How to get cited in ChatGPT, Perplexity, Gemini

Getting cited by ChatGPT, Perplexity, or Gemini is not a creative exercise. It is not about guessing what AI likes or crafting viral-ready prompts. It is about making three technical and editorial decisions correctly, then checking whether they worked. The founders I work with who get this right do not treat citation as a mystery. They treat it as a crawl-index-retrieve problem they can instrument and fix.

The work splits into three parts. First, you allow the right crawlers and block the ones that cost you bandwidth without sending traffic back. Second, you write pages that answer one question clearly enough that an AI system can extract, attribute, and cite the claim without ambiguity. Third, you measure whether your content is showing up in answers to the queries that matter to your business, and you adjust when it is not.

Allow the search crawlers, block the training scrapers

AI companies now run separate bots for training and for live retrieval. OpenAI runs GPTBot for training and OAI-SearchBot for ChatGPT search. Anthropic runs ClaudeBot for training and Claude-SearchBot for live answers. The crawlers you want are the search ones. Those are the bots that retrieve your page when someone asks a question, extract the answer, and send you referral traffic when the user clicks through.

Your robots.txt should allow OAI-SearchBot, ChatGPT-User, Claude-SearchBot, Claude-User, PerplexityBot, and Perplexity-User. Block GPTBot, ClaudeBot, CCBot, Google-Extended, and Meta-ExternalAgent if you do not want to contribute training data without getting citations in return. The configuration looks like this:

User-agent: OAI-SearchBot
Allow: /

User-agent: ChatGPT-User
Allow: /

User-agent: Claude-SearchBot
Allow: /

User-agent: Claude-User
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: Perplexity-User
Allow: /

User-agent: GPTBot
Disallow: /

User-agent: ClaudeBot
Disallow: /

User-agent: CCBot
Disallow: /

User-agent: Google-Extended
Disallow: /

Blocking GPTBot does not affect whether ChatGPT can cite you in search answers. That separation is deliberate. You can refuse to feed the training corpus and still show up in live retrieval. The same logic applies to Claude. For Perplexity, there is no separate training bot. PerplexityBot handles both, so if you block it you disappear from Perplexity entirely.

If your site renders critical content with JavaScript and you have not set up server-side rendering or static generation for bot traffic, the AI crawlers will see empty divs and skip you. I wrote about how to handle JavaScript rendering for AI crawlers in detail. The short version is that OAI-SearchBot and PerplexityBot do not execute JavaScript reliably, so if the answer lives in a React component that mounts after page load, you need to render it server-side or pre-render it statically.

Write pages that make one claim per paragraph, with proof

AI engines cite pages that make extraction easy. That means a page structured around a single question, with the answer in the first paragraph, followed by supporting detail that backs the claim. In the citation audits I run for B2B SaaS companies, definition pages consistently outperform other formats across Perplexity, Claude, and Google AI Overviews. Comparison pages also do well in Perplexity and AI Overviews. Pricing pages and forum threads perform inconsistently.

The pattern that works is this. Open with a one-sentence definition or direct answer. Follow with three to five short paragraphs, each making one claim, each claim supported by a data point, a source link, or a specific example. Use descriptive H2 or H3 headings that mirror how someone would phrase the question in a search box. Keep paragraphs to three or four sentences. If you mix three claims in one paragraph, the AI system has to choose which claim gets the link, and it usually picks none.

Bullet lists and tables help when the content is genuinely structured. A comparison table with three columns and five rows is easier to extract than five paragraphs describing the same information in prose. But do not force it. A bulleted list that wraps ten sentence-long items is harder to parse than three short paragraphs.

Back your claims with inline links to primary sources. If you cite a percentage, link to the study. If you reference a product feature, link to the vendor documentation. Perplexity values content that creates a web of mutual verification, and that applies across engines. Pages that cite their own sources get treated as more credible than pages that make unsupported claims.

Update your strategic content quarterly. Perplexity prioritizes freshness more than ChatGPT does, and visibly dating your articles with a last-updated timestamp reinforces the signal. A page published in 2024 and never touched will lose citation share to a competitor page updated last month, even if your domain authority is higher.

Measure citation rate, not traffic alone

The citation happened before the click. When you see utm_source=chatgpt.com in your analytics, it means ChatGPT already cited you and someone clicked through. That traffic is valuable, but it is a lagging indicator. The leading indicator is whether you are named in the answer at all, whether or not the user clicks.

Run a fixed list of 15 to 20 buyer-intent queries across ChatGPT, Perplexity, Claude, and Gemini every week. Record which engine cited you, in which position, and whether it linked to your domain or just named your brand. Track five metrics. Citation rate measures whether the AI mentions your brand at all. URL citation rate measures whether it links to your site. Sentiment captures how the AI describes you. Prominence tracks your position in the source list. Share of voice compares how often you appear relative to competitors.

You can do this manually in 30 minutes a week. Open an incognito window, run each query in each engine, screenshot the results, log the data in a spreadsheet. If you want to automate it, tools like Profound, LLMsRefs, and Cloro track this. I run manual tracking for clients in my consultancy engagements because the judgment call about which queries matter and how to score ambiguous mentions is hard to automate without losing signal.

Set up a custom channel group in GA4 so AI traffic does not get buried in referral or direct. Go to Admin, then Data Display, then Channel Groups. Create a new channel called AI Traffic. Set the Source condition to match this regex:

chatgpt\.com|perplexity\.ai|claude\.ai|gemini\.google\.com|copilot\.microsoft\.com|openai\.com

Place it above Referral in the channel list. ChatGPT now appends utm_source=chatgpt.com to many citation links, and most citation clicks pass that parameter. Perplexity, Claude, and Gemini send referral traffic without UTM tags, so the custom channel group catches them by source domain.

True AI influence is likely two to three times what GA4 reports. Mobile app visits and zero-click interactions where someone reads the AI answer and never clicks through do not pass attribution, but they still moved the buyer. If your citation rate is climbing but your AI traffic in analytics stays flat, that gap is probably zero-click answers doing their job.

What getting this right looks like in practice

In the AI search audits I run for Series A and B companies, the pattern is consistent. The companies that get cited reliably have allowed the search bots in robots.txt, they publish definition and comparison pages with inline sources and quarterly updates, and they measure citation rate on a fixed query list every week. The ones that do not get cited usually have one of three problems. Either they blocked PerplexityBot or OAI-SearchBot without realizing it, or their content answers five questions in one 2,000-word post instead of writing five focused pages, or they are measuring traffic in GA4 and concluding AI does not work because they cannot see the zero-click influence.

If you want to move faster on this, I run a four-week AI search engagement that includes a bot-access audit, a 20-query citation baseline, page-structure templates, and a measurement dashboard that tracks all five citation metrics. The companies that finish that sprint usually double their citation rate in the next 90 days because they fixed the three decisions that were costing them visibility.

Getting cited is not about being smarter than the algorithm or writing content that sounds like an AI wrote it. It is about making your pages crawlable, extractable, and verifiable, then checking whether the engines that matter to your buyers are actually naming you when it counts.