How to Get Cited by ChatGPT, Perplexity & AI Overviews

To get cited by ChatGPT, Perplexity, and Google AI Overviews, you need to publish content that an AI system can extract a clean answer from, verify against other sources, and trust enough to name. That means direct answers in the first fifty words of a section, question-shaped headings, real statistics with sources attached, schema markup that labels what the page is, and a refresh cycle measured in weeks rather than years. The mechanics differ by platform, and the gap between them is wider than most teams expect.

Here is the process we use, and the parts of it that are still guesswork.

Updated August 2026: three things moved since we first published this. The link between ranking and citation weakened sharply, the measured payoff for being cited got much larger, and Google shipped a publisher control worth knowing about. Details below.

Why This Is Now a Separate Job From SEO

Search stopped being a list of links for a large share of queries. We covered the strategic shift in SEO is not enough anymore — welcome to GEO; this piece is the operational version. BrightEdge put AI Overview coverage at roughly 48% of queries in February 2026, and Google itself has said about half of US queries now trigger one.

AI Mode, Google's fully generative interface, is the bigger story. Google announced at I/O 2026 that it had passed a billion monthly users, and the rollout now covers close to 200 countries in 98 languages. Where AI Overviews summarise above the links, AI Mode mostly replaces them.

The traffic effect is real. SparkToro's 2026 analysis found zero-click searches reaching 68%, so fewer than one in three Google searches sends a click to any website at all. Ahrefs, measuring across 300,000 keywords in December 2025, found position-one click-through rates down somewhere between 34.5% and 58% when an AI Overview appears. Pew Research, working from browsing data rather than rank data, found users clicked a traditional result 8% of the time with an AI Overview present against 15% without. The studies disagree on magnitude. None of them disagree on direction.

So ranking first is no longer the finish line. Being named inside the answer is.

That reframe has a payoff attached, and it is larger than we reported a month ago. Seer Interactive's April 2026 update, drawn from 53 brands and 5.47 million queries, found cited brands earning about 120% more organic clicks per impression than uncited brands on the same AI Overview queries. Their earlier work put the paid CTR advantage at 91%. Citation does not replace ranking. It compounds it.

The Ranking Link Is Weaker Than It Was

This is the update that should change what you do. Six months ago the honest summary was that classic SEO qualifies you for citation. That claim is now shaky.

A page at position one still carries roughly a 53% chance of being cited, per Ahrefs data from March 2026. But BrightEdge found that 83% of AI Overview citations came from pages sitting outside the organic top ten. ALM Corp, using a different method, put the same figure at 62%. Both are dramatic moves from mid-2025, when the top ten supplied the clear majority of citations.

Read those two findings together, because they are not in conflict. Ranking first is still the single best thing you can do for your odds on any given query. There are simply far more pages outside the top ten, and the models are increasingly willing to pull from them.

The practical consequence: a page that ranks fourteenth for a competitive term is no longer out of the running. Rewriting it to answer cleanly may earn a citation long before it earns a first-page ranking.

The Three Engines Do Not Share a Playbook

This is the part teams get wrong most often. They optimize once and assume it carries everywhere.

It does not. When Profound ran 100,000 prompts through ChatGPT and Perplexity, only 11% of cited domains appeared in both. A third were cited only by ChatGPT, half only by Perplexity. Whitehat's independent study of 118,000 responses landed on the same 11%.

The reason is architectural:

• ChatGPT runs a blended stack. It launched on Bing's index and still draws on it, but OpenAI now operates its own crawler, OAI-SearchBot, and its own index alongside. Two jobs follow from that: confirm you are indexed in Bing Webmaster Tools, and confirm your robots.txt is not blocking OAI-SearchBot. Plenty of teams have never checked either.
• Perplexity uses its own index and weighs recency heavily. Reddit alone accounts for a large share of what it cites, so community presence matters more here than anywhere else.
• Google AI Overviews and AI Mode pull from Google's index, so conventional SEO signals carry the most weight of the three.

One playbook will not cover all three. Verify your Bing indexation and crawler access, put your fastest-moving content on a monthly cycle for Perplexity, and keep doing technical SEO for Google.

The Process

1. Answer in the first fifty words

Every section should open by answering the question in its heading, plainly, before any context or setup. AI systems extract passages, not pages. A paragraph that spends four sentences warming up gives the model nothing clean to lift.

Write the answer. Then explain it.

2. Make headings match real questions

Use the phrasing people actually type or say. "How much does a fractional CMO cost?" gets extracted. "Pricing Considerations" does not. Question-shaped headings map onto how these systems retrieve information internally.

3. Attach a source to every number

This is the highest-return habit on the list, and the most commonly skipped.

Language models prefer information they can cross-check against something else. A statistic with a named source and a date can be corroborated. A bare number, floating in a sentence, cannot be, so it gets passed over in favour of one that can.

Cite inline, name the source, and link it. If you cannot source a number, cut it or state it qualitatively. Never invent one.

4. Publish something only you have

Original data is the strongest citation asset available to a small company. Survey your customers. Publish the results of your own audits. Report what you actually observed across your client base, anonymised.

You cannot out-authority Wikipedia on a definition. You can be the only source for a number that does not exist anywhere else.

5. Add the schema that describes the shape of your content

FAQPage schema lines up with the question-and-answer structure these systems use internally. HowTo schema suits instructional queries. Article schema with a real author, a real datePublished, and an accurate dateModified gives the machine something to trust.

Schema will not rescue thin content. It makes good content legible.

6. Understand where authority actually sits

The tidy version of this advice used to be that domain authority predicts citation. The 2026 research does not support it cleanly. Surfer analysed five million citations and found domain authority predicted almost nothing; other studies put the correlation at the top of their list. When credible datasets disagree this hard, treat the metric as unsettled and stop optimising for the score itself.

What the data does agree on is concentration. The 5W citation source index found roughly 68% of AI citations going to the top fifteen publisher domains. Wikipedia dominates what ChatGPT cites. Reddit dominates Perplexity.

The read for a small company is not to chase a domain rating. It is to be present where the citations already concentrate: the Reddit threads, the YouTube explainers, the industry publications your buyers read. Being quoted on a domain the model already trusts is worth more than another link to your own.

7. Make your entity legible

An AI system has to work out who you are before it will name you. That sounds abstract; the fix is not.

Describe the business the same way everywhere — site, LinkedIn, directories, press. If your About page says "fractional CMO firm" and your LinkedIn says "growth consultancy" and a directory lists you as a "digital agency", you have handed the model three entities instead of one. Organization schema with a consistent name, URL, and logo ties those together. So does an author page for the person whose name sits on the articles.

This is quiet work with no immediate payoff. It is also what separates a company the model can confidently name from one it describes vaguely and never credits.

8. Refresh on a schedule, not on a whim

Content updated inside the last 30 days earns roughly 3.2 times the citations of older pages, according to research from Goodie AI reported by Adweek. Perplexity shows the strongest recency bias of the three engines. Put your most competitive pages on a 30-day cycle and everything else on 90.

Update the substance and the dateModified field together. Changing the date without changing the content is a trick that stops working the moment anyone checks.

Holding that cadence by hand is the part teams underestimate; we run ours with agentic workflows so the calendar does not depend on anyone remembering. The scheduling matters more than the ambition. A 30-day cycle you actually hold beats a rewrite plan that slips two quarters.

9. Claim your preferred-source button

New as of August 2026, and cheap to do. Google rolled out Preferred Sources to AI Mode and AI Overviews in May, letting people mark publishers they want to see more of. On 20 August it released an embeddable button so you can ask readers to do that from your own site.

Google reported that people are twice as likely to click through to a source they have marked as preferred, and that more than 345,000 unique sources had been selected by the May launch. This will not do much for a company nobody has heard of. If you have an audience that already reads you, it converts that goodwill into search visibility, which is otherwise hard to do.

What Does Not Work

Betting on llms.txt. The picture here has improved slightly and still does not justify a strategy. Anthropic and Perplexity have both publicly confirmed that they read the file. Google has said plainly that it does not support it and has no plans to, and OpenAI has confirmed nothing either way. So it is a real convention with partial adoption and no standards body. Publish one — ours took twenty minutes. Do not build a plan around it, and be wary of anyone selling it as the key to AI visibility.

Keyword stuffing, in any modern form. These systems work on meaning, not term frequency. Repetition reads as low quality.

Chasing every platform at once. Pick the two where your buyers actually are. For most B2B companies that is Google and ChatGPT.

Writing for machines. The extraction-friendly structure above is genuinely good writing: direct answers, clear headings, sourced claims. Content that reads as though it were assembled for a crawler performs badly with both audiences.

How to Tell Whether It Is Working

Rank tracking will not show you this. You need to ask the engines directly.

Run your ten most valuable questions through ChatGPT, Perplexity, and Google monthly. Record whether you are cited, which competitors are, and which specific URL got picked. That log becomes your baseline.

Three things worth logging beyond a yes or no. Whether the answer named your brand or merely linked you, because being named is the outcome with positioning value. Whether the cited URL is the page you would have chosen. And which competitor keeps appearing — that tells you whose content the model currently trusts more than yours.

Do not expect Search Console to settle it for you. Google folds AI Mode clicks and impressions into the same totals as classic results, with no way to separate them. Your reporting will show a blend, and you cannot break it apart.

Then watch referral traffic from chat.openai.com and perplexity.ai in analytics. The volume will look small next to organic. Judge it on conversion rate instead — that traffic tends to arrive further along in the decision, because the buyer has already had the explaining done for them.

Expect the numbers to be noisy. These systems are non-deterministic; the same question can return different sources on consecutive days. One month of data tells you very little. Three months tells you whether you are trending in the right direction.

Frequently Asked Questions

How long does it take to get cited?

For content on an already-indexed domain, weeks. For a new domain, longer, because the authority and entity signals take time to accumulate. Perplexity tends to pick up fresh content fastest.

Do I have to choose between SEO and AI citation?

No, though the relationship loosened through 2026. Position one still carries roughly a 53% citation probability, so ranking helps. But most AI Overview citations now come from pages outside the top ten, which means a mid-ranking page rewritten to answer cleanly is a live opportunity rather than a lost cause.

Does schema markup guarantee a citation?

No. It improves the odds that your content is parsed correctly. Thin content with perfect schema stays uncited.

Is being cited worth it if nobody clicks?

Cited brands earn about 120% more organic clicks per impression than uncited ones, per Seer Interactive. Even setting clicks aside, being named as the source in an answer your buyer trusts is a positioning win that a blue link ten rows down does not deliver.

Where to Start

Pick your five highest-value commercial questions. Rewrite those pages so each section answers its heading immediately, source every number, add FAQPage schema, and set a 30-day refresh reminder. Log your citation baseline before you change anything, so you can prove what moved.

Then widen the list. Given how many citations now come from outside the top ten, your ranking pages twelve through thirty are the most undervalued assets you own.

Most of this is unglamorous. It is also, so far, the part that works.

If you want an outside read on where your content stands with AI search today, that is what our marketing audit covers.