How to get cited by ChatGPT: what actually decides which sites it links

Most advice on getting cited by ChatGPT is one of two things. Some of it is regular SEO with "AI" added to the headings. The rest is a list of tricks, like special files, schema types, or keyword patterns, that no AI company has said it uses.

This guide sticks to what the engines themselves document, plus two large independent studies. The mechanics turn out to be simpler than the advice around them. It covers ChatGPT first and notes where Perplexity, Claude, and Google's AI features work the same way.

How ChatGPT decides what to cite

ChatGPT can mention your brand in two different ways, and only one of them comes with a link.

From its training data. The model learned about many companies during training, and it can mention your brand from memory. That knowledge comes from crawls like GPTBot's, and it's months old by the time a model ships. There's no link attached.

From a live search. When ChatGPT searches the web, it runs one or more queries, reads the pages it gets back, and cites the ones it uses. OpenAI says ChatGPT search uses third-party search providers, including Microsoft Bing, along with its own crawler, OAI-SearchBot, and content from partners. Every link in a ChatGPT answer comes from this path.

A single question often turns into several searches. Ask "what's the best CRM for a 10-person agency" and ChatGPT might search for agency CRM comparisons, pricing pages, and reviews separately, then combine what it finds. Ahrefs calls this "query fan-out." It explains a result that surprises most SEO teams: in Ahrefs' study of 15,000 queries, only about 8% of ChatGPT's in-text citations ranked in Google's top 10 for the original prompt. Across the four AI assistants tested, the average was 12%. Perplexity was highest at 28.6%.

So ranking for the head keyword isn't the target. The target is being the best page for one of the narrower questions ChatGPT searches for while it builds an answer.

Step 1: Let the right crawlers in

Every AI company runs separate bots for training, search, and fetches made on a user's behalf. You can block one and allow another. Most sites that are never cited are blocking the wrong one, often through a CDN or firewall setting rather than their own robots.txt.

OpenAI is explicit: "Sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers." Blocking GPTBot only means your content isn't used for training, and it doesn't affect citations. Anthropic says the same about Claude-SearchBot: blocking it "may reduce your site's visibility" in Claude's search results.

To stay out of training while staying eligible for citations, use:

User-agent: OAI-SearchBot
Allow: /

User-agent: Claude-SearchBot
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: GPTBot
Disallow: /

User-agent: ClaudeBot
Disallow: /

Then check the layers above robots.txt. Cloudflare and other CDNs have AI-crawler blocking switches, and security plugins and firewall rules often challenge unfamiliar user agents. Search your server logs for OAI-SearchBot. If it never shows up, or only gets 403 responses, something upstream is blocking it.

Step 2: Get indexed in Bing

Because Bing is one of ChatGPT search's providers, a page Bing hasn't indexed has one fewer way into an answer. Many sites have spent years watching Google Search Console and never opened Bing Webmaster Tools.

1. Verify your site in Bing Webmaster Tools. You can import it from Google Search Console in a few clicks.
2. Submit your XML sitemap.
3. Turn on IndexNow so Bing hears about new and updated URLs within minutes rather than waiting for a recrawl. Many CMSs and CDNs support it with one setting.
4. Look at the pages Bing reports as "discovered but not crawled" or excluded. They can't be cited.

Step 3: Serve the answer in the initial HTML

Vercel's analysis of AI crawler traffic found that none of the major AI crawlers render JavaScript, including OAI-SearchBot, ChatGPT-User, GPTBot, ClaudeBot, and PerplexityBot. They download your JavaScript files but don't run them.

If your pricing table, docs, or FAQ loads after client-side rendering, the crawler sees an empty page. To check, fetch the page with JavaScript disabled, or run curl and search the response for the sentence you want quoted. If it isn't there, render that content on the server.

Two related fixes:

• Speed matters more than usual. User-triggered fetches like ChatGPT-User happen while someone waits for an answer. A slow page can time out and get skipped.
• Don't put key facts only in images or PDFs. A pricing screenshot can't be quoted. A pricing table in HTML can.

Step 4: Write pages that answer the sub-questions

ChatGPT quotes passages, not whole pages. A passage it can lift cleanly is short, direct, and complete on its own. That's what "answer-first" means:

• Match headings to real questions. Use "How much does X cost for a 10-person team?" rather than "Pricing philosophy."
• Answer in the first two to four sentences under each heading, then add detail. A reader who only sees those sentences should still get the answer.
• Be specific. Numbers, prices, dates, limits, and named alternatives are what AI answers repeat. "Affordable for small teams" gets skipped, while "$19/month for up to 5 seats" gets quoted.
• Say who it's not for. Buyers ask comparison questions ("X vs Y for agencies"). Pages that honestly cover trade-offs and competitors match those searches. Pure sales copy doesn't.
• Date your facts and keep them current. ChatGPT often searches for the current year. A page last updated two years ago loses to one that says "as of September 2026."

To choose which questions to write for, ask ChatGPT and Perplexity the questions your buyers ask and read the sources they cite. Every cited page you could do better than is a page to write. Every question where a competitor is cited and you aren't is a gap.

Step 5: Get mentioned where AI engines already look

Your own site is only part of it. Ahrefs looked at 75,000 brands and found that YouTube mentions had the strongest correlation with AI visibility (about 0.74), followed by branded web mentions (0.66–0.71) and branded anchor text. Traditional backlink counts correlated much more weakly. The study also found that ChatGPT relied less on established authority signals than Google's AI Mode did, which gives newer brands a better chance there.

In practice, that means getting your product discussed in the places fan-out searches pull from:

• Comparison and "best X for Y" articles on other sites, including your competitors' roundups
• Reddit threads and community forums where your buyers ask for recommendations
• YouTube reviews, tutorials, and walkthroughs
• Review sites like G2 and Capterra, and directories in your category
• Podcasts and industry newsletters that publish transcripts or show notes

None of this is new, but the reason to do it is. The goal isn't a link. It's having your brand name appear next to the problem in text that AI engines read.

Skip the shortcuts. Google says you don't need AI text files, special markup, or schema to appear in its AI features, and no AI company has said it uses llms.txt to pick citations. Structured data still helps regular search, but it doesn't get pages cited.

Step 6: Measure it

You can't improve what you only check by hand now and then. At a minimum, track three things.

Referral traffic. ChatGPT adds utm_source=chatgpt.com to the links it cites, so those visits appear under that source in Google Analytics. Perplexity, Claude, Gemini, and Copilot show up as referrers too. Look at which landing pages get these visits. They're the pages already being cited.

Crawler activity. In your server or CDN logs, count requests from OAI-SearchBot and Claude-SearchBot (indexing) and from ChatGPT-User, Claude-User, and Perplexity-User (live fetches during an answer). A rise in live fetches on a page usually shows up before a rise in referral traffic.

Prompt-level citations. Traffic only shows the clicks. To see where you're cited, or where a competitor is cited instead, write down 20 to 40 questions your buyers actually ask and re-run them across engines on a schedule. Answers vary from run to run, so compare trends over several runs, not single results. You can do this by hand in a spreadsheet. Dedicated tools like Ahrefs Brand Radar and EchoRanked do it automatically. Disclosure: EchoRanked is made by Geistwerk, the company behind DeskFerry. It tracks citations across ChatGPT, Claude, Gemini, Grok, and Perplexity, reports ranges instead of single numbers because AI answers are noisy, and flags pages that AI crawlers can't reach.

Whatever you use, make the check a recurring task. A weekly routine that re-runs the prompt set, pulls the ChatGPT referral numbers from analytics, and flags any page that lost a citation is easy to hand off. It's the kind of recurring cross-app job you can give to an AI agent instead of a person.

A 30-day plan

Then repeat weeks 2 to 4 each month.

Related reading: How to Create an AI Agent · AI Agents for Marketing · AI Agent Workflows · SEO Automation with AI Agents