"Ranking factors" is one of the most searched phrases in SEO, and one of the most padded. Lists of 200 factors mix things Google has confirmed with things someone once correlated in a spreadsheet. You do not need the list. Google documents its ranking systems publicly, and once you know what each one rewards, most of the folklore falls away.
This post goes through the systems Google names, what they reward in practice, the factors people still chase that Google says it does not use, how AI answers choose sources differently, and a short plan for raising rankings and traffic.
What does Google say about how it ranks pages?
Google's ranking systems guide says its systems "look at many factors and signals about hundreds of billions of web pages and other content" to present the most relevant, useful results. It then names the systems it uses. They group into five jobs.
Understanding the query and the page
BERT, MUM, neural matching and RankBrain are language systems. They work out what a query means, including words it does not contain, and match it to pages that answer that meaning. Passage ranking lets Google find the relevant section inside a long page.
What it rewards: pages that answer the question in clear language, with the specific thing near the top, rather than pages that repeat the exact keyword.
Judging quality and reliability
Reliable information systems promote authoritative, high-quality sources, especially on topics where accuracy matters. Original content systems elevate the primary source over pages that only cite it. The reviews system rewards in-depth, first-hand reviews over thin summaries. Google's guidance on helpful, people-first content describes the same idea from the writer's side.
What it rewards: first-hand experience, original data and evidence, a clear author, and sources for claims.
Weighing other sites' judgement
Link analysis systems and PageRank use links between pages to understand what pages are about and which might be most helpful.
What it rewards: links from relevant, independent sites that chose to link because the page is worth it.
Keeping results fresh and varied
Freshness systems favour recent content for queries where recency matters, such as news or anything with a year in it. Site diversity limits how many results come from one domain, and deduplication hides near-identical results.
What it rewards: keeping time-sensitive pages current, and one strong page per question rather than several overlapping ones.
Removing what should not rank
Spam detection systems and removal-based demotion systems demote pages and sites that break Google's spam policies or attract many valid legal removal requests. The exact match domain system stops a domain name that matches the query from getting undue credit on its own.
What it rewards: not cutting corners. Most ranking drops people blame on "an update" are one of these doing its job.
Which popular ranking factors are myths?
- The keywords meta tag. Google says plainly that Search does not use it.
- Keyword density. Language systems understand meaning; a target percentage makes pages worse to read and does nothing else.
- Word count. Long pages rank when length serves the question. A 3,000-word answer to a two-sentence question is padding.
- Exact-match domains. Google has a named system to stop them getting credit for the match alone.
- Social shares as a direct factor. Shares can bring links and attention, which matter; the share count itself is not a documented signal.
Why does ranking first matter so much?
Because most clicks go to the top few results, and the top is now earning fewer of them. Ahrefs' study of 300,000 keywords, published in February 2026 from aggregated Search Console data, found the first result's average desktop click-through rate on informational keywords fell from 7.6% in December 2023 to 3.9% in December 2025, and to 1.6% where an AI Overview appeared. Ranking first still matters, but on question-style queries it no longer guarantees the visit, which is why being named in the answer above the results counts as well.
How do AI answers choose sources differently?
Google says there are no additional requirements or special optimizations needed to appear in AI Overviews or AI Mode: a page must be indexed and eligible to show with a snippet. But the selection is different. AI features use what Google calls query fan-out, issuing related searches across subtopics, and pick passages that answer each part. Ahrefs found that in March 2026 only 37.9% of URLs cited in AI Overviews also ranked in the top 10 for the same query; the rest came from positions 11 to 100 and beyond.
Two things follow:
- Quotable passages matter more. A page that answers a sub-question in one clear, self-contained paragraph can be cited without ranking first for the head query.
- Third-party pages matter more. For "best" and "vs" questions, answers lean on roundups, reviews and threads written by others. Being named there counts even when your own page is not cited.
Our own site shows how literal relevance still is. We track the question "GEO tool that opens pull requests", which describes what our product does. In the month to 4 October 2026, ChatGPT cited one of our pages in 4 of 12 answers, yet neither engine named us in any of 24. The first deciding source on both engines was the pull-requests page of an open-source GEO research repository on GitHub: a page that matches the words "GEO" and "pull requests" exactly and answers a different question. Relevance is decided on the words before quality is weighed, so a page that never states plainly "this tool opens pull requests" loses to one that merely contains the phrase.
The table at the top of this page shows the other force, reliability, on five how-to questions about ranking and crawling that we track, read on ChatGPT from 5 September to 4 October 2026 (189 answers). On "how to rank in ChatGPT for B2B SaaS" and "how to get recommended by AI", the most-cited source was OpenAI's own help centre (72 of 321 and 392 of 1,976 citations). On "how to track AI search visibility" and the two questions about AI crawlers, it was Google's developer documentation (322 of 2,366, 679 of 1,104 and 608 of 1,507). On a question about how a platform ranks or reads pages, the engine goes to the platform's primary source first, which is what Google's original-content and reliable-information systems reward too. Literal word matching can still win over all of it: asked for "Profound alternatives for small businesses", ChatGPT often read "profound" as an adjective, and over the 30 days to 4 October 2026 its most-cited source was a dictionary, with 204 of 1,516 citations across 60 answers.