A mention is not a ranking
Ranking asks which page is best overall. Citation asks a narrower question: for this sentence I am about to write, which source states it cleanly enough that quoting it is safe?
That difference explains an outcome businesses find baffling — a comprehensive, well-ranked guide gets ignored while a short FAQ answer from a smaller site gets quoted. The guide buries the claim inside three paragraphs of context. The FAQ states it in two sentences that survive extraction.
Wallat, Heuss, de Rijke and Anand (2025) set out the underlying constraint: a system that cites sources is expected to ground its claims in those sources faithfully. Faithfulness is easier to achieve against a passage that means the same thing on its own as it does in context. Anything requiring the surrounding paragraphs to be understood correctly is a risk, and gets skipped.
The format effect, measured
Iyappan (2026) measured citation rates by content shape. The spread is wide enough to be a strategy on its own:
| Content shape | AI citation rate |
|---|---|
| Long-form contextual content | 92% |
| Entity-rich content | 89% |
| Structured-data-heavy pages | 85% |
| FAQ-formatted content | 67% |
| Keyword-focused thin content | 41% |
The bottom row is the one worth staring at. Pages written to hit a keyword density target cite at 41% — less than half the rate of content written to carry meaning. Two decades of SEO habit produce exactly the shape that generative engines cite least.
Kargaev (2026) adds the sentence-level version of the same finding: statistics present in content score NIS 0.747 and citations within content 0.671, against 0.043 for content length. What gets quoted is not more words. It is a checkable claim with a number and a source attached.
What a quotable passage looks like
Four properties, all testable on your own pages.
It stands alone. Read the paragraph with everything around it removed. If "this" and "as we saw above" leave it meaningless, it cannot be lifted.
It is specific enough to be wrong. "We improve visibility" cannot be verified or falsified, so quoting it adds nothing. "Lead times of four to six weeks for orders under 500 units" can be checked, and is therefore worth attributing.
It carries its own evidence. A number, a date, a named source. This is the NIS 0.747 finding in practice: a claim that brings its evidence is safer to repeat than one that asserts.
It names the entity. "We" is useless in an extracted quote. A passage that says who is making the claim survives the trip into an answer with the attribution intact.
Why your own site is the weaker half
You control the shape of your passages, which is real leverage. You do not control the thing that decides whether the model trusts the claim enough to attribute it: whether anyone else says something compatible.
de Oliveira (2026) describes the mechanism as an authority loop — stronger semantic associations between an entity and a topic raise the probability of future citation, which in turn strengthens the association. The loop has to be started from outside. A trade publication, a sector association, a named client, a review platform: each is an independent statement the model can weigh against yours.
This is why digital PR reads as an AI visibility tactic rather than only a link-building one. The link raises classic authority; the sentence about you raises citability.
Rewriting a page for citation
Take one page that already gets impressions and no clicks.
- Put the direct answer in the first 60 words, as a plain paragraph, before any context.
- Turn every "this improves results" into a statement with a number, a timeframe or a named condition.
- Replace "we" with the company name in the claims that matter.
- Add a real FAQ — the questions buyers actually ask, answered in two or three sentences each, not padded.
- Check every paragraph in isolation. Fix the ones that collapse.
Leave the length alone. Length scores 0.043; nothing is gained by padding, and the long-form advantage in the table above comes from contextual depth, not word count.
Frequently asked questions
How do I get mentioned in ChatGPT?
Write passages that can be lifted without distortion: self-contained, specific, carrying their own evidence, and naming the entity. Then get independent sources to state something compatible, because attribution follows corroboration.
What is the difference between being mentioned and being recommended?
A mention is your name appearing in an answer, including as background or as one of several sources. A recommendation puts you forward as the choice. The selection mechanics behind the second are in getting recommended by ChatGPT.
Does adding an FAQ section help?
Measurably. FAQ-formatted content cites at 67% against 41% for keyword-focused content (Iyappan, 2026). The gain comes from the shape — short, self-contained answers — so an FAQ padded into essays loses the advantage. That the rich result itself is gone does not change this: Google retired FAQ snippets, but the format still earns citations. The same mechanics across every engine, not only ChatGPT, are in getting mentioned in AI search.
Do statistics really make a difference?
They are the second-strongest signal in Kargaev's (2026) hierarchy at NIS 0.747, behind only entity mentions. A claim arriving with a number and a source is safer to repeat than one that asserts.
Why does ChatGPT quote my competitor's shorter page?
Because citation is decided per passage. A shorter page whose claims stand alone beats a longer page whose claims need their context to make sense.
Can I stop ChatGPT mentioning my brand incorrectly?
Not directly, but an incorrect description is usually assembled from thin or contradictory sources. Making the correct version consistent across the places the model finds is the available remedy.
Do mentions in ChatGPT bring traffic?
Less volume than search, better quality — AI-referred traffic has been measured converting at 14.2%. Many mentions produce no click at all, which is why measuring citation rather than sessions is the honest metric.
How do I track whether I am being mentioned?
Run a fixed set of buyer questions monthly with web search enabled and record whether you appear and how you are described. The fixed list matters more than the tooling; changing the questions resets the trend.
If you want the passage-level check done across your site, with the pages that already get impressions prioritised, the free scan is where to start.
