The Search↔AI gap report joins two datasets that normally live in separate tools: what Google Search Console says you rank for, and what AI engines actually said when asked your tracked prompts. Neither half produces this view alone.
It reports two disagreements and a quantity of uncertainty. The uncertainty is not a footnote — it is roughly half the value of the report, because the most tempting misreadings of this data are all confident conclusions built on missing evidence. This article covers what each list means, how the join works, and where the report stops short on purpose.
What the report is joining
The Google side comes from Search Console: queries, clicks, impressions and average position for your verified property. The AI side comes from your scans: each tracked prompt, how many engines returned a usable answer, and how many of those answers mentioned your brand.
Two details about the Google numbers matter before you read anything into them.
Average position is impression-weighted, not a plain mean across rows. A query with three impressions cannot pull the average as hard as one with thirty thousand. A plain mean is the single most common way to get this figure wrong, and it is the number a marketer screenshots and challenges.
The day window is computed in Pacific time, because Search Console reports in America/Los_Angeles. Using UTC boundaries shifts every figure by a day for part of each day.
A third detail is about freshness rather than correctness: Search Console data lags roughly two days, so the most recent day in the report is never today. The report derives that boundary from the rows themselves rather than guessing at the lag.
Search Console is available on every plan, including free, and connecting a property backfills history immediately — about sixteen months of daily property totals, and a shorter recent window of per-query rows, which are far bulkier and whose long tail Google withholds anyway. So the trend line is deep on day one while the query-level joins below start from the recent past and lengthen as syncs accumulate. The why rankings and citations disagree article covers the conceptual background if the phenomenon itself is new to you.
How a query is matched to a prompt
A Search Console query and a scanned prompt are different shapes of the same intent. Queries are terse (“crm startups”); prompts are questions (“What is the best CRM for startups?”). The join accounts for that with a containment rule.
Each string is reduced to significant tokens — lowercased, punctuation stripped, single characters and a small stopword list removed. A query matches a prompt when every significant token of the query appears in the prompt. Containment runs query→prompt, never the reverse: requiring the prompt’s tokens to appear in the terse query would match almost nothing.
Two consequences worth knowing:
- The stopword list is deliberately small. An aggressive list starts eating real product terms, and the containment rule is already strict.
- A query made entirely of stopwords matches nothing. With an empty token set, “every token appears” would be vacuously true and the query would match every prompt you track.
Each row in the report shows the prompts it matched, so you can judge whether the match is fair rather than taking it on faith.
Ranked but not cited
A query lands here when it clears the impression floor, sits at Google position 10 or better, matched at least one tracked prompt, and no engine mentioned your brand in any of those matched prompts. The row carries the query, its clicks, impressions and weighted position, the prompts it matched, and how many engine answers were involved.
The list is sorted by impressions, highest first — the largest audience finding you in Google and hearing nothing about you from an answer engine.
This is the most actionable output in the product, and the reason is that you already own the asset. You rank. Crawlability, authority and topical fit are established. What is usually missing is the structural work that makes a page quotable: a direct answer near the top, self-contained passages, unambiguous entity signals. Writing for AI citation is the practical follow-on, and citation intelligence shows which sources the engines named instead of you for those same prompts.
Cited but not ranking
The mirror list: a tracked prompt where at least one engine mentioned your brand, matched to a Search Console query that ranks worse than position 10. Sorted by Google position, deepest first — the widest disagreement between the two surfaces at the top.
This is usually a content gap rather than a ranking problem. Engines already associate your brand with the topic, and you have no page positioned to capture the search demand that exists alongside it. Publishing against those prompts is unusually cheap, because the authority work that normally gates a new page is already done.
Note the constraint in the definition: this list only includes prompts where a matching query exists and ranks poorly. That is not an accident, and it leads directly to the next section.
What the report refuses to conclude
Two categories of row are counted and shown as coverage, never asserted as findings.
Queries that matched no tracked prompt. These are not reported as “AI ignores you.” Nobody asked the engines about that query. The report has no evidence either way, and manufacturing a finding from an unasked question would be wrong in a way you would eventually catch.
AI-mentioned prompts that matched no query. These are not reported as “you don’t rank.” Search Console withholds queries issued by only a few dozen users over a two-to-three month period, so an absent query is frequently a privacy filter rather than a ranking fact. Google’s own deep dive into Search Console performance data filtering and limits documents this directly, including why summing the rows of the query table never reaches the chart total.
That same filter shows up as a separate figure: query coverage, the share of your property’s real clicks that named queries account for. It commonly undercounts by 18–50%. The report states the number rather than hiding it, because the discrepancy is Google’s privacy filter and cannot be engineered away — and because a dashboard that sums the query table and calls it “total clicks” is wrong in a way a customer can catch by opening Search Console in the next tab.
The general principle: a confident claim built on absent data is worse than a stated gap. The whole value of the Google side of this join is that it is checkable.
The URL-level view and when it is available
There is a sharper version of the same comparison that works on pages rather than questions: which URLs Google ranks for you against which URLs engines actually cited. Own-domain page-one URLs no engine cited, and cited own-domain URLs Search Console shows no ranking for.
It only works on grounded scans. When an engine answers ungrounded there is no retrieval step and no link — the analyzer extracts a domain from prose, and the stored citation carries no URL at all. With zero citation URLs to match against, “pages that rank but are not cited” would return every ranking page: a confident, wrong, and very plausible-looking finding.
So the derivation reports a urlsAvailable flag alongside the lists. When it is false, the empty lists mean no URL data, not no gaps found, and the interface says so rather than rendering an impressive-looking table of nothing. Read that flag before you read the table.
When URLs are available, matching normalizes both sides — scheme and www. dropped, query string and fragment discarded, trailing slash removed — because Search Console and an engine citation routinely differ on all of those while pointing at identical content. Cited URLs are also filtered to your own domain first: a competitor’s cited URL is not evidence about your pages.
A reading order that works
- Check query coverage first. If named queries account for a small share of your clicks, every conclusion below is drawn from a partial sample. That is fine, as long as you know it.
- Read ranked-but-not-cited top down. Highest impressions first. Open the matched prompts on each row and confirm the match is fair before acting on it.
- Read cited-but-not-ranking as a publishing queue, not a diagnosis. These are topics you have already earned the right to write about.
- Treat the two unmatched counts as a to-do for your prompt set, not as findings. A high unmatched-query count usually means your tracked prompts do not cover the intent your Google traffic already demonstrates — see choosing tracked prompts.
- Check
urlsAvailablebefore reading the page-level table at all.
This report sits alongside — not inside — your AI answer position data. Answer engine ranking means position within an AI answer; Google position means position in a results page. Merging the two into one “rank” number destroys both. Teams running this comparison across multiple clients will find the workflow notes in solutions for SEO professionals useful.
Frequently Asked Questions
Why do some of my Search Console queries show no gap analysis at all?
They matched none of your tracked prompts. The report counts these as unmatched queries and stops there, because it has no engine evidence about that intent — nobody asked. A large unmatched count is a signal to expand your prompt set, not a signal about your AI visibility.
Why does the page-level comparison show nothing?
Because none of the stored citations carried a URL, which is what happens when scans run ungrounded — the analyzer extracts a domain from prose rather than following a link. The derivation reports this explicitly rather than returning empty lists that read like “no gaps found.” Grounded scans produce real cited URLs and the page-level view becomes available.
Why don’t my query clicks add up to my total clicks?
Google withholds queries issued by only a few users to protect privacy, so those clicks appear in the property total but never as a row in the query table. The gap is typically 18–50%. The report shows it as query coverage rather than papering over it, because the two figures genuinely measure different things.
Is “cited but not ranking” bad news?
No — it is usually the best-value list in the report. It means engines already associate your brand with a topic where you have nothing published to capture the search demand. The authority work is done; what is missing is a page. That is a far shorter project than earning trust on a topic from zero.