AI Visibility — is the customer in the core

The question this dimension answers

Who is recommended by AI in this category, and is the customer one of them?

There is a core of companies that AI answers name again and again, and the rest are named seldom. The core is stable — it changes seldom, and a limited number of questions is enough to detect it. Asking more only confirms it.

The customer cares about one thing: are they in the core. The verdict is one of three, in the product's own words. Use them verbatim:

• Core — recommended in at least a quarter of AI answers, even allowing for how few answers there are.
• Too early to tell — recommended sometimes, but too few answers so far to say whether that is often or rarely. Here they can fight.
• Rarely recommended — recommended in fewer than a tenth of AI answers, even allowing for how few answers there are. That is different work.

Never a fourth word, and never "tail" or "edge" to a reader.

Everything else this dimension produces — presence, ranges, per-engine splits, endorsement, the blended score — is the mechanism that decides membership, not the answer.

Before you start

Read references/reading-the-data.md. It is not optional here: this dimension has more ways to be read wrong than any other in the platform, and the two worst are leading with a score and ordering two brands whose ranges overlap.

Steps

1. Resolve the project. list_projects → get_project for the prompts and per-dimension freshness → list_competitors for the roster (the customer's own domain is in it, marked isOwn: true).

2. Read the market map. get_ai_visibility_dashboard. Go to summary.marketMap. It arrives compact: marketMap.brands is one page — the top rows, at least the whole core, plus the customer's own row and every tracked competitor's — and marketMap.brandsPage.total is every row on the map: the companies the models named, plus the customer's own row when no answer named it. A tracked competitor missing from the page was named in no answer; any other company missing from it is on another page (mapOffset) or was named in no answer. Reading several pages, de-duplicate by domain — the kept rows repeat. The response opens with readingGuide, the platform's rule for each field in it; read it first.

3. Check `summary.promptMarket` before you use the map.

Unless its state is rivals_named_in_most_answers, say the prompts may not describe this project's market, and do not lead with the map. A map built from questions that return registries rather than vendors is measuring the wrong thing, and reporting it as the market is the worst error available here. This check comes first, every time.

4. State membership.

"Nine companies make up this market as the AI models draw it. The customer is one of them, tied for 7th of 9 by how often it is named."

or

"Eight companies make up this market. The customer is not one of them — named in 1 of the 40 answers this check, against the leader's 14."

Find the customer's row by isOwn: true — it is always on the page. Use marketMap.coreSize, rankByPresence, and the presence figure with its range and its count. A null rankByPresence means no answer named the customer: say not named in any answer — never a place, never not measured.

The verdict word comes from that row's zone, and from nothing else: named_in_a_quarter_or_more_of_answers → Core · share_not_yet_separable → Too early to tell · named_in_under_a_tenth_of_answers → Rarely recommended.

While marketMap.tailIsProvable is false, say no brand can be ruled out of this market yet.

5. Read the per-engine split before calling anything core. A brand core to one model and a brand core to all five look identical on the pooled figure. Read perEngine. A brand named everywhere is a different finding from a brand one model likes.

6. Only now, the supporting reading.

• get_ai_visibility_trend — one row each for the customer, every tracked competitor and up to 3

untracked companies: a reading now, a reading at the window's start, and whether they are separable. now is the latest map, pooling its checksAnalysed checks — never the latest check alone. A digest, not a plot; detail: "series" adds up to 12 points per company if you genuinely need them. Either way, if the ranges overlap that is two readings, not a movement.

• get_ai_visibility_history → get_ai_visibility_check_detail for a specific past check;

get_ai_visibility_history with limit: 1 for the latest check alone.

• What the models actually said: get_ai_visibility_check_detail with includeAnswers: true and a

filter — provider= with promptIndex= for one model's answer to one question, brand= for one domain across every answer. An answers read there leaves the summary out unless includeSummary: true. Unfiltered it runs to hundreds of thousands of characters on a large market; the figures are in references/platform.md § Answer sizes.

Reporting rules specific to this dimension

Never lead with the AI Visibility Score. It is a blended figure and it answers a question nobody asked. A brand named once can read 2; a brand named often and placed low can read 1. If you mention it at all, mention it after membership, and say what it is: a 0–100 figure over the top five positions only.

Never report a position or an average position. The field does not exist, by design. Order is by presence — how often a brand is named — and nothing else.

Never order two brands whose ranges overlap. They are tied. Say so.

Check the sign on `mentionRateGap` — it is the customer minus the leader, so negative means behind. Reporting a trailing brand as leading is the most damaging mistake on this dimension.

Google AI Overviews names companies and ranks nothing. Any order you see there is CompetLab's order of first mention. Its rows carry no description, which is why marketMap.profileEngines lists four engines, not five.

An engine absent from a check was not asked. Older checks ran three or four engines. That is not a brand's absence from that engine.

Model prose is the model's. "Claude described them as…", never "they are…".

Output

Short. The answer is a membership statement, not a dashboard.

# AI Visibility — {Project}
*{n} answers on the {date} check; {N} pooled across {k} checks in the current window.*

## The verdict
{One sentence: how many companies make up this market, and whether the customer is Core, Too early to tell, or Rarely recommended.}

## The market as the models draw it
{coreSize companies make up the market; the models named n in all — brandsPage.total, less the customer's row if no answer named it.}
| Company | Named in | Presence | Monitored |
|---|---|---|---|
{rows from the page you read — presence with its range; ties marked as ties}

## Per engine
{Which engines name the customer and which never have. Name the ones that produced nothing.}

## What this means
{2–4 sentences. Membership first. What would have to change for it to move.}

## What we did not measure
{Engines not asked, checks that came back short, prompt-market caveat if it fired.}

What NOT to do

• Do not lead with a score, a mention rate, a position, or a citation count.
• Do not compute a percentage from a small answer set without its count and range beside it.
• Do not call a difference a rise or a fall when the intervals overlap.
• Do not pool engines, and do not draw them as slices of one whole.
• Do not report Google AI Overviews' order as a ranking Google gave.
• Do not present model prose as fact.
• Do not tell the user what to do about the sources behind the answers — that is

competlab-ai-sources, and it has its own limits.

Decision questions

End with 3–5 questions whose answers would change the recommendation. Tie them to what the data actually showed — not generic clarifications. State both branches where you can:

"If the goal is entering the core, the work is off-property and slow. If it is holding ground against one specific rival, it is narrower and faster. Which is it?"