What "ranking" means in AI search
AI search engines do not rank ten blue links. They choose a handful of sources to read and cite, and the factors that decide which pages make the cut fall into three groups: retrieval (can the engine find your page for the searches it runs), authority (is your page safe to repeat), and extractability (can it lift a clean answer from your text).
Classic search ranks whole pages for a query. AI engines such as ChatGPT, Perplexity, Gemini, Claude, and Google AI Overviews search, read, and then write an answer, attributing each claim to the passage it came from. Your page is competing for a citation inside that answer, not for a position on a results page. For a step-by-step walk through that process, see how AI assistants choose which sources to cite. This post is the factor list: what each one checks and what you can change.
The ranking factors at a glance
The table below lists the factors that consistently decide whether a page gets cited, grouped by the stage where they act. No engine publishes its weights, so the groups and order reflect how the pipeline works rather than measured coefficients.
| Factor | Group | What it checks | What you control |
|---|---|---|---|
| Crawl access | Retrieval | Whether AI and search crawlers may fetch the page | robots.txt rules, firewall and bot settings |
| Indexing | Retrieval | Whether the page is in the index the engine searches | Bing and Google indexing, sitemaps |
| Server-rendered text | Retrieval | Whether the content exists without running JavaScript | Rendering setup for key content |
| Semantic relevance | Retrieval | How closely the meaning matches the searches the engine runs | Topic coverage, follow-up questions, natural wording |
| Information density | Authority | Specific, checkable facts per paragraph | Numbers, dates, names, original data |
| Corroboration | Authority | Whether claims agree with or cite reliable sources | Links to primary sources |
| Entity clarity | Authority | Who published and who wrote the page | Organization and Person details, named authors |
| Mentions elsewhere | Authority | Whether other sites associate you with the topic | Reviews, directories, press, community answers |
| Freshness | Authority | How current the page is, for time-sensitive topics | Real updates and honest update dates |
| Answer-first structure | Extractability | Whether each section opens with the answer | Section openings, question headings |
| Lists and tables | Extractability | Whether steps and comparisons are explicit | Formatting of enumerable content |
Read the table top to bottom: a page that fails a retrieval factor never reaches the authority and extractability checks.
Retrieval factors: can the engine find your page?
Retrieval factors decide whether your page enters the shortlist at all, and they are the most common reason good pages are never cited.
Crawl access and indexing
Each engine retrieves from a search index or its own crawl. ChatGPT search draws on third-party search indexes, with Bing the most commonly cited partner. Gemini and Google AI Overviews use Google's index. Perplexity runs its own crawler. If your robots.txt or firewall blocks the relevant crawler, or the page is not indexed where the engine looks, it cannot be retrieved.
Server-rendered text
Many crawlers read the HTML your server sends and do not run JavaScript. Text that only appears after scripts run, such as content inside client-side tabs or lazy-loaded widgets, may be invisible to them. Keep the content you want cited in the initial HTML.
Semantic relevance
Engines match meaning, not exact strings. They usually rewrite a conversational question into one or more shorter searches, then look for passages whose meaning is closest. Pages that cover a topic fully, including its common follow-up questions and the terms practitioners actually use, match more of those searches. Repeating a keyword does nothing here.
Classic rank still feeds retrieval
Because engines search before they answer, pages that rank well in a search index for the engine's searches are the first candidates. Classic SEO is not replaced. It is the entry ticket.
Content factors: density, depth, and consensus
Once retrieved, pages with specific, verifiable facts beat pages with general statements, because a model summarizing a topic needs something concrete to say and attribute.
- Information density. The ratio of unique, checkable facts to total words. Replace "many businesses see results" with what happened, to whom, over what period, measured how. Cut filler: models discard it anyway.
- Original information. First-hand data, a worked example, or a documented process gives the engine something no other page has. Commodity explanations compete with thousands of near-identical pages.
- Depth on the topic. Answering the question and its follow-ups in one place makes your page useful for more of the searches the engine runs.
- Consensus and corroboration. Models favor claims that agree with other reliable sources. When your view differs from the consensus, link the evidence so the claim can be checked rather than dismissed.
Extractability factors: structure a model can lift
A relevant, trusted page still loses the citation if its answer is hard to extract, so structure is a ranking factor in its own right.
- Answer first. Open each section with a direct answer to the question in its heading, then support it.
- Question or statement headings. Headings that say what the section answers help the engine map the page. Avoid vague labels such as "Overview".
- Self-contained sections. Each section should make sense on its own, because engines extract passages, not whole pages.
- Explicit lists and tables. Steps belong in numbered lists and comparisons in labeled tables, each introduced by a sentence saying what it shows.
- Clean HTML hierarchy. One H1, H2s for main sections, H3s beneath them, with no skipped levels.
Freshness: when it matters and when it does not
Freshness is a strong factor for time-sensitive topics and a weak one for stable topics. Prices, software features, regulations, and "best tools" lists go stale quickly, and engines lean toward recently updated sources for them. A definition of a stable concept does not need monthly edits.
When you update a page, change the content meaningfully and update the visible date and the dateModified value in your Article markup together. Changing the date without changing the page is easy to detect and undermines trust.
What changed from classic SEO, and what did not
The foundations of classic SEO still apply; what changed is the unit of competition, the definition of a win, and how the query is formed.
What still matters: crawlability, indexing, relevance to the searcher's intent, content quality, and links. Google states that there are no special requirements to appear in AI Overviews beyond being indexed and eligible to show a snippet in Search, so its AI features guidance reads much like its general guidance.
What changed:
- Passages compete, not pages. One strong section can earn a citation for a page that ranks poorly overall, and one buried answer can lose it for a page that ranks well.
- The win is a citation or a mention, not a click. Being named in the answer has value even when the user does not visit.
- One question becomes several searches. Engines fan a question out into related searches, so coverage of follow-up questions matters more.
- Results vary run to run. The same question can produce different sources on different days, so visibility is a rate you measure over time, not a fixed position.
For a full plan that turns these factors into work on your site, see the step-by-step guide to showing up in AI search.
What is not a ranking factor
Several popular tactics have no evidence behind them and are not worth your time.
- Keyword repetition. Semantic matching makes keyword density irrelevant, and stuffed text reads as low quality.
- Word count by itself. Length helps only when it adds answers to real questions. Padding lowers information density.
- An llms.txt file. Google says Search ignores it, and no major AI engine has confirmed using it to choose citations. It can still help tools that read it, but it does not rank pages.
- Small speed gains. A page must load reliably enough for crawlers to fetch it. Beyond that point, shaving milliseconds does not change which source an engine cites.
Frequently asked questions
What is the most important AI search ranking factor?
Being retrievable. If the engine cannot crawl, index, and read your page for the searches it runs, no other factor applies. After that, a direct answer in clear, self-contained sections and a clearly identified, credible source matter most.
Do backlinks still matter for AI search?
Yes, indirectly. Backlinks help crawlers discover your pages and support the classic rankings that feed retrieval. They carry less direct weight in the final citation choice than relevance, clarity, and trust in the passage itself.
Are AI search ranking factors the same on every engine?
The three groups apply everywhere, but each engine retrieves from a different index. Google AI Overviews and Gemini depend on Google's index, ChatGPT search relies heavily on third-party indexes such as Bing, and Perplexity uses its own crawler, so the same page can rank differently across them.
Is page speed an AI ranking factor?
Only up to a threshold. Very slow responses can cause crawlers to time out and skip the page. Once the page loads reliably, faster load times do not make an engine more likely to cite it.
How quickly do changes affect AI citations?
It depends on the engine. Engines that search live can use an updated page as soon as their index has the new version, while others refresh more slowly. Re-check the same questions over several weeks rather than judging from a single answer.
