Lost AI citations: queries you used to rank on and dropped
AI answers aren't static. A citation you held last month can vanish without warning — and you'll never notice from a single snapshot. Here's how to catch the losses.
A lost AI citation is a query you used to be cited or mentioned on that you no longer appear on. AI answers are volatile — models update, retrieved sources shift, a competitor publishes something fresher — so slots you held can disappear without any change on your side. The catch is that you cannot detect a loss from a single report: a snapshot only shows where you are now, never what dropped off. Detecting losses requires comparing two points in time — today’s coverage against a recorded baseline — so a query that was present before and is absent now is flagged as lost. That is the difference between a free one-off Domain Check and an ongoing monitored project: the project re-scans on a cadence, keeps the history, and tells you the moment a citation is gained, lost or regained, at the level of the specific query and model.
Why do AI citations disappear in the first place?
Unlike a static index, AI answers are regenerated. ChatGPT, Gemini and Grok update their retrieval and ranking, the live sources they ground on change, and competitors keep publishing. Any of those can knock you out of an answer you held, with no edit to your own site. This is the same volatility that makes share of model drift — each model changes on its own schedule.
The consequence is that the queries you win are not a fixed asset. They’re a position you hold only as long as the underlying signals keep winning, which means a slot can erode quietly.
Why can’t a single report show me what I lost?
A reverse AI search report is a snapshot: it lists where your domain appears right now. A row that dropped off simply isn’t there — there’s no “ghost” of a citation that used to exist. To see a loss you need two snapshots and a comparison between them. Without a recorded baseline there is nothing to diff against, so losses are invisible by construction. This is the single biggest reason a one-off check, however useful, can’t do the whole job.
How does loss detection actually work?
It’s a set comparison at two grains — the domain and the specific cited URL. The system records which queries and URLs each model cites you on, re-scans on a cadence, and diffs the new set against the stored one. Every row lands in one of a few states:
| State | Meaning | What to do |
|---|---|---|
| Held | Cited last scan and this scan — the slot is stable. | Defend it: keep the source page fresh and accurate. |
| Gained | Not cited last scan, cited now — a new win. | Note what changed so you can repeat it elsewhere. |
| Lost | Cited last scan, absent now — a citation dropped. | Investigate: who replaced you, and on which model. |
| Regained | Lost earlier, cited again now — fluctuation, not a permanent drop. | Confirm stability before treating it as recovered. |
What do I do when I find a lost citation?
Start at the query and model grain — a loss on one of three models is a different problem from a loss on all three. Then look at who the model cites now instead of you. If a rival took the slot, that’s a direct lead into why ChatGPT recommends your competitor and not you. Prioritise losses by the intent of the query — a dropped pricing or alternatives citation is more urgent than a dropped definitional one.
Why does this require ongoing monitoring?
Because the baseline has to come from somewhere. A monitored project re-scans the queries that matter to you on a schedule, keeps the history, and surfaces the gained/lost/regained events as they happen — so a dropped citation becomes an alert, not something you discover months later when a deal goes cold. The free check is the snapshot; the project is the time series.
Start tracking before you lose more
Run the free Domain Check to capture today’s coverage across ChatGPT, Gemini and Grok — that becomes your first baseline. Turning it into a monitored project is how you catch the next loss the day it happens.