Analyze results

Citation gap report

Which prompts cite everyone except you: queries you rank for in Google or Bing that AI engines never cite you on, and the reverse. Found under Reports → Opportunities.

Open in ChatGPT Open in Claude

Definition

Citation-ranking gap

The mismatch between where your site ranks in organic search and where AI engines cite it as a source, reported in both directions.

Where to find it: Reports → Opportunities → Citation gap

What it measures#

The report card asks: “Which prompts cite everyone except us?” It joins two data sets that normally live apart: your organic rankings (from Search Console or Bing Webmaster Tools) and the citation rate of your domain on your tracked prompts. The result comes on two tabs:

Invisible winners. Queries you rank at or near the top of that the assistants never cite you for. You win the search result and lose the answer.
Inverse gap. Your own URLs that the assistants do cite, but that rank poorly or not at all. Engines trust the content; search engines do not rank it.

Rank and citation are near-independent

A top-three ranking does not mean an engine will cite you, and many cited pages rank outside the top ten. The report treats the gap as a finding in itself. It does not claim that closing one side moves the other.

How prompts meet keywords#

A tracked prompt and a search query rarely use the same words, so each prompt is paired with organic queries by one of three rules, and the rule is shown on every row:

exact. The prompt text itself is the keyword.
fan-out. One of the prompt’s fan-out queries is the keyword.
semantic. The page proposed as the prompt’s owner by prompt coverage is the page that ranks for the keyword.

Rankings are read over the last 90 days of organic data to decide which page a keyword ranks on.

How the citation rate is calculated#

Citation rate

runs where your domain was cited ÷ runs of the matched prompt, per engine

A rate is published only after 20 runs of a prompt on an engine. Below that, the row is marked Too few runs and hidden unless you tick Include low-confidence rows. “Cited in 0 of 2 runs” is not a finding.
Every published rate carries a 95% interval (Wilson score), so you see how wide the claim is. At 0 citations in 20 runs the upper bound is roughly 16%.

Worked example#

You rank at average position 2.1 for “helpdesk for shopify” with 1,400 impressions. The matched prompt ran 30 times on ChatGPT and your domain was cited once.

1 ÷ 30 = 3.3%: under the default 5% ceiling, above the 20-run floor. The row appears as an invisible winner.

Controls and defaults#

Tab	Control	Default	Meaning
Invisible winners	Max position	3	Weighted average position over the window, at or better than this
	Min impressions	100	Enough searches for the position to mean anything
	Max citation rate	5%	“Zero or near zero”
Inverse gap	Min position	20	Past page two, where rank stops earning traffic; URLs with no organic data are always included
	Min citations	2	Cited more than once, so it is a pattern

Both tabs take a date range (last 7 days up to last 16 months, or custom) and an optional comparison period that adds Citation-rate change and Citation change columns. Export CSV downloads the rows.

Reading a row#

Open an invisible-winner row to see:

Matched prompts and the match type for each.
By assistant. Runs, rate and 95% interval per engine.
Cited instead of you. Up to five hosts the engines cite for those prompts.
AI-readiness inputs. Web Vitals, agent accessibility and llms.txt findings for your ranking page, from the Audit section.

What to do with it#

Invisible winners: compare your page with whoever is cited. The difference is usually structure: a direct answer at the top, named entities, a date and a quotable fact. Use Compare a URL on the prompt detail page.
Check readiness first. A page engines cannot fetch, that loads slowly, or that llms.txt steers them away from will not be cited however well it ranks. See domain audit and AI accessibility.
Inverse gap: do ranking work on pages engines already trust. On-page and internal-link work has something to build on there.
Re-read after a few runs. A citation rate over one or two runs is noise. Track the prompts that matter and come back once they pass the 20-run floor.

Needs organic data

Without Search Console or Bing Webmaster Tools connected, the report is empty. Connect one under Data Sources: see Search Console and Bing.

Search and traffic reports The rest of the Reports directory. Why isn't my page cited? Checklist for a page that ranks but is never a source. Prompt coverage Which sub-questions your ranking page fails to answer. Sources Who is cited instead of you, across every prompt.