How to read your reverse AI search report (field by field)
Every column in the report answers a different question. Read it as a worklist — what to defend, what to study, what to close — not as a single number.
A reverse AI search report is one row per query your domain is cited or mentioned on, with a handful of columns that each answer a different question. The query tells you what a buyer actually asked; intent tells you where in the funnel that question sits; models citing tells you which of ChatGPT, Gemini and Grok named you, since they routinely disagree; cited vs mentioned tells you whether you were linked as a source or just named in prose; competitors named tells you who the model treats as your substitute; and last seen tells you how fresh the row is. Read top to bottom, each column converts into an action — defend a slot, study a rival, close a gap, or investigate a drop. The number at the top is context; the rows are the work.
What is the report, exactly?
A reverse AI search report is the output of looking up your domain in the query–domain index backwards: instead of typing a question and seeing who gets named, you start from your domain and read out every question the models have cited or mentioned you on. The result is a table — one row per query — and each column carries a distinct piece of meaning. If you read it like a dashboard you’ll miss the point; read it like a worklist.
The full set of rows is your AI citation footprint. This page is about the columns inside it.
What does each field mean — and what should I do with it?
Here is every field, what it tells you, and the action it implies. Work the table column by column the first time, then row by row after that.
| Field | What it tells you | What to do with it |
|---|---|---|
| Query | The literal buyer question a model was asked (e.g. queries like “best CRM for small agencies”). | Read the phrasing to judge value; a sharp commercial question outranks a broad informational one. |
| Intent | Where the question sits in the funnel — discovery, best-of, alternatives or pricing. | Sort ready-to-buy intents to the top; that is where a citation is worth the most. |
| Models citing | Which of ChatGPT, Gemini and Grok actually named you on that query. | Treat single-model coverage as fragile; aim to widen it. See share of model. |
| Cited vs mentioned | Whether you were a linked source (citation) or named in prose (mention). | Protect citations as referral paths; work mentions toward becoming linked sources. |
| Competitors named | The other domains named in the same answer — the model's substitute set for you. | Study who you appear beside; this is your competitor query overlap to defend or attack. |
| Last seen | The most recent scan where you appeared on that query. | Re-check stale rows; a row that vanishes is a candidate lost citation to investigate. |
How do I prioritise the rows?
Two columns drive priority: intent and models citing. A pricing or alternatives query cited by all three models is a high-value asset you must defend. An informational query named by one model is lower stakes. Use the intent taxonomy to rank the value of each question, then use model coverage to judge how durable your position is.
What about the competitors column?
The domains named alongside you are the brands a model treats as interchangeable with you on that question. Run a rival’s domain through the same check and the rows where they appear and you don’t become your AI keyword gap. Where they out- rank you on a query you should own, dig into why ChatGPT recommends your competitor and not you.
Why does “last seen” matter so much?
AI answers are volatile — models update, sources shift, and a slot you hold today can flip next month. A single report is a snapshot in time. The last-seen date is your freshness check: old dates mean re-verify, and disappearances mean a possible lost citation. Tracking that over time is the job of a monitored project rather than a one-off check.
Where do I start?
Run the free Domain Check on your own domain to generate the report across ChatGPT, Gemini and Grok, then read it column by column using the table above. For the broader research context on how often these rows change, see our State of AI Citations work.