AI content ranking

What does AI content ranking look like on one question?

Take one question we track for attensira.com, "ChatGPT rank tracker", and look at which domains each engine leaned on between 5 September and 4 October 2026. We read the deciding sources over the Attensira MCP on 4 October 2026.

• ChatGPT

  Answers

  12

  Citations attached

  248

  Most-cited domain

  keyword.com, 33 (13.3%)

• Perplexity

  Answers

  12

  Citations attached

  116

  Most-cited domain

  geoptie.com, 15 (12.9%)

Two engines, one question, two different winners. Of each engine's six most-cited domains, only rankscale.ai appears on both lists. Neither engine let one domain dominate: the top source took about an eighth of the citations on each side. attensira.com was cited zero times on either engine, and Attensira was named in none of the 24 answers.

That is AI content ranking in practice: a selection made per question and per engine, invisible to the reader, that decides which handful of sources the answer is built from. The answer at the top of this page is one of those 24, and its five inline links are the result of that selection.

How does an AI system rank content?

In a retrieval pipeline, four orderings happen one after another, and each can drop you.

1. Candidate generation. A search or vector lookup produces the pool. A page that is not in the pool cannot be ranked, however good it is. Crawler access decides this stage; see crawling and indexing.
2. Relevance scoring. Each chunk is scored against the question. Vector similarity handles paraphrase well and exact identifiers badly, which is why many systems add a lexical score such as BM25.
3. Budget selection. Only some passages fit into the model's context. A reranking step often decides which. Anthropic documents that a simple factual question typically uses one to three searches while comparative research can use ten or more, so the size of the pool varies with the question.
4. Citation choice. The model writes, then attributes. A passage can shape an answer without being cited, and the order of the attached links is its own number; see citation position.

Treating these as one "AI ranking" is the most common mistake in the field. The fixes differ at each stage, and only the first two respond directly to what you publish.

Is AI ranking different from Google ranking?

For eligibility, no. Google's AI features documentation says there are no additional requirements to appear in AI Overviews or AI Mode, and that a supporting link must be indexed and eligible to show with a snippet. It also notes these features may use query fan-out, issuing several related searches across subtopics to build one response.

For selection, yes. A results page orders pages for a query. An answer picks passages for a question and may pull them from searches the person never typed. High classic rank mostly helps by making you a candidate. It does not guarantee your passage is the one chosen.

What makes a page lose at each stage?

• Not fetchable. Blocked to the engine's search agent, rendered only in the browser, or behind a login. It never enters the pool.
• The answer is spread out. A page whose answer runs across four sections loses to a page that answers in one paragraph. The retrieved chunk does not carry the rest of your article with it.
• The qualifier lives elsewhere. A caveat two paragraphs above a claim does not travel with the claim.
• Snippets are blocked. Google lists nosnippet, data-nosnippet, max-snippet and noindex as controls that limit what is shown from a page, including in AI features. A page that forbids a snippet has opted out of being quoted.
• The engine prefers a different kind of source. For "ChatGPT rank tracker", both engines leaned on the trackers' own product pages and on roundups, not on general SEO blogs. No amount of polish on a page of the wrong kind changes that preference; being present on the sources it prefers does.

What do people get wrong about AI content ranking?

"We rank first on Google, so we will be in the AI answer." Ranking makes you a candidate. Being selected as evidence and being the answer are different outcomes, and they come apart in both directions. On "ChatGPT rank tracker", Rankscale was named in 18 of 24 answers, yet rankscale.ai drew only 14 of ChatGPT's 248 citations: the engine named the brand mostly on the strength of other sites' pages.

"There must be a trick that gets you ranked." No engine publishes one. Google says no special optimization exists. The levers are access, a passage that answers the exact question, and presence on the sources each engine already prefers.

"One ranking for all AI engines." The table above shows the opposite: one shared domain in two top-six lists. Each engine has to be read on its own.

How does Attensira work with AI content ranking?

For every question you track, Attensira records which domains and pages decided the answer on each engine, whether you appear on them, and what the next move is: fix your own page, or get onto the sources the engine already trusts. The crew then does that work for you, with your approval, and measures the result per engine with the sample size. Citation tracking shows the same per-engine source report for your own questions.