How to Rank on ChatGPT (What 1.4M Prompts Reveal)

ChatGPT read your page last week. It just didn't tell anyone.

That gap is the whole problem. Most teams asking how to rank on ChatGPT are optimizing for the wrong finish line, because there are two finish lines and only one of them is visible.

Ahrefs analyzed 1.4 million ChatGPT prompts and found the model cites only about half the URLs it retrieves. Getting pulled into the answer is one contest. Getting named in it is a separate one.

This playbook covers both. By the end you'll know which channel ChatGPT actually cites from, why your page title matters more than your word count, and what to stop doing today.

Key Takeaways

• ChatGPT pulls roughly 16 cited and 16 non-cited URLs per prompt, and only 49.98% of retrieved URLs get cited (Ahrefs, April 2026). Retrieval is not visibility.
• Pages arriving through ChatGPT's general search channel get cited 88.46% of the time. Pages from its dedicated Reddit feed get cited 1.93%. Ranking in search is still the entry ticket.
• Titles do the selecting. Cited pages scored 0.602 on title-to-prompt similarity versus 0.484 for non-cited ones, and natural language URL slugs hit an 89.78% citation rate versus 81.11% without.
• Off-site brand signals beat links. Across 75,000 brands, branded web mentions correlated with AI visibility at 0.66 to 0.71 while total site pages sat near 0.194 (Ahrefs, 2026).
• Earned media accounts for 84% of AI citations. Paid and advertorial content accounts for 0.3% (Muck Rack, May 2026). Your blog is not the main event.

Ranking on ChatGPT Isn't One Contest, It's Two

Here's what happens when someone asks ChatGPT a question that triggers a web search.

The model breaks the prompt into smaller sub-questions, often called fan-out queries. It pulls back candidate pages for each one. Then it decides which of those pages to open, read, and credit with a numbered blue link.

Two gates. Retrieval, then citation.

The Ahrefs study found ChatGPT pulls about 16.57 cited URLs and 16.58 non-cited URLs per prompt. So for every page that gets named, roughly one other page got read and quietly used to shape the answer with no credit at all.

Almost every AI SEO checklist you've read optimizes for gate one, then wonders why nothing shows up. Crawlability, schema, sitemaps: all of that gets you retrieved. None of it gets you cited.

The second gate is where the work is.

88% of ChatGPT's Citations Come From Plain Old Search Results

ChatGPT tags every source it retrieves with an internal label for the channel it came through. Ahrefs found five: search, news, reddit, youtube, and academia.

The citation rates are not close.

The general search index wins on both volume and citation rate, and 88% of the URLs ChatGPT ends up citing come straight out of it.

Read that again if someone told you AI search made rankings irrelevant.

It didn't. It changed what a ranking buys you. Ranking well no longer just wins clicks, it wins a seat in the pool where nearly every citation gets handed out.

Now the finding that should reset a few strategies.

Reddit gets read constantly and credited almost never

Reddit has its own dedicated retrieval feed inside ChatGPT, and it is enormous: over 16 million data points in the dataset. Yet 67.8% of all non-cited URLs come from Reddit.

The model mines Reddit to understand what people think and where consensus sits, then hands the citation to a more presentable source. It learns from the crowd and cites an institution.

So Reddit still has value. It shapes how ChatGPT talks about your category, and your brand name can land in the answer text. What it won't reliably buy you is a visible link.

The same pattern hits YouTube and academic sources through their dedicated feeds. Pulled in at scale, almost never surfaced as citations.

Your Title Is Doing More Work Than Your Content

This is the finding that should change your next content brief.

When ChatGPT retrieves a result, it gets back a title, a URL, sometimes a short snippet, and an ID number. It uses that metadata to decide which pages are worth opening at all. There's a gatekeeping layer running before ChatGPT reads a single word of your actual page.

Your 3,000-word masterpiece never gets opened if the title doesn't clear that bar.

How much titles matter, in numbers:

• Prompt versus cited page title: 0.602 similarity
• Prompt versus non-cited page title: 0.484
• Best-matching fan-out query versus cited title: 0.656

Notice the gap widens when you compare against the fan-out queries instead of the original prompt. That's the tell. You aren't writing titles for what the user typed, you're writing them for the sub-questions ChatGPT invents to answer it.

URLs matter too, and this one is embarrassingly cheap to fix. Search results with natural language slugs got cited 89.78% of the time, against 81.11% for those without.

If your CMS still ships /p?id=48219, that's nine points of citation rate sitting on the floor.

Write for the Sub-Questions, Not the Prompt

Fan-out queries are the practical center of this playbook, so here's how to work with them.

Take a prompt you want to win. Something like "best white label link building agency for small agencies." Now write out the sub-questions a researcher would need answered to reply well:

• What does white label link building include?
• How much does it cost per link?
• What minimum order sizes do these agencies have?
• How do they handle reporting and client-safe branding?
• Which ones work with agencies under 10 people?

Every one of those is a candidate H2 with a direct answer beneath it. Any one with real search volume behind it is a page title of its own.

Then check your page against the list honestly. Not "do we mention pricing" but "does a model skimming this page find a specific number attached to a specific offer."

Three habits do most of the work:

1. Front-load the answer. Give the direct response in the first 40 to 60 words of a section, then explain. A model borrowing a claim takes the clean, self-contained sentence over the one that needs three paragraphs of setup.
2. Make claims quotable. One idea per sentence, specific numbers, no pronouns pointing back at earlier paragraphs. If a sentence can't survive being lifted out of context, it won't get lifted.
3. Match the language of the question. If people ask "how much does digital PR cost," your H2 says that, not "Investment Considerations."

You can pull real fan-out queries out of AI visibility tools now, including Ahrefs Brand Radar, which shows the sub-questions generated for a prompt next to the URLs that got cited. Comparing your coverage against a competitor's on the same prompt is the fastest diagnosis available.

Freshness Is a Tiebreaker, Not a Strategy

The received wisdom is that AI loves fresh content. It's half right, and the wrong half will waste your quarter.

Across 17 million citations, Ahrefs found ChatGPT cited URLs 458 days newer than Google's organic results, the strongest freshness preference of any platform they tested. So yes, ChatGPT skews recent compared to Google.

Inside a single prompt's retrieval set, the picture flips. The median cited page from the search index was around 500 days old, some cited pages ran past 2,700 days, and the non-cited pages were overwhelmingly very young.

Both things are true at once. The overall population of AI citations skews younger than Google's, and within any given answer the established page usually beats the brand-new one.

The lesson: a new page that matches the fan-out queries well gets cited, and a new page that doesn't gets retrieved and ignored. Relevance does the heavy lifting. Freshness only breaks ties.

Where freshness genuinely rules is news. In that channel, title relevance scores for cited and non-cited pages were nearly identical, so the model fell back on age, and cited news pages skewed to a median around 200 days versus 300 for non-cited.

For your calendar: republish and deepen evergreen pages instead of churning thin new ones, and treat news content as a separate motion with a separate speed requirement.

Links Got You Rankings. Mentions Get You Answers.

Now the uncomfortable part for anyone whose entire AI plan lives on their own domain.

Ahrefs studied 75,000 brands to see which signals correlate with visibility across ChatGPT, AI Mode, and AI Overviews. Branded web mentions correlated at 0.66 to 0.71. YouTube mentions came in strongest at about 0.737. Backlinks trailed well behind, and the number of pages on a site had almost no relationship with AI visibility at roughly 0.194.

Publishing volume is close to noise. Sit with that for a second.

ChatGPT specifically is the least impressed by classic authority metrics. Branded search volume correlated at 0.352 and Domain Rating at 0.266, both noticeably weaker than for Google's AI Mode.

Fair warning, and Ahrefs flags it themselves: correlation is not causation. Starting a YouTube channel tomorrow does not summon citations. The thing underneath all these metrics is brand authority, and the metrics are symptoms of it.

Independent data points the same way. Muck Rack analyzed more than 25 million cited links across ChatGPT, Claude, and Gemini and found earned media accounts for 84% of all AI citations, journalism alone makes up 27%, and paid or advertorial content accounts for 0.3%. Across three editions of that study going back to July 2025, earned media has ranged from 82% to 89%.

That range is the useful part. It means this isn't a quirk of one model update. It's how these systems source.

So the budget question answers itself. If you're spending 90% of your AI visibility effort on your own blog, you're competing for the thinnest slice of citations on offer. Digital PR, original data, and expert commentary put your name inside the 84%.

The Playbook, in Order

Do these in sequence. The later steps don't work without the earlier ones.

1. Get retrievable. Allow the AI crawlers in robots.txt, including OpenAI's search crawler, and confirm your pages are indexed beyond Google. Submit sitemaps in Bing Webmaster Tools. It costs an afternoon.
2. Fix your URLs. Human-readable slugs everywhere. Cheapest win on the list.
3. Rewrite titles against fan-out queries. For each priority page, list the sub-questions, then make the title match the strongest one. Vague brand-voice titles are the biggest silent loss in most content libraries.
4. Restructure for extraction. Question-shaped H2s, direct answer in the first 40 to 60 words of each section, one claim per sentence, real numbers.
5. Raise fact density. Cite real sources with real figures and name them in the text. Models pick pages that let them borrow verifiable claims.
6. Go earn mentions. Original data studies, expert commentary, journalist responses, category roundups you can honestly appear in. Target the publications your category's answers already cite, which any AI visibility tool will show you.
7. Refresh on a cadence. Quarterly on money pages. Update the substance, not the date stamp.

Steps 1 to 5 are a few weeks of work. Step 6 is the program. Step 7 is the habit.

What to Stop Doing

Chasing llms.txt. No major AI platform has confirmed using it for retrieval or ranking. Add it if you want. Don't call it strategy.

Publishing more. Page count barely correlates with AI visibility. Ten pages that answer real sub-questions beat a hundred that answer none.

Buying placements. At 0.3% of citations, paid content is a rounding error in AI answers.

Reporting mentions alone. Being named in an answer and having your site cited as the source are different outcomes with different fixes. Track both.

How to Measure Without Fooling Yourself

Most teams can't yet. Semrush's 2026 AI Visibility Index, built on 126 million US AI search prompts from January through April 2026, found 45% of marketing leaders cannot accurately measure brand visibility in AI answers, and only 9% have tools covering every relevant metric.

Three rules keep reporting honest.

Measure per engine. ChatGPT cites an average of 15 sources per response while Gemini cites three. A strategy tuned to one can look like failure on the other.

Separate mentions from citations. Every report, every time.

Re-measure often. Citation patterns move fast enough that quarterly is the floor, not the ideal.

And connect it to everything else. Semrush found teams running AI visibility as part of existing SEO, content, and brand programs reported better results than teams treating it as a standalone project.

FAQ

How long does it take to rank on ChatGPT?

The structural work shows up fastest. Title, URL, and formatting changes can affect citation within a crawl cycle or two, often weeks. Mention-building runs on PR timelines, so expect a quarter or more before it moves your visibility in a way you can see in reporting.

Does ChatGPT use Google or Bing?

ChatGPT's live search layer draws on its own crawler plus third-party search infrastructure rather than Google's index. The practical takeaway is that Google rankings alone don't guarantee retrieval, so cover Bing indexing and allow OpenAI's crawlers as basic hygiene.

Do backlinks still matter for AI visibility?

Yes, but indirectly. Links help you rank in the search channel that supplies most ChatGPT citations. What they don't do is predict AI visibility as well as unprompted brand mentions, which is why link building and digital PR now need to run together rather than compete for the same budget.

Is FAQ schema required to get cited?

No, and no study has isolated schema as a citation driver. It's cheap and it helps machines parse your page, so it's worth having. It won't rescue a page whose title doesn't match the questions being asked.

Why does ChatGPT mention my competitor but link to a magazine?

Because that's the normal pattern. Most citations land on earned media rather than brand domains, so the model names brands while crediting the publications that wrote about them. Getting into those publications is the fix.

The Short Version

ChatGPT isn't a mystery box. It reads a lot, cites about half, and picks the pages whose titles and URLs match the sub-questions it invented, from a pool that mostly comes out of ordinary search rankings. Then it credits the publications that wrote about you more often than it credits you.

So the work splits cleanly. Structure your own pages to be extractable, then spend the rest of your effort becoming the brand credible publications mention without being asked.

If that second half is your gap, it's what digital PR and AI visibility exist for, and it's the same muscle as link building aimed at a different target.

Which are you fixing first, the titles or the mentions?