What actually gets a page cited by an AI answer
Generative and answer engines now sit between a question and the pages that answer it. This is a reference on how they select, cite and attribute sources: what the platforms document, what independent work has measured, and which widely repeated tactics have nothing behind them at all.
Every page leads with an evidence verdict
One judgement, stated before any prose: how much evidence stands behind the claim the page is about. It is the thing a reader came for, and it is the reason this reference is worth linking to.
| Verdict | What it means | Entity pages |
|---|---|---|
| Documented | The platform documents this, or independent replication supports it. | 3 of 24 |
| Observed | Consistent practitioner observation, not confirmed by the platform, and it may change without notice. | 6 of 24 |
| Unsupported | Widely repeated, with no evidence behind it. | 15 of 24 |
Six pages that set the argument
The three best-documented surfaces, and three techniques that are sold constantly and supported by nothing.
Surfaces, and what publishers do about them
AI search surfaces
The products themselves — AI Overviews, AI Mode, ChatGPT search, Perplexity, Gemini, Copilot and the smaller assistants. What each one retrieves from, what it discloses about source selection, and what it reports back to publishers.
Techniques and signals
What a publisher does, or is told to do, to be cited: schema markup, answer-shaped writing, chunking, entity grounding, crawler directives, tracking and attribution — each held to the evidence rather than to the consensus.
The guides
Twelve guides, grouped under three pillar guides: what AI search is, how to write and build for it, and how to measure it without fooling yourself.
A reference, not a pitch
Nothing here is for sale and no tool is ranked. Where a widely sold tactic has no evidence behind it, the page says so and explains why people believe it anyway.