AEO Content Coverage

Overview

AEO Content Coverage is the gtmcron answer-engine-optimization loop. It has two parts. The top of the page is AI search: the questions your buyers ask before they choose, what an answer engine says when they ask, and the fix for each question you lose. The bottom is coverage scoring: paste any list of questions and see which ones your published content answers, biggest gaps first.

What it is for, on the page. Under the headline, this page says what it produces, what makes it run, and where the output lands. Every Work page reads that line from one registry, so the three answers cannot drift from the page.

Cited by answer engines

Once a month, in the first week, the answer-engine check asks your buyer questions of Claude Haiku 4.5 with web search, one search per question. For each answer it stores which domains were cited, whether the answer named your company, whether a source sat on your own domain, and which of the competitors on your brief the answer named. One more question, "what is your company?", is compared with your brief, your last website scan and your Brand & Company name, and any difference in name, offer or category is listed.

The page shows the newest stored month, lost questions first: the AI result, the competitors named instead, Google figures beside it, why it was lost and the fix, and a Rerun button. Against the month before it lists the mentions gained and lost. Questions are compared only when they were asked in both months, so a changed question set does not read as a lost mention. A month checked before competitors were recorded says so on each row rather than showing none.

The results describe one engine, which the page names. ChatGPT and Perplexity may cite differently, and neither offers an API we can call on your behalf, so we do not report figures for them.

If the check skipped a month (no confirmed brief, no Anthropic key, a spend cap reached), a banner says which and the newest month that did run is shown. If it has never run for your account, the page says so instead of showing zeros. Each run is metered like any other AI call and is counted against your caps.

Your questions

The check asks your question set, which you edit under Your questions. Until you change it, the set holds up to 10 questions drawn from your confirmed brief: your category, the problem you solve, and comparisons with your competitors. You can add the questions your buyers really ask, pin the ones that matter most, edit the wording, and retire the ones that do not apply. The next monthly check asks exactly the active questions.

Your edits always win. A question you retire is never added back, even when your brief would produce it again, and a question you rewrote is not added back in its old wording. Pinned questions are listed first. You can keep up to 25 active questions. The derived questions only top the set up to its limit (10 unless you change it), so your monthly cost does not rise unless you add questions or raise the limit. The page shows the estimated cost of the next check at your current size. Under Your other domains, list any domain that is yours besides the website on your brief (a docs site, a second brand), so a citation there counts as your own.

Why a question was lost

Every question where the answer neither named nor cited you gets one cause, checked in this order:

• Not indexed. A page of yours matches the question, but Search Console shows no impressions for it in the last 28 days or does not list it. Fix the indexing before you rewrite anything. This cause needs Search Console connected; without it, the page says indexing was not measured and no question gets this cause.
• Missing page. Nothing you publish answers the question. Draft a page in Content Studio.
• Weak page. A published page covers the question, but the engine cited something else. The row names the matched page so you can check the match. Improve that page rather than writing a new one.
• Third-party sources. The fallback when your content could not be scored: the answer relied on other sites. The source map shows which.

The diagnosis is deterministic. It reuses the same term-overlap coverage as the scoring section below, against your published content today, and the stored Search Console read. It can match the wrong page; when it does, log a fix and tick "The diagnosis was wrong".

Indexing

A question is diagnosed as not indexed when the page that matches it has had no impressions in Google Search Console in the 28-day window, or Search Console does not list the page at all. Open the page in Search Console's URL inspection, request indexing, and check that it is not blocked by robots.txt or a noindex tag. Then rerun the question after Google has crawled the page. Search Console connects under Settings, Connect; it is a preview connection.

Source map

Every domain the engine cited in the month sits in one of four classes, with how many questions cited it, so you work on the sources that shape the most answers first:

• Own. Your website and the other domains you listed. Improve the page that is cited, or the one that should be.
• Claim. Listings, directories and profiles you can take over or update yourself.
• Earn. Editorial, review and community coverage you have to deserve: pitch, contribute or get reviewed.
• Contest. A source behind the engine's own description of you, when that description disagrees with your brief. Ask for a correction. Months checked before this was recorded have no contested sources.

The claim list is a short list of listing hosts kept as data, and it will age. If a domain is in the wrong class, move it with the control beside it. Your choice holds for every later check.

Google beside AI

When Search Console is connected, each question shows the clicks, impressions and average position over the last 28 days for the Google queries that match it. A query matches when most of its words appear in the question. The Google figures and the AI result sit side by side and are never combined into one score, so a question can gain Google impressions while it stays lost in AI search, and you see both. Without Search Console the column reads Not connected.

Rerun one question

After you fix a page, press Rerun on that question to ask it again now instead of waiting for next month. A rerun asks the same engine the same way as the monthly check, is metered, runs on your own Anthropic key when you saved one, and counts against your AI budget like the monthly run. You can rerun each question once a day and 30 times a month in total; the page shows how many are left and when the next one is allowed. A rerun is shown beside the monthly result and never changes the monthly figures or the month-over-month comparison.

Fix log

Log what you changed for a question ("rewrote the pricing page", with an optional link). The log pairs each fix with the question's result before it and the first result after it, from a rerun or the next monthly check, and labels it moved, unchanged or not rerun yet.

Signups from AI search

The tile counts signups whose first visit came from an answer engine. It reads the referrer we record at signup, so only the platform operator's own account sees a number there. Your signups happen on your site, not ours, so for your account the tile asks you to connect analytics; reading this count from your own analytics is not built yet. A failed read says so and is never shown as zero.

How coverage is scored

Coverage is deterministic: for each question, it finds your best-matching piece of content and measures the share of the question's significant terms that the content covers. The same questions and content always produce the same coverage, so the board is defensible and repeatable. A question scores as covered, thin, or a gap. Short, common words are ignored so coverage reflects the substantive terms a question and an answer share, not filler.

Where the answers come from

By default the score runs against the content you have published from Content Studio. Published only, because the question is what an answer engine can actually see: a draft you have approved but not published is not counted. If you have published nothing yet, or you want to test a specific set, paste answers directly: one per line, an optional title, a colon, then the answer text. Either way, a question with no matching content is shown honestly as a gap with no best match, never filled with a guessed answer. Turning a gap into a drafted answer stays with the content skills, which narrate and draft; the coverage number itself is this deterministic function.

If you run it with the Answers box empty and you have published nothing, the page says so and scores nothing. It does not report every question as a gap, because "you have no content" and "your content does not answer this" are different findings and only one of them is a reason to go and write something.

What a publication status means

Content Studio binds approval to the exact saved revision. An edit after approval sends the piece back for review. WordPress and Webflow first create a provider draft, then make that same object public. Notion creates a page that is visible inside the selected workspace, which does not mean the page is public. Mailchimp and Customer.io create provider drafts that remain unsent.

Each provider object keeps its stable ID. If a later step fails, Retry resumes that object and shows its current visibility instead of creating a duplicate or claiming that staged content was delivered.

What it does not do

The page does not publish content or crawl the live web, and it is not a live ranking. The AI search part reads what the monthly check and your reruns stored; the coverage board reads the questions you supply and the content you store or paste. The design mockup sketched a transcript-to-page publishing pipeline and a structure score; those are omitted because nothing here measures them, so showing them would mean inventing the numbers.