allow OAI-SearchBot to crawl your site. If you want ChatGPT to name your product when buyers ask for tools, read how to get your SaaS recommended by ChatGPT.

Perplexity

Perplexity’s help centre says it searches the web in real time, summarises what it finds, and adds numbered citations so you can open the sources. Its crawler docs recommend allowing PerplexityBot in robots.txt to appear in its results. The steps, including firewall rules, are in how to get cited by Perplexity.

Claude, Gemini and Grok

These tools also answer from a mix of training data and live retrieval, depending on the product and mode. Treat them the same way for writing: clear answers near the top, plain facts about who you are, and pages that show up in normal search. Then test each tool with the same buyer prompts so you see who they name today.

Be on the pages models already pull from

If no page of yours answers the question, the answer quotes someone else. Models lean on:

• Pages that already rank in Google or Bing for the question
• Docs and help centres that state facts clearly
• Comparison posts and “best X for Y” listicles
• Reddit threads, directories, YouTube transcripts and public posts on X

So the first job is coverage. List the questions buyers ask before they hire you or buy. Write one page per question. Link those pages from your homepage and from each other. Our SEO for startups with no budget guide covers how to pick those terms without paid tools.

Write pages a model can quote

Answer first

Put the answer in the first two or three sentences. Most pages warm up before they answer. AI answers tend to quote the page that says it first.

Say who you are

If your site never states plainly what you do and for whom, the answer cannot name you with confidence. One short line on every key page is enough. Example shape: “Ignura is a growth team for software founders.”

One question per page

A page that tries to cover five questions answers none of them well. Comparison and “best X for Y” pages work when they pick a real buyer choice and state the trade-offs with sources.

Facts with links

Link every number to a public source. Models and readers both check. Invented stats get you ignored or corrected.

Titles people can scan

Write a title and meta description that match the page. Our free SERP preview tool shows how they may look in Google on desktop, with pixel widths. Google may still rewrite them, but a clear title helps both search and humans.

llms.txt and markdown copies

llms.txt, proposed by Jeremy Howard in September 2024, is a markdown file at the root of your site. It gives agents a short summary of who you are and links to your key pages. The same proposal suggests a clean markdown copy of each important page at the same URL with .md added.

Google says llms.txt isn’t needed for Google Search, and that adding one “will neither harm nor help your site’s visibility or rankings in Google Search” (Google). No major AI provider has confirmed that its crawlers read other sites’ llms.txt. Some tools and agents can fetch it when a user or developer asks them to. It is about an hour of work. We publish llms.txt and .md twins for our pages. Treat it as a cheap proposal, not as a ranking factor. We cover who reads it in What is llms.txt.

Structured data that matches the page

Google’s AI features guide says structured data should match the text on the page, and that there is no special schema required for AI Overviews or AI Mode. Its AI optimization guide says structured data isn’t required for generative AI search, but is still worth using for rich results.

So add schema for Google Search, not for AI answers. Common types for a startup site:

• Organization and WebSite on the homepage: name, URL, logo, sameAs links
• SoftwareApplication on product pages that show real reviews
• BlogPosting on posts

Since August 2023, Google only shows FAQ rich results for well-known, authoritative government and health sites (Google), and its update log says they stopped appearing from 7 May 2026. You don’t need to remove FAQ markup, but don’t expect the dropdowns. Schema markup for SaaS sites covers each type and what Google requires. Skip markup you cannot back with on-page copy.

Third-party mentions

Answers often cite pages you do not own. Useful places:

• Honest Reddit replies from a real account, useful without a pitch
• Directories and listicles where your category already lives
• YouTube explainers and walkthroughs with a clear transcript
• Public posts on X that state a fact someone might ask about

Do not astroturf. Say who you are when it matters. A thin spam mention helps nobody and can hurt trust.

If content is your weak layer, our content page covers how we write for one audience with a tracked next step.

How to test every week

GEO without a log is guesswork. Run a simple loop.

1. Write 20 to 30 prompts buyers actually ask, in their words.
2. Ask each one in ChatGPT, Perplexity, Claude, Gemini, Grok and Google (AI Overview when it appears).
3. Log the date, the prompt, who got named, which URLs were cited, and whether you appeared.
4. Note gaps. Missing page, weak first lines, no third-party mention, or a competitor page that answers better.
5. Fix one gap, then re-ask those prompts the next week.

Count visits and sign-ups from AI tools when you can see them in analytics. Do not count mentions as the result. Mentions move. Visits and sign-ups are what you can run a business on.

For the service version of this work, see GEO for startups. For classic search next to it, see SEO for startups.

FAQ

What is GEO?

Generative engine optimisation: getting your company named and your pages quoted when buyers ask AI tools which product or agency to use. It uses many of the same pages as SEO, written so the answer comes first.

Can you guarantee ChatGPT will cite us?

No. Nobody honest can. The answers change. Write the pages, check the prompts every week, and count the visits and sign-ups that arrive.

Do I need llms.txt for Google AI Overviews?

No. Google says its AI features need no extra files or markup, and that an llms.txt file “will neither harm nor help your site’s visibility or rankings in Google Search”. No major AI provider has confirmed reading it either. Some tools and agents can fetch it on request, and it is cheap to add.

Is GEO different from SEO?

The same pages often serve both. SEO is about ranking in the list of links. GEO is about being the page an AI answer quotes, or the company it names. In practice that means answering early and saying plainly what you do. More on our SEO page.

How often should we test prompts?

Weekly is enough for most startups. Answers drift. A monthly check is better than none. A weekly log shows when a competitor takes the answer.

Nobody finds you in AI answers?

That is a search-layer problem we fix on our GEO page. In a free 15-minute Growth Check, tell us which buyer questions matter. We name what is behind it, live, and you leave with a free growth breakdown and strategy.

Book your free Growth Check