How to get cited by ChatGPT and Perplexity: what the evidence actually supports

Search for how to get cited by ChatGPT and you will find hundreds of confident guides, most of them quoting statistics that trace back to nothing. The honest position is that the AI vendors do not publish how they choose sources, so every tactic sits somewhere on a scale from documented to invented. This guide puts each tactic on that scale, so you can spend your effort in the right order.

Tier 1: be reachable (documented by the vendors)

Nothing else matters if the crawler cannot read your page. These are the only steps the vendors themselves document.

• Allow the search crawlers. OpenAI says sites that opt out of OAI-SearchBot will not appear in ChatGPT search answers, and that a robots.txt change can take about 24 hours to take effect. Perplexity recommends allowing PerplexityBot so your site can appear in its results. Anthropic says blocking Claude-SearchBot may reduce your visibility in search results. Our study of 1,069 top sites found 12.3% block at least one of these.
• Be indexed and eligible for a snippet. For Google's AI Overviews and AI Mode, Google says a page must be indexed and eligible to appear in Search with a snippet, and that there are no additional technical requirements, no special markup and no new files to add.
• Put the content in the HTML. A Vercel and MERJ log study found that none of the major AI crawlers it tracked, including GPTBot, ClaudeBot and PerplexityBot, rendered JavaScript. The study dates from December 2024, so retest, but a page that is blank without JavaScript is a risk. See why AI crawlers can't see your site.

Tier 2: what controlled research measured

The only peer-reviewed work on this question is GEO: Generative Engine Optimization by researchers at Princeton, Georgia Tech and the Allen Institute for AI, published at KDD 2024. They built a benchmark of 10,000 queries, rewrote web pages using nine different tactics, and measured how much of a generated answer drew on the rewritten page. The score is position-adjusted word count: how many words of the answer are attributed to your page, weighted by how early they appear. The baseline with no change scored 19.3.

Three results stand out. First, the tactics that worked best add substance: quotes, numbers and citations. Second, keyword stuffing scored below the no-change baseline, so the oldest SEO habit actively hurt. Third, the effect depended on where you started: citing sources raised visibility by 115.1% for sites ranked fifth in the underlying search results and lowered it by 30.3% for sites ranked first. Substance helps most when you are not already the default answer.

Read this with care. The paper tested an academic generative-engine pipeline plus a smaller check on Perplexity, not ChatGPT, and the abstract reports gains of up to 40% overall. Effects differed by topic. We quote the paper's table as we read it; check the figures against the paper before building a business case on them.

Tier 3: plausible, but nobody has measured it

These follow from how retrieval systems work, and you will see them in many guides. None has a published, controlled measurement behind it, so treat them as reasonable bets rather than rules.

• Answer the question in the first two sentences under a heading that matches how people ask it. A retrieval system that lifts a passage needs the passage to stand alone. See how to write a page AI answers can quote.
• Describe yourself consistently. Use the same name, one-line description and category on your site, your profiles and your directory listings, so a model assembling an answer meets one coherent entity.
• Be mentioned on pages the engines already trust. Answers are built from the web, so what other sites say about you can matter as much as what you say.
• Keep dated content fresh. Engines that search the live web have something recent to retrieve. Show real publication and update dates.
• Use structured data to remove ambiguity. Organization, Article and FAQ markup state facts about the page plainly, though Google says no special schema is required to appear in its AI features.

Tier 4: folklore to ignore

• llms.txt as a ranking lever. No major AI vendor has said it reads the file for citations, and an Ahrefs study of 137,210 domains found 97% of llms.txt files received no requests in May 2026. It is cheap, not powerful. See llms.txt explained.
• Keyword density targets. The GEO paper's worst performer was keyword stuffing.
• Hidden text written for bots. It adds nothing a crawler can use and risks a penalty from search engines.
• Invented statistics. The research rewards real numbers with sources, and a fabricated figure is a liability the first time someone checks it.
• Any guide quoting a precise percentage with no named study. If you cannot find the study, the number is marketing.

A 30-minute checklist

1. Open your robots.txt and confirm OAI-SearchBot, PerplexityBot, Claude-SearchBot, Googlebot and the user-triggered agents are not blocked. Check your CDN's bot settings too.
2. View the source of your three most important pages and confirm the key text is in the HTML, not injected by JavaScript.
3. Pick one page that should be cited. Rewrite its opening so the answer comes first, and add at least one sourced number and one attributed quote where they fit.
4. Add Organization and Article JSON-LD, and make sure the dates are real.
5. Write down 10 buyer questions and record which engines cite you today. Our guide to checking AI citations shows how.
6. Repeat the questions in two weeks. Repeat each at least three times, because answers vary between runs.

How to tell whether it worked

Measure both citations and visits. Citations come from asking the engines your target questions on a fixed schedule. Visits come from analytics: see how to track ChatGPT, Perplexity and Gemini traffic in Google Analytics 4. Expect noise. A single run proves nothing, and AI referral traffic is small for most sites today.

Frequently asked questions