L7 · AI-answer visibility, the prompt-basket lever for AI citations

Ask an assistant your category's main buying question today. Ask it again next Tuesday. Two answers, two different lists of sources underneath. Nothing tells you whether the gap was you or a competitor.

L7 · AI-answer visibility is the lever people expect to be the whole method. It is one of eight kinds of work. It runs seventh. It sits behind a gate. Vendors sell this one as the entire product, and running it seventh is why its numbers survive a hard question.

It is primary on R3 · Referenced, the third question: are you the source answers are built from? It is secondary on R4 · Represented, the fourth question: does AI describe you the way you would describe yourself, in your own positioning? R4 is only partially instrumented today, and that caveat travels with it.

The entry criteria are blunt. A tracker configured. A prompt basket built, meaning the set of questions we replay in AI engines to measure you. And the AI-bot check passed, because you cannot be cited if you were never crawled. A citation means an AI answer naming one of your pages as a source.

The prompt basket is the strategy

Everything this work produces inherits from one decision. Which prompts do you measure? Get that wrong and every number downstream is decorative. Quattr builds the set from your first-party demand rather than an afternoon of brainstorming, then replays it on a schedule.

Four sources feed it. Each earns its seat. Your search queries and intents ground the category questions. Your log files and AI referrals surface the questions assistants already send you. Your sales conversations contribute the words buyers use first.

The named anti-pattern is the hand-picked list. It drifts toward the brand's own name and flatters whoever wrote it. Where answers carry no citations, the measurement falls back to Google. The prompt baskets article covers the recipe.

We sample the set and size it to the tracker tier rather than being exhaustive. A stable sample you can replay beats a heroic one-time crawl of everything, and a set that changes shape between replays produces a trend line about itself.

See also: why hand-picked prompt lists lie →

Replay turns screenshots into metrics

A one-off check of what ChatGPT says today is a screenshot. Interesting, unrepeatable, expired by Friday. The same basket replayed on a schedule is a trend line with a denominator underneath it, and that denominator is the only thing separating an anecdote you can tell in a meeting from a number your team can be judged on for a year.

Cut by engine before you blend anything. ChatGPT, Perplexity, Gemini and AI Mode pick their sources differently, and an aggregate hides the trades. AI Mode is Google's conversational search surface, separate from the AI Overview, the AI-written block Google puts on top of a classic results page.

A flat blended line usually means one engine gained while another lost. Treat the blended view as a summary and keep the per-engine cuts visible.

Trend reads inherit the reporting rule. Separate the market's growth from your share of it before you credit the work. These surfaces are still rolling out, and adoption moves the raw counts on its own.

Prompt gaps become briefs

The output is a worklist rather than a wall of charts. Prompts where you never appear go to the ai-prompt-coverage-gaps skill, a packaged analysis you run by name in your assistant. It shows who is cited there and which pages do the answering, and that becomes the brief for net-new content.

You protect and extend the pages that already earn citations. Then you study the prompts a competitor owns. What does their cited page do that yours does not? That question is answerable.

The Represented connection rides the same replays at no extra cost. The answers that cite you also describe you, so description drift surfaces in your basket before it surfaces anywhere else.

The workflow that does this: Prompts that cite us →

Crawlability gates every conclusion

Every conclusion here assumes the bots could fetch you. A blocked crawler produces the same zero as a content gap. Only your logs tell them apart. That is why R1 · Reachable runs first, the first question: can crawlers and AI bots fetch, render, and index your pages?

The check settles arguments that would otherwise run a month. The replays say invisible. The logs say never fetched. The fix is a line in robots.txt, and two checks together stop a team writing its way out of a technical problem.

Easy to state, hard to sell. So the method documents it once instead of renegotiating it every engagement. Run seventh, the citation number arrives with its excuses already eliminated.

The article on not being cited if you were never crawled covers that join. On the cross-map this is where the AI-answer work concentrates, and it still refuses to be first.

See also: why you cannot be cited if you were never crawled →