Answer Radar
The Content Gap View
A recommendation gap says an engine recommended someone else for a question. The content gap view asks the next question: which of your existing pages was supposed to answer it, and what does that page not say that the answers did. It scores each tracked page against a requirement set derived from the admissible answers, names what is missing, and stops there.
Quick Answer
For each measured question, the engines' answers are reduced to a requirement set: phrases at least two different engines used, plus the products their judged mentions named. Each tracked page is scored on which requirements it contains. A question no page covers is flagged as a content gap and the closest page is named. It reports what is missing and writes no advice.
- Agreement is counted in engines, not samples, so one engine repeating itself does not create a requirement.
- The question's own words, publications, features and your own brand are excluded.
- Coverage is binary per requirement: the page says it or it does not.
How a requirement is derived
The admissible samples for one prompt are read for two kinds of requirement. Terms come from the answer text: a phrase counts when at least two different providers used it, because one engine sampled five times repeats its own phrasing and distinct providers agreeing is the better question. Entities come from the judged mentions: the products the answers named. The two are reported separately and counted together only as a plain total, because they carry different version stamps and a weighted blend would be one number with two incomparable versions.
Three exclusions
The prompt's own words are excluded: every answer to 'best CRM for a small business' contains 'crm' and 'small business', and a page that restates the question has covered nothing. Publications and features are excluded: half the entity names a judge returns are cited sources rather than products, and requiring a page to cover a review site's domain is nonsense. Your own brand is excluded: requiring your page to mention you is free coverage, and free coverage inflates the score of exactly the pages that need work.
How a page is scored
Each tracked page's stored text has its URLs stripped, then each requirement is matched as a whole word. Coverage is binary per requirement. Counting occurrences or weighting a heading above a footer would be a judgement about prominence, and that judgement is left to you, once the view has told you the page says the words at all.
Three refusals mirror the metric layer. A page whose text could not be read is unusable with the reason, never coverage zero. A prompt whose requirement set came out empty is unusable for the same reason. A rate with an empty denominator is absent from the payload rather than 0.0.
Using the view
Filter by url to see one page against every question, by prompt to see every page against one question, by only_gaps to see the questions no page covers, or by a coverage band to find the pages that are close. Scope to one run or a period. Evidence mode returns the per-requirement verdicts so you can see exactly which phrase or product a page misses.
The view names the closest page for a question nobody covers, which is usually the page to extend rather than a new one to write. What to write is your call, or your assistant's; the tool reports what is missing and writes no advice about it.
- Start with only_gaps: questions with no covering page.
- Then the low coverage band: pages that are almost there.
- Then draft a fix against the gap, which pulls the same evidence.
FAQ
Which pages are tracked?
Pages stored in the project's evidence store: pages fetched as cited sources during audits, and pages you fetch deliberately with answer_sources_fetch. A page must be stored before it can be scored.
Why is a competitor's product a requirement for my page?
Because the engines named it when answering the question, and a page that never mentions the alternatives buyers are weighing reads as incomplete to an engine composing a comparison. Whether to name a competitor on your page is your editorial decision; the view only reports that the answers did.
Does coverage mean the page will be cited?
No. Coverage is an observation about text, not a prediction. The methodology page lists what no measurement here can tell you, and why an engine chose a source is on that list.
Related help pages
Answer Radar
Drafting a Fix From the Evidence
How a fix goes from evidence skeleton to approved artifact: draft, write, check, save, review. What the faithfulness guard rejects and who can approve.
Answer Radar
The Query Fan-Out View
What query fan-out is, how Answer Radar records the searches each engine issued while answering your prompts, how it is grouped, and the two limits it reports.
Answer Radar
Reading the Visibility Metrics
What presence rate, recommendation rate, share of voice, sentiment mix and citation share mean, which samples count, and why a segment can report no number.
Answer Radar
The Action Queue
What the Actions tab holds: content gap and competitor won cards derived from measured runs, their evidence and impact counts, the five statuses, and why no.