# AI Search Ranking Factors: What the Studies Show

> AI search ranking factors are the signals that make ChatGPT, Perplexity and Google's AI features cite a page or name a brand. The best evidence points to brand mentions across the web, pages that answer the many sub-questions an engine searches for, fresh content, and passages with quotes and statistics. Backlinks matter far less than in classic SEO.

By Othmane Khadri, founder of Earleads. Published 2026-10-18. Canonical: https://www.earleads.com/blog/ai-search-ranking-factors/

## What are AI search ranking factors?

AI search ranking factors are the signals that make an AI engine cite your page or name your brand in its answer. ChatGPT, Perplexity, Gemini, Claude and Google's AI Overviews and AI Mode don't publish them. So everything you read on the topic, including this article, is inferred from studies, from what the engines' owners document, and from watching answers change over time.

That's why the evidence matters more than the list. Most "AI ranking factors" articles give you 7 or 20 factors with equal weight and no sources. Some of those factors have large studies behind them. Some come from a single lab experiment. And some are repeated from blog to blog without anyone showing they work.

This guide goes through the studies one by one. Each figure below was checked on the page of the company that published it. We'll grade what's documented, what's measured at scale, what's been tested in a lab, and what's only been repeated. Then we'll add what we see in practice. [Earleads](https://www.earleads.com/) runs AI search visibility programmes for B2B and B2C brands, so the last part comes from tracking buyer questions across engines every month, not from a study.

## The evidence at a glance

| Factor | What the evidence shows | Source | Evidence type |
|---|---|---|---|
| Indexable, snippet eligible pages | Required to appear as a supporting link in Google's AI features | [Google Search Central](https://developers.google.com/search/docs/appearance/ai-features) | Documented by the engine |
| Answering the sub-questions (query fan-out) | Only 38% of AI Overview citations rank in the top 10 for the query, down from about 76% | [Ahrefs, March 2026](https://ahrefs.com/blog/ai-overview-citations-top-10/) | Documented + measured |
| Brand mentions across the web | Correlate at 0.664 with AI Overview visibility, against 0.218 for backlinks | [Ahrefs, May 2025](https://ahrefs.com/blog/ai-overview-brand-correlation/) | Measured (correlation) |
| YouTube mentions | Strongest correlation (about 0.737) across ChatGPT, AI Mode and AI Overviews | [Ahrefs, December 2025](https://ahrefs.com/blog/ai-brand-visibility-correlations/) | Measured (correlation) |
| Freshness | AI cited pages are 25.7% fresher than organic results | [Ahrefs, July 2025](https://ahrefs.com/blog/do-ai-assistants-prefer-to-cite-fresh-content/) | Measured |
| Being on the domains engines read | Each engine has its own favourite sources, and they shift | [Semrush](https://www.semrush.com/blog/most-cited-domains-ai/), [Peec AI](https://peec.ai/blog/top-domains-cited-by-ai-search-analysis-based-on-30m-sources) | Measured |
| Quotes, statistics, cited sources | Lifted visibility in generative answers by up to 40% | [Aggarwal et al., KDD 2024](https://arxiv.org/abs/2311.09735) | Lab experiment |
| llms.txt, special AI markup | Google says neither is needed for its AI features | [Google Search Central](https://developers.google.com/search/docs/appearance/ai-features) | Repeated, not shown |

A word on the third column. A correlation across 75,000 brands is strong evidence that two things travel together. It isn't proof that one causes the other. The Ahrefs authors say so themselves, and we'll keep that caveat attached to every number.

## How AI search engines choose sources

Before the factors, the mechanics. An AI engine doesn't rank ten blue links. It reads a question, runs one or more searches, retrieves a pile of pages, and writes an answer that cites some of them.

Google describes the middle step openly. Its [AI features documentation](https://developers.google.com/search/docs/appearance/ai-features), last updated December 10, 2025, says AI Overviews and AI Mode "may use a 'query fan-out' technique, issuing multiple related searches across subtopics and data sources to develop a response." One buyer question becomes many searches. A question about project management software might fan out into pricing, alternatives, reviews and "for small teams."

That changes what ranking means. In July 2025, Ahrefs found about 76% of pages cited in AI Overviews also ranked in Google's top 10 for the same query. When it reran the analysis on 863,000 search results in March 2026, that share had fallen to 37.9%. Another 31.2% of cited pages ranked between positions 11 and 100, and 31.0% ranked beyond the top 100 ([Ahrefs, March 2026](https://ahrefs.com/blog/ai-overview-citations-top-10/)). The authors point to fan-out, noting that since AI Overviews moved to Gemini 3, "fan out queries are playing a bigger role in source selection."

![Diagram showing one buyer question splitting into four sub-queries about pricing, alternatives, reviews and small teams. Pages that show up across many sub-queries are highlighted and lead to being cited, while the top 10 for the original question only leads to citation less often](/blog/ai-search-ranking-factors/diagram-2.svg)
*Engines search the sub-questions. Pages that answer several of them get cited, whatever they rank for the head term.*

The assistants outside Google show the same pattern, more strongly. Across 15,000 long-tail queries, Ahrefs found only 12% of the URLs cited by AI assistants ranked in Google's top 10 for the original prompt. Perplexity overlapped most, at 28.6%. ChatGPT's in-text citations overlapped at 8%, Gemini at 8.6% and Copilot at 8.2% ([Ahrefs, August 2025](https://ahrefs.com/blog/ai-search-overlap/)).

So the first factor isn't a factor at all. It's a shift in unit. You don't win AI search one keyword at a time. You win a buyer question by having good answers to the questions around it.

## Factor 1: brand mentions beat backlinks

This is the best supported finding in the field, and the one that most changes how budgets should be spent.

Ahrefs studied 75,000 brands and measured how closely different signals tracked their visibility in Google's AI Overviews. Branded web mentions came first, with a Spearman correlation of 0.664. Branded anchors followed at 0.527 and branded search volume at 0.392. The classic link metrics trailed, with Domain Rating at 0.326, referring domains at 0.295 and the number of backlinks at 0.218 ([Ahrefs, May 2025](https://ahrefs.com/blog/ai-overview-brand-correlation/)).

![Bar style diagram of six signals and their correlation with AI Overview brand visibility. Branded web mentions at 0.664 is highlighted, followed by branded anchors 0.527, branded search 0.392, Domain Rating 0.326, referring domains 0.295 and backlinks 0.218](/blog/ai-search-ranking-factors/diagram-1.svg)
*Being talked about tracks AI visibility about three times more closely than being linked to.*

A follow-up in December 2025 extended the study to ChatGPT and AI Mode. YouTube mentions showed "the strongest correlation with AI visibility (~0.737), outperforming every other factor across ChatGPT, AI Mode, and AI Overviews." Branded web mentions stayed near the top everywhere, at 0.709 in AI Mode, 0.664 in ChatGPT and 0.656 in AI Overviews ([Ahrefs, December 2025](https://ahrefs.com/blog/ai-brand-visibility-correlations/)).

Both studies carry the caveat in the authors' own words. "Correlation isn't causation." Big brands get mentioned more and get named more, so some of the relationship is simply size. Still, the direction is consistent across three engines and two studies. Engines write answers from what the web says about you, and a link without a sentence around it says very little.

In practice this means your earned work should aim at pages that already appear in answers to your buyer questions, not at the highest authority site you can get a link from. We cover that outreach in [how to get cited by AI](/blog/how-to-get-cited-by-ai/).

## Factor 2: freshness, which matters more outside Google

Ahrefs analysed 16.975 million URLs cited by seven AI platforms and found AI cited content was 25.7% fresher than pages in regular Google results. The average cited URL was 1,064 days old, against 1,432 days for organic results ([Ahrefs, July 2025](https://ahrefs.com/blog/do-ai-assistants-prefer-to-cite-fresh-content/)).

The spread between engines is the useful part. ChatGPT's citations were the youngest, at 958 days on average. Google's AI Overviews were the oldest, at 1,432 days, which is about the same as organic results. As the author notes, Google is the least influenced by freshness, and Google is still where most searches happen.

Two cautions. The average cited page is still almost three years old, so freshness is a tiebreaker, not a substitute for being the best answer. And changing a date without changing the content is the wrong lesson. Refresh the facts, prices, screenshots and examples on the pages that answer your buyer questions, then update the date.

## Factor 3: being on the domains each engine reads

Engines don't read the web evenly. Each one has favourite sources, and they move.

Semrush tracked over 230,000 prompts across ChatGPT search, Google AI Mode and Perplexity for 13 weeks, covering more than 100 million citations. ChatGPT cited the largest discussion forum in close to 60% of its responses in early August 2025, then around 10% by mid September ([Semrush, November 2025](https://www.semrush.com/blog/most-cited-domains-ai/)). One product change, and a source that was in most answers dropped out of most of them.

Peec AI analysed 30 million cited sources in the US across ChatGPT, AI Mode, Gemini, Perplexity and AI Overviews. A discussion forum, YouTube, LinkedIn and Wikipedia led overall, with Forbes and G2 close behind. But the order changes by engine. ChatGPT's top source was Wikipedia, while AI Mode and AI Overviews put YouTube first ([Peec AI](https://peec.ai/blog/top-domains-cited-by-ai-search-analysis-based-on-30m-sources)). Their own advice is to look at which domains matter "for your specific industry and your specific prompts, not following a generic list."

This is the factor most lists get wrong. "Get on Wikipedia" or "post on forums" is advice for an average that no single brand lives in. A B2B software buyer's answers lean on review sites, comparison articles and practitioner threads. A B2C buyer's answers lean on video, retailer pages and consumer reviews. The only way to know is to run your own questions and read the sources.

## Factor 4: how the page is written

The one controlled experiment in this list comes from academia. Researchers from IIT Delhi and Princeton built a benchmark of queries and tested how rewriting a source changed its visibility in generative answers ([Aggarwal et al., GEO, KDD 2024](https://arxiv.org/abs/2311.09735)). They report that these methods "can boost visibility by up to 40%."

The winners were adding quotations, adding statistics and citing sources. Keyword stuffing, the old SEO habit, offered "little to no improvement." The most interesting result was who gained. Lower ranked sources gained most, with a source in fifth position improving by 115.1% when it cited its sources, while the top source lost ground.

It's a lab result on a test engine, not a measurement of ChatGPT in the wild, so treat the percentages loosely. The lesson holds up in practice though. Engines lift passages that make a specific, checkable claim, attribute it, and stand on their own when quoted. That's also why every number in this article has a named source next to it.

## What Google says you don't need

Some popular factors have no evidence behind them, and in Google's case the owner has said so directly.

Google's documentation says "there are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary." It goes further. "You don't need to create new machine readable files, AI text files, or markup to appear in these features. There's also no special schema.org structured data that you need to add" ([Google Search Central](https://developers.google.com/search/docs/appearance/ai-features)).

What Google does ask for is ordinary. Let crawlers in, including through your CDN. Link important pages internally. Put important content in text, not only in images or PDFs. Make sure your structured data matches the visible text. To appear as a supporting link, a page must be "indexed and eligible to be shown in Google Search with a snippet."

So llms.txt files and "AI schema" packages aren't ranking factors on any evidence we've seen. They may be harmless. They shouldn't be what you pay for.

## Why "ranking" is the wrong word for AI search

Classic SEO has positions. AI search doesn't, and this is the factor most studies skip.

SparkToro and Gumshoe had 600 volunteers run 12 prompts through ChatGPT, Claude and Google's AI a combined 2,961 times. They found "a <1 in 100 chance" that ChatGPT or Google's AI would return the same list of brands in any two responses, and roughly 1 in 1,000 that two lists would come back in the same order ([SparkToro, January 2026](https://sparktoro.com/blog/new-research-ais-are-highly-inconsistent-when-recommending-brands-or-products-marketers-should-take-care-when-tracking-ai-visibility/)). Their verdict on tools that report a ranking position in AI was blunt. "Full of baloney."

That has a practical consequence for everything above. You can't test a ranking factor with one screenshot. Any change you make has to be measured as an appearance rate across many runs, per engine, before and after. Our guide to [AI share of voice](/blog/ai-share-of-voice/) covers how many runs make a number trustworthy.

## ChatGPT ranking factors vs Google AI ranking factors

The studies agree on direction but differ by engine, in measurable ways.

| | ChatGPT | Perplexity | Google AI Overviews and AI Mode |
|---|---|---|---|
| Overlap with Google's top 10 | 8% (in-text citations) | 28.6% | 37.9% for AI Overviews, down from about 76% |
| Freshness of cited pages | Youngest, 958 days on average | 1,166 days | AI Overviews oldest, 1,432 days |
| Top cited source (Peec AI) | Wikipedia | A discussion forum | YouTube |
| Strongest brand signal (Ahrefs) | YouTube mentions | Not studied | YouTube mentions, then branded web mentions |

Sources: [Ahrefs overlap](https://ahrefs.com/blog/ai-search-overlap/), [Ahrefs top 10](https://ahrefs.com/blog/ai-overview-citations-top-10/), [Ahrefs freshness](https://ahrefs.com/blog/do-ai-assistants-prefer-to-cite-fresh-content/), [Peec AI](https://peec.ai/blog/top-domains-cited-by-ai-search-analysis-based-on-30m-sources), [Ahrefs brand factors](https://ahrefs.com/blog/ai-brand-visibility-correlations/).

Read it as a set of tendencies, not rules. Perplexity is closest to classic search, so pages that already rank do well there. ChatGPT rewards recent content and leans on reference sources. Google's AI features still start from Google's index but now cite deep into the long tail through fan-out. If you want more on the first engine, see [how to get ChatGPT to recommend your business](/blog/how-to-get-chatgpt-to-recommend-your-business/).

## What we see in practice

Studies average across millions of answers. Running buyer questions for specific brands every month adds a few patterns the studies don't show.

**Reach isn't the signal.** Only a small minority of the community threads a brand takes part in ever get quoted by AI, and they're rarely the viral ones. The threads that earn citations are low-view threads whose title is the literal buyer question. Comparisons, migrations ("what are people moving to before X shuts down") and "X alternative" threads do best. An agency reporting reach is reporting the part that doesn't compound.

**Question threads survive.** In discussion forums, posts written as a question tend to survive moderation where story posts get removed. Threads removed after they were indexed often keep being cited. And one topic needs one thread, because cross-posts on the same question compete with each other.

**Comparison pages win on craft.** The site pages that get extracted most are comparison pages where the brand is one of the two names. Put the comparison table first, then a 40 to 60 word verdict on who each option is for, including who you're not for, then honest weaknesses. Competitors' winning comparison pages often rank on very few referring domains, which fits the mentions over links finding above.

**Referral traffic is the wrong scoreboard.** AI sends citations, not clicks. Judge a programme by referrals and you'll cancel the channel buyers use to build shortlists. We measure share of answers on questions that don't name the brand, branded search lift and what buyers say at signup. [GEO KPIs](/blog/geo-kpis/) covers the full set.

## Step 1: Write down the buyer questions, not keywords

Start with 30 to 60 questions your buyers actually ask an AI assistant, taken from sales calls, support tickets and won deals. Have someone in sales sign the list off, so nobody picks questions that flatter a dashboard. Include questions that don't name your brand, because that's where shortlists form.

## Step 2: Run each question live and read the sources

Run every question natively in each engine you care about, several times. Record whether you appear, how you're described and which pages are cited. This gives you the only list of ranking factors that matters, which is the sources your engines use for your questions. Our article on [what an AI visibility audit should show you](/blog/what-an-ai-visibility-audit-should-show-you/) goes through what to capture, and an [AI visibility audit](/#audit) runs this check across engines for you.

## Step 3: Fix the entry ticket first

Check that AI crawlers can reach your pages, that important content renders as text, and that pages are indexed and snippet eligible. None of the other factors count if an engine can't read you.

## Step 4: Answer the sub-questions on your own site

For each buyer question, list the questions an engine would fan out to (pricing, alternatives, integrations, limits, who it's for) and make sure a page of yours answers each one in a passage that stands alone. Add named sources, quotes and dated facts. Refresh those pages when the facts change.

## Step 5: Earn mentions on the pages already cited

Take the cited sources from Step 2 and work through them. Get named in the comparison articles, review listings, videos and community threads engines already read for your questions, with concrete detail that gives an engine a reason to pick you. Then re-run the questions and compare appearance rates, per engine, against your baseline.

![Diagram grading AI ranking factors by evidence. Documented by the engine leads to do it, it's the entry ticket. Measured at scale, highlighted, leads to bet on it and measure your own. Tested in a lab leads to use it in how you write. Repeated but not shown, such as llms.txt, leads to don't budget for it](/blog/ai-search-ranking-factors/diagram-3.svg)
*Spend in proportion to the evidence. The measured factors are where the budget belongs.*

## What changes for B2B and B2C brands

The factors are the same, but where they play out differs.

For B2B, buyer questions are long and specific, so fan-out reaches deep into comparison pages, integration docs and practitioner threads. Mentions on review sites and in industry write-ups carry a lot of weight, and a wrong description of your pricing or product can reach sales before you do. If that's a concern, see [AI reputation management](/blog/ai-reputation-management/).

For B2C, answers lean more on video, retailer and marketplace pages, and consumer reviews, which fits YouTube's strong correlation in the Ahrefs data. Product facts need to be consistent across your site, your listings and your structured data, because engines compare them.

In both cases the order of work is the same. Find the questions, read the sources, fix access, answer the sub-questions, earn mentions where engines already look, and measure across many runs. Anyone promising a fixed position in ChatGPT is promising something nobody controls.

## Frequently asked questions

### What are the most important AI search ranking factors?

The strongest published evidence points to four. Brand mentions across the web, which Ahrefs found track AI visibility far more closely than backlinks. Pages that answer the sub-questions engines search behind a query. Content that's been kept current. And passages that carry quotes, statistics and cited sources. Google adds the baseline, which is a page that's indexed and eligible to show a snippet.

### Do backlinks matter for AI search?

Less than in classic SEO. In Ahrefs' study of 75,000 brands, branded web mentions correlated with AI Overview visibility at 0.664, against 0.218 for the number of backlinks. Links still help pages get crawled, indexed and ranked, which feeds retrieval. But being talked about on pages engines read matters more than counting links to your domain.

### What are ChatGPT ranking factors?

ChatGPT doesn't publish ranking factors. Studies show it cites pages that often don't rank in Google's top 10 for the same prompt (8% overlap in Ahrefs' test), leans toward fresher content than Google, and has relied heavily on Wikipedia and forum threads. Its sources also swing month to month, so measure your own questions rather than copying a generic list.

### Does llms.txt help you rank in AI search?

There's no published evidence that it does. For its own AI features, Google says you don't need to create new machine readable files, AI text files or markup to appear, and that there's no special schema.org structured data you need to add. Treat llms.txt as optional housekeeping, not a ranking factor, and spend the budget on pages and mentions.

### Why do AI search results change every time I ask?

Because AI answers are generated, not looked up. SparkToro and Gumshoe ran 12 prompts 2,961 times and found less than a 1 in 100 chance that ChatGPT or Google's AI gives the same list of brands twice. That's why there's no stable AI ranking position to chase. Measure how often you appear across many runs instead.

### Are AI search ranking factors the same for every engine?

No. Ahrefs found Perplexity's citations overlap with Google's top 10 far more than ChatGPT's do, and that ChatGPT leans hardest toward fresh content while Google's AI Overviews lean least. Peec AI found ChatGPT's most cited domain is Wikipedia, while Google's AI features lean on YouTube. Check which sources each engine cites for your own buyer questions.