AI ranking factors: what the evidence says in 2026

You're probably here for one of three reasons. You asked ChatGPT or Google about your category and it named a competitor. Your boss asked whether the company shows up in AI answers. Or an agency has pitched you an "AI optimisation" package and you want to know what you'd be paying for.

Either way, you want the same answers. Which factors decide whether AI tools cite you? Which of them can you change? And what should you do first?

The lists that rank for "AI ranking factors" don't answer that well. Most give 6 to 20 factors written as facts, and few link to any evidence.

This article grades each factor by the strength of the evidence behind it, from official documentation down to opinion. Every study is linked where it's used, so you can check the numbers yourself.

What AI ranking factors actually are

AI ranking factors are the signals that make an AI tool more likely to cite your page. No AI company publishes them. What exists is platform documentation, which says what's required rather than what helps, plus large correlation studies and surveys of what search professionals believe.

To see why some factors matter more than others, it helps to know how an AI answer gets built.

How AI tools build an answer

Most AI search tools search the web first, then write the answer from what they find. Google says AI Overviews and AI Mode may use "query fan-out", running several related searches across subtopics. ChatGPT does the same when it needs current information. The model then picks which pages to use, and adds what it already learned in training.

A page has to get through steps 3 and 4 to have any chance at step 6. That's why the factors with the strongest evidence are about being found, not about how the page is written.

Where AI tools get their information

AI tools draw on two places: what the model learned in training, and what it finds by searching at the moment you ask. For most buying questions the live search decides which sources get cited, and those sources have changed sharply in 2026.

Training memory changes slowly. In Cyrus Shepard's survey of 131 search professionals, how well the model already knows your brand ranked third of 13 factors. You build it the same way you build mentions: by being written about.

Reddit is no longer one of ChatGPT's favourite sources

Reddit and Wikipedia were ChatGPT's most-cited sites for much of 2025 and early 2026. That has changed.

• Ahrefs tracked 2.4 to 3.8 million US ChatGPT answers a month. Reddit was cited in about 1 in 5 answers in March 2026 and about 1 in 43 by late September, a fall of 88%. Wikipedia fell 80% and YouTube 27%. All three are still among ChatGPT's most-cited sites.
• The biggest drop came in one week. Promptwatch measured Reddit at 3.83% of ChatGPT search citations from 18 July to 7 August 2026, then 0.52% from 14 to 17 August, a fall of 86%. OtterlyAI saw a 73% drop across 16 brand reports over the same days, while other AI tools barely changed.
• It has happened before. Semrush saw Reddit fall from close to 60% of ChatGPT responses in early August 2025 to around 10% by mid-September 2025.

OpenAI hasn't publicly explained the change. Promptwatch noticed that on 8 August 2026, the share of ChatGPT's searches aimed at one named website jumped from 0.37% to 16.8%. Its reading is that ChatGPT started checking brands' own sites directly, but it calls that a plausible link, not a proven one.

What replaced it: official sources

The sources gaining ground are official ones: brand websites, government bodies and regulators.

• In an Ahrefs test of 20,000 US prompts, 84.4% of answers to factual questions (prices, features, company details) cited at least one official source. Official websites got 63.4% of their citations. For comparison questions, the figures were 68.3% and 46.4%.
• Between March and September 2026, PubMed rose from 64th to 4th among ChatGPT's most-cited sites, and the US FDA entered at 8th. Explainer sites such as WebMD and Investopedia lost citations, as did Wikipedia and Britannica.
• OtterlyAI found official brand and competitor pages gained citations in 10 of the 15 reports where they appeared.
• Sites in OpenAI's publisher partnerships rose from 7.5% of citations among the top 200 cited domains in March to 15.8% in July, then 12.5% in September.

Google's AI answers draw on different sources

The Reddit drop was mostly a ChatGPT change. Over the same August days, Promptwatch saw Reddit's share fall only 11% in AI Overviews and 31% in AI Mode, both gradually. In AI Overviews, Ahrefs found YouTube is the most-cited site, at 5.6% of cited pages and up 34% in six months.

How strong is the evidence for each factor?

Here's the short version. The sections below explain each row and link the sources.

Scores quoted below from Cyrus Shepard's review are his own ratings of the evidence out of 10, not statistical measures.

Strong evidence: AI tools have to reach your page

If an AI tool can't fetch and read your page, nothing else on this list matters. This is the one factor the platforms state outright.

• Google: to appear as a link in AI Overviews or AI Mode, a page must be indexed and eligible to show with a snippet. Google says there are "no additional technical requirements".
• ChatGPT: OpenAI says sites that opt out of OAI-SearchBot "will not be shown in ChatGPT search answers". Blocking GPTBot, its training crawler, is a separate setting and doesn't remove you from search.
• JavaScript: in Glenn Gabe's October 2026 tests, ChatGPT, Claude and Perplexity couldn't render JavaScript in any test. Gemini could. Text that only appears after JavaScript runs is invisible to three of the main AI tools.

Shepard's review of 54 studies puts access first, at 9.5 out of 10, and snippet controls such as nosnippet fourth, at 9.2.

Strong evidence: you rank in search, for more than one query

AI answers are built on search results, so search rankings are one of the best predictors of being cited. The twist is that the main keyword isn't enough.

Ahrefs studied 863,000 keyword results and about 4 million AI Overview citations (published March 2026). Only 37.9% of cited pages ranked in the top 10 for that keyword, counting ads and other search features as positions. Another 31.2% ranked between 11 and 100, and 31.0% ranked lower or not at all.

Ahrefs's 2025 study had found 76%, but its method has changed since, so the two figures aren't directly comparable. Its explanation for the gap is fan-out: Google cites pages that rank for the related searches it runs, as well as for the question the user typed.

ChatGPT shows the same pattern. In Ahrefs's study of 1.4 million ChatGPT prompts, pages found through its web search were cited 88% of the time. Reddit threads it pulled in were cited less than 2% of the time.

Shepard scores search rank at 9.4 and fan-out rank at 9.3, second and third on his list.

Strong evidence: other sites talk about you

Mentions of your brand on other sites are the off-page factor with the strongest link to AI visibility. Backlinks matter less than most SEO checklists suggest.

In Ahrefs's study of 75,000 brands, web mentions had a correlation of 0.66 with AI Overview visibility. Backlinks had 0.22. A correlation of 1 would mean the two move in lockstep, so 0.66 is a moderately strong link and 0.22 a weak one. Neither proves cause.

A Semrush study of 1,000 domains, run with Kevin Indig, points the same way. Nofollow links (0.34) were as closely tied to AI mentions as followed links (0.33). That suggests being referenced counts for more than passing link value.

There's a catch for self-promotion. Lily Ray checked 100 B2B "best [category]" queries in AI Overviews. When a brand's own "best tools" list was cited, the brand was left out of the recommendation 69% of the time. The AI used the page as a source and recommended the better-known competitors on it.

Moderate evidence: the page answers the question plainly

Once a page is found, the clearer and more specific its answer, the more likely it is to be used. You control this factor today, unlike rankings and mentions.

• Relevant titles: in Ahrefs's ChatGPT study, cited pages had titles closer in meaning to the prompt than pages that weren't cited (similarity 0.60 against 0.48). Pages with plain-language URL slugs were cited 89.8% of the time, against 81.1% without.
• Answer first: Shepard scores query-answer match at 9.2 and an answer near the top of the page at 8.8.
• Specific and sourced: specific facts score 8.3, citing sources 8.0 and passages that make sense on their own 8.0. The expert survey ranks "citable, specific facts" fourth.

This doesn't mean cutting pages into tiny chunks for machines. Google says there's "no requirement to break your content into tiny pieces". It means writing the way a busy reader wants: the answer under the heading, then the detail, with numbers and named sources.

Mixed evidence: freshness

Recently updated pages are cited more often, but relevance still beats age. The evidence depends on the platform and the type of question.

Seer Interactive looked at 47,097 citations across ChatGPT, Gemini and Perplexity (July 2026). 75% of cited pages had been updated within the last year: 78% on Gemini, 73% on ChatGPT and 65% on Perplexity. The data came from four brands, so it may not hold for every industry.

Ahrefs's ChatGPT study adds a counterweight. Among pages found through search, the typical cited page was about 500 days old, and relevance mattered more than age. Only for news did younger pages win when relevance was similar.

Weak or no evidence: schema, length, authority and llms.txt

These are the factors most often sold as AI fixes. The evidence for them is thin.

• Structured data (schema): Shepard finds a small, consistent positive link (5.6 out of 10), and experts rate it low (+0.80 on a scale of −3 to +3). Google says there's "no special schema.org markup you need to add". Experts in Shepard's survey made one exception: product data, which feeds Google Shopping.
• Domain authority: 5.0 out of 10 in Shepard's review, often a weak relationship. It overlaps with mentions, which are the stronger signal.
• Length: 6.7, with inconsistent evidence. Longer pages sometimes do better, but long pages can also have less of their text used.
• llms.txt: 2.0 out of 10, the lowest score, with no credible evidence that it affects citations. Google says it doesn't use these files. I explain why in llms.txt: what it is and whether you need one.

What changes if your buyers are in Southeast Asia

The same factors apply, but language and location change which sources get picked. Most of the studies above use US and English-language data, so treat them as a starting point for this region.

Shepard lists language as a factor (6.3 out of 10): AI tools lean towards sources in the language, and sometimes the location, of the question. Ahrefs's fan-out study found that "2026" appeared in 22.9% of its English fan-out queries in September 2026 and "price" in 8.9%. Country names such as India, Australia and the Philippines appeared more often over the same period.

My reading, which the studies don't test directly: ChatGPT often searches with a year, a price or a country attached. A page that names the market and gives current local prices matches those searches better than a generic global page. And if your buyers ask in Thai, an English-only site is competing for a search it may not match.

Where to start

Work through the factors in order of evidence, not hype. Think of access as the entry fee: without it, you're not in the game.

1. Check access. Allow OAI-SearchBot and other AI search crawlers, make sure key pages can show snippets, and keep the main text in the HTML.
2. See where you stand. Ask AI tools the questions your buyers ask and note who gets named. My AI visibility audit guide shows how.
3. Rank for the question cluster. Find the related questions behind your main keyword and make sure a page answers each one.
4. Rewrite your key pages to answer first. Specific numbers, dates and sources, with each section making sense on its own.
5. Earn mentions. Reviews, press, partners and industry lists. This is slow, so start early.

For the step-by-step version of steps 2 to 5, read how to get your brand cited in AI answers. If you'd rather hand it over, this is what my AI search optimisation service covers.

FAQs

About the author Guy Sanohlam, Founder

Guy founded Wincon Marketing in Bangkok after nearly ten years in SEO, content and paid media, both agency-side and in-house. He moved early into AEO as AI assistants started answering buyers' questions, and now helps brands get found on Google and in AI answers.

Connect on LinkedIn →