Why do AI Mode citations differ from AI Overviews?

Query fan-out. AI Mode decomposes the prompt into many sub-searches, so it can cite sources that never rank for the prompt itself.

Do you measure follow-up turns?

No. We measure turn one only, because follow-ups change the citation set and destroy reproducibility.

Google AI Mode citations tracking

google_ai_modecitationsAnonymous

AI Mode cites inline and through a source panel, and because it fans one question out into as many as sixteen parallel searches, its citation set routinely contains domains that rank nowhere for the literal prompt.

How it renders

Inline linked phrases inside the answer plus a right-hand source panel.

Sampling

Citation sets are the least stable field we return. The same prompt run n times will produce overlapping but non-identical URL sets, which is why we report frequency across samples rather than a list from one run.

Gotcha

AI Mode and AI Overviews disagree on the same query more often than they agree. Reporting one as a proxy for the other is the single most common error we see in reporting decks.

What the API returns

A single collection call against Google AI Mode for the citations field, sampled n times and returned as one object per prompt.

{
  "engine": "google_ai_mode",
  "field": "citations",
  "prompt": "best ai visibility tracking tool",
  "country": "GB",
  "language": "en",
  "n": 30,
  "citations": [
    {
      "url": "https://nostimates.com/engines",
      "domain": "nostimates.com",
      "position": 1,
      "sample_frequency": 0.63
    },
    {
      "url": "https://example.org/guide",
      "domain": "example.org",
      "position": 3,
      "sample_frequency": 0.4
    }
  ]
}

GET /v1/collect?engine=google_ai_mode&field=citations

Response fields

Key	Type	Notes
url	string	Absolute URL as the engine rendered it, redirects resolved.
domain	string	Registrable domain, normalised for grouping.
position	integer	Order within the citation set for that sample.
sample_frequency	float	Share of the n samples in which this URL appeared.
anchor_text	string \| null	Visible anchor or chip label, when the engine renders one.
first_seen	date	First run in which this URL appeared for this prompt.

What it costs

1 credit per prompt per sample. Citations are included in the base collection — no surcharge.

See credit pricing →

Google AI Mode citations FAQ

Why do AI Mode citations differ from AI Overviews?: Query fan-out. AI Mode decomposes the prompt into many sub-searches, so it can cite sources that never rank for the prompt itself.
Do you measure follow-up turns?: No. We measure turn one only, because follow-ups change the citation set and destroy reproducibility.

Measure AI Mode as its own engine

If you're reporting AI Overviews only, you're missing a billion-user surface that disagrees with it. Let's scope both.

Other Google AI Mode data

Brand mentions Sources Answer text Share of voice Sentiment Prompt volatility Position Locale variance

Across every engine

Compare how citations behaves on every engine we track.

Citations across all engines →