How to Get Cited by AI: A Practical Guide for Earning Citations

If your URL never appears in AI answer engines’ source panel, it means you’re failing at one of three gates: retrieval, extraction, or attribution. Each gate points to a different underlying issue and requires a different fix.

You ask ChatGPT a question your best page answers better than anything else on the internet. The answer comes back, and the source panel links to a competitor, an aggregator you have never heard of, and a four-year-old Reddit thread. Sometimes your brand name is even sitting there in the answer text, and the link still isn’t yours.

The reflex is to assume the fix is more content or better rankings, but neither reliably works anymore. Ahrefs measured the share of Google AI Overview citations coming from top-10 organic results at roughly 38% in early 2026, down from about three-quarters the year before.

Ranking well no longer protects you, and your competitors’ rankings no longer block you. Instead, a page has to pass three distinct gates before it can appear as a cited source. Those gates are retrieval, extraction, and attribution. Each can fail for different reasons, and most teams waste time fixing the wrong one.

Below, I explain how I run an audit to find which gate is failing, and what to do to fix each one.

Key Takeaways

• A page fails to earn a citation at one of three gates: retrieval, extraction, or attribution.
• Most teams overinvest in extraction because it is the only gate their tools can score.
• Citations flow to origins. If you summarise someone else’s data, the model links to them instead.
• Retrieval and extraction fixes can move within 30 days. Attribution work compounds over 90 or more.

What’s the Difference Between an AI Citation and a Brand Mention?

A citation is a linked, attributed source the model pulled a passage from. A mention is your brand name appearing in the answer text with no link back. Both are valuable, and they are earned differently.

Outcome	What it looks like	What causes it
Citation	Your URL appears in the source panel or as an inline link	One passage on your page was the best available source for one claim
Mention	Your brand is named in the answer body, no link to you	The wider web associates your brand with the category

The practical consequence is that the two outcomes require different strategies:

• Mentions are mostly driven by what the rest of the internet says about you, which makes them a brand and PR problem.
• Citations are driven by whether your specific URL is the strongest source for a specific passage, which makes them a content and technical problem.

While my 3Bs Framework for AI Visibility covers both, here I’ll focus specifically on earning citations.

How Do AI Search Engines Choose Which Pages to Cite?

After retrieval, a citation is selected at the passage level from the pages that were shortlisted. That means three sequential gates. Failing any one of them means failing to get cited, but for completely different underlying causes.

AI search runs on retrieval-augmented generation (RAG). The system fetches live documents, drafts an answer from what it read, and attributes individual claims back to the documents it used. This means that a page can be eliminated at three separate stages:

1. Retrieval: Your URL never enters the candidate set. The crawler cannot reach it, cannot render it, or the system does not consider it fresh or relevant enough to fetch.
2. Extraction: Your page gets fetched, but no passage inside it is cleanly quotable. The answer to the question is spread across four paragraphs, hedged into vagueness, or dependent on context from three sections earlier.
3. Attribution: Your passage is perfectly quotable, but you are not the origin of the claim. You cited someone else’s research, so the model links to them instead of you.

Gates 1 and 2 are the easiest to address and where most teams spend most of their time and resources. Gate 3 is the toughest but the most likely to move the needle.

How to Get Cited By AI: My 3-Step Process

To earn AI citations, work the gates in order. There is no point making a page more quotable if the crawler can’t reach it, and no point publishing original research on a page no model can parse. Diagnose first, then fix in sequence.

Step 1: Find Out Whether AI Systems Can Retrieve Your Page

💡 Retrieval failures are the quickest to fix and the easiest to miss. If your URL is not fetchable, renderable, and recent enough to be worth fetching, nothing else in this article applies to it.

Start with crawler access and then check rendering to find out whether your page is fetchable. You can assess both with Screaming Frog, but you need to customize the configurations since it doesn’t run these crawls by default.

To check crawler access, navigate to Configuration > HTTP Header > User-Agent, choose Custom, and enter the User-Agent string of the bot you want to test.

After running the crawl, analyze the Response Codes tab. If your CDN/WAF rules are blocking those bots, you will see 403 Forbidden, 429 Too Many Requests, or 503 Service Unavailable instead of 200 OK.

To look into client-side rendering issues, navigate to Configuration > Spider > Rendering and change the Rendering Mode from Text Only to JavaScript.

After running the crawl, analyze the JavaScript tab at the top. Use the Contains JavaScript Content filter to identify pages where core text exists only in the rendered DOM, or check the Word Count Difference filter to flag large variances between raw HTML and rendered HTML.

If there is a massive gap, it means that AI crawlers are seeing an empty page, since they rely on raw HTML.

A few more eligibility questions worth running through:

• Is the content behind a login, a form, or a cookie wall?
• Is it locked in a PDF rather than an HTML page?

However, being fetchable doesn’t guarantee your page will be chosen. To avoid wasting compute, crawlers look at your page’s cover to decide whether they should open and parse it.

So, other questions worth asking include:

• Is any page in your website’s content inventory a plausible candidate? Deep, specific pages get cited. Homepages and broad category hubs almost never do.
• When was the page last meaningfully updated? Ahrefs found that 76.4% of ChatGPT’s top-cited pages had been updated within the previous 30 days.

Step 2: Ensure That Passages on Your Page Are Worth Quoting

💡 Being “extractable” means creating content structured so that AI engines can isolate, interpret, and surface clear facts, concise definitions, and direct answers without needing complex reasoning or parsing through fluff.

This step is one of the easiest to implement (though time-consuming in some cases), but research has shown that it’s one of the most effective strategies for improving brand visibility in AI answer engines.

Some techniques that have been proven effective include:

• Answer first, in 40 to 60 words. Every section opens with a direct answer written to survive being cut out of the page.
• Headings that state the claim, not the topic. “Earned coverage only earns you a citation when it links back” beats “Off-site signals.”
• Sections that stand alone. The test is whether a section (ideally, even a paragraph) still makes sense with no antecedent.
• Specific numbers with named sources. Models reach for verifiable claims over qualitative ones.
• Tables for anything comparative. They chunk cleanly and get pulled intact.

Then there is the question of which page gets the link. Search Engine Land found that 82.5% of AI citations pointed to deeply nested pages rather than top-level ones. The implication is about scope. A statistic buried in section nine of a 5,000-word guide is a weaker candidate than the same statistic on a page built around it.

Regarding schema, a contentious topic in AEO, GEO, and AIO debates, it’s worth noting that it’s strongly correlated with citations, because pages that get cited tend to be well-built pages.

But Ahrefs ran a controlled test across 1,885 pages in 2026 and found that adding schema on its own did not increase the number of citations. Treat it as disambiguation infrastructure that makes your content unambiguous to machines, not as a lever you pull for citations.

Lastly, consistency across a whole site is also important. A documented editorial QA workflow rather than trusting it to happen page by page helps.

Step 3: Be the Source of the Claim by Creating Original Content

💡 Citations typically flow to origins. When your page reports a statistic from someone else’s study, the model has a choice between your summary and the original, and it will usually take the original.

It is also the reason so much genuinely good content earns no citations. A well-researched synthesis of five sources is a candidate for zero citations, because every individual claim in it has a home somewhere else. Being an excellent librarian is not the same as being worth quoting.

Here are some ways you can become a source worth citing:

• Publish claims that have no other origin. Original data from your own book of business, benchmarks, survey results, or a teardown of something you actually ran. The minimum viable version is not a forty-page industry report. It is one number nobody else has, stated cleanly, with a methodology note attached.
• Own a definition or a framework. Named concepts create claims that can only be sourced back to you.
• Give origin claims their own page. If a proprietary number is worth citing, it should not be sitting in section nine of another page.
• Keep origin pages current. A stale figure gets superseded by a fresher one from someone else.
• Link out generously to originals. Counterintuitive, but it keeps your page positioned as the origin of the claims you did originate rather than a competitor for claims you didn’t.

Additionally, identical content earns citations 4.5x times more often when published on a third-party site than on a brand’s own domain — 7.6% against 34% across around a thousand prompt-platform combinations. But in that scenario, the citation goes to the publisher’s URL, not yours.

Guest posting and digital PR are excellent mention tactics. They only become citation tactics for your domain when the third-party piece attributes a claim back to a page you own. So pitch your original data, but ensure the link to your source page is a deliverable.

How Citation Behaviour Differs by AI Platform

💡Citation behaviour varies enough between engines that averaged reporting hides the picture. ChatGPT cites often and names rarely, Gemini does the reverse, and Perplexity leans more on community sources.

Before the table, a caveat: the majority of what works is shared across every platform. Structure and factual density move all of them, for example. Platform-specific tactics are worth maybe the last fifth of your effort, and they are the wrong place to start.

But the differences are real, still. Profound ran 100,000 prompts across both ChatGPT and Perplexity and found only 11% of cited domains overlapped between the two, with 37.4% cited exclusively by ChatGPT and 51.6% exclusively by Perplexity.

The source mixes below come from Profound’s separate 680-million-citation dataset unless noted.

Platform	Source mix	Citation behaviour	Practical implication
Google AI Overviews	Reddit around 21% of top sources. Roughly 43% of citations point to Google-owned properties, including YouTube.	Only 37.9% of citations come from top ten organic results (Ahrefs data).	Existing SEO carries over more here than anywhere else, but 2 out of 3 citations come from outside page one.
ChatGPT	Wikipedia is dominant at 47.9% of top sources; Reddit is far lower at 11.3%.	Cites in 87% of brand appearances (Semrush data)	Encyclopaedic and third-party validation beat your own blog. Expect ghost citations.
Perplexity	The most community-skewed engine. Reddit around 46.7% of top source share.	Profound finds review aggregators such as G2 and Gartner prominent on commercial queries.	Narrow, current, specific pages and strong review profiles do the work.
Gemini	Blogs, news, and video, with YouTube mentions the strongest single correlate Ahrefs measured.	Names brands 83.7% of the time but cites them only 21.4% (Semrush).	The inverse of ChatGPT. Optimise for mentions, not for citations.
Claude	Training data plus live retrieval, with a smaller published dataset than the others.	Less independently measured than the other four.	Not enough platform-specific guidance for a recommendation.

Platform-specific work becomes worth doing when your audience concentrates on one engine. A developer tooling company whose buyers live in Claude should be weighted differently than a consumer brand whose category gets researched in Gemini. You can find out which AI platforms are driving traffic to your website on GA4.

Looking to Get Cited By AI Engines? Here’s Where to Start

To get cited by AI, first ask, “Which gate am I failing, and am I actually the source of anything?”

Pick ten to twenty prompts your buyers genuinely ask. Input them into your AI citations tracking tool and find out who is getting the link. Quickly, you’ll know whether the problem is that models cannot reach you, cannot quote you, or have someone better to point to.

Most teams discover it is the third one, which is uncomfortable, because the fix is the toughest. It means publishing something nobody else can claim. That is slower than restructuring content. It’s also the only part that competitors can’t ever copy from you.

If you want a second pair of eyes on where your citation gap actually sits, that is the kind of work I do. Take a look at my AEO services page to find out how I can help.

FAQs