How to Get Your Website Cited by AI

TL;DR: Three things decide it. The engine has to be able to fetch your page, the page has to answer a question somebody actually asked, and it has to answer it in a form that can be lifted out. Most sites fail the first check or the third, and almost everybody spends their time on the second.

What "cited" means

When an AI engine searches before answering, it reads a handful of pages and builds the answer from them, often with links. Being cited means being one of those pages.

That's different from being named. You can be named in an answer built from somebody else's article, which is a separate route. This post is about making your own page the one that gets cited.

Be fetchable

The engines use named crawlers, like GPTBot, ClaudeBot and PerplexityBot, and a robots.txt written for Googlebot doesn't cover them. GPTBot is OpenAI's training crawler. ChatGPT search uses OAI-SearchBot, so check that one by name too. Google-Extended is a robots.txt token, and it doesn't control AI Overviews.

Blocking is rare. In the crawler study, 2.9% of 818 cited domains block GPTBot (details in LLM SEO). It takes a couple of minutes to check, and the AI crawler checker tests each crawler for you.

Then check rendering. If your content only exists after JavaScript runs, a crawler that doesn't run it sees an empty page. Server-render anything you want quoted. The title and meta description are part of what a crawler reads on the page, and the meta description checker shows you what yours say.

Answer a question somebody asked

Pick a question, in the words a person would use, instead of a keyword. Good sources are your Search Console queries, your sales calls and the first email a new customer sends. If three customers have asked the same thing in their own words, that's a page.

Engines also break one question into several searches before they answer. How ChatGPT finds your page covers how that works and what it means for the pages you write.

What a citable page looks like

Here are two versions of the same page.

The version that doesn't get cited

Enterprise Invoicing Solutions

In today's fast-paced business environment, managing invoices efficiently has become more critical than ever. Companies across industries are discovering the transformative power of modern invoicing platforms. A comprehensive solution leverages cutting-edge technology to streamline your accounts receivable workflow...

There's no claim a model can lift, no number, no stated limitation, and the first sentence that matters is four paragraphs away. A model summarizing this page writes a sentence about invoicing in general and names nobody.

The version that does

How do I invoice for retention on a construction job?

Retention is typically 5% of each invoice, held until practical completion and released in two stages. Most general invoicing tools can't track it, since they treat an invoice as either paid or unpaid. You need either a construction-specific tool or a spreadsheet alongside your invoicing.

The three common approaches are compared below, with what each one costs and where it breaks down.

The answer is in the first two sentences. It has specifics (5%, two stages) and it says what the common tools can't do. A model pulling from this page gets a usable claim.

The extraction checklist

Run any page you want cited through this.

• The answer appears in the first two sentences
• At least 3 specific numbers, dates or named things
• Each H2 states a conclusion, like "Most teams pay $40 to $90 a month", instead of a label like "Pricing"
• Sentences short enough to lift whole
• At least one stated limitation
• Renders without JavaScript

You can score a page against most of these with the AEO checker.

Perplexity

Perplexity nearly always searches, and it shows its sources next to every answer. That makes it the easiest engine to check first. Ask it your question, read the sources it lists, and see whether your page is one of them. If it isn't, look at which pages are and compare them against the checklist above.

When your own page isn't the way in

How much your own content matters depends on the market. Peak Answer measured 14 of them. In 10, at least 80% of the cited sources were third-party articles rather than a vendor's own site. A few examples:

Measured 10/02/2026. The full set is on the research page.

In markets at the top of that table, getting cited usually means getting mentioned in the articles the engines already quote. That's outreach work, and Why AI Recommends Your Competitor covers how to find those pages and what to send.

The order to do it in

1. Check crawler access.
2. Check that the engines search on your questions at all.
3. Write or rewrite the pages that answer them, using the checklist.
4. Re-run the questions and keep what moved.

What doesn't work

Keyword density. Schema markup on its own, which helps a page get parsed correctly but won't get it cited. Publishing a lot of pages without a target question for each.

Publish the number nobody has

If you have usage data, or you can survey your customers, publish the result. Put it in a table, state the method you used, and put it under a heading written as a question. Original data is the type of content that gets cited as the source.

Measure the answer

If you only report positions and impressions, add the questions your buyers ask and whether you're named in the answer. You can track that with the AEO tracker. Ranking and being named in the answer can come apart, and the answer is the part the buyer sees.

Related study: Who Does AI Cite as Questions Get More Specific?