How to read a visibility score report
A section-by-section guide to a monthly AI visibility score report: method, headline, prompt movement, competitors, citations, crawler access and AI-referred traffic, plus the questions worth asking.
A monthly visibility report should let you answer three things in about ten minutes: did the score move, why, and what happens next month. If it takes longer, or if you finish reading and still cannot say why, the report is not doing its job.
This is how we suggest reading ours, section by section. The same approach works for any AI visibility report, including one you build yourself.
Start with the method box, not the headline
Before the number, look for a short block that says how it was measured. It should state the size of the prompt set, any prompts added or retired this month, the engines covered, the markets, roughly how many checks each prompt got, and what counts as being named.
If that block is missing, you cannot compare this month with last month. A score that rose because ten easy prompts were added is not the same as a score that rose on an unchanged set. In our reports, a changed prompt set always comes with a bridge: the score on the old set and the new set for the month the change landed.
The headline: per engine first
The headline should show each engine's score next to last month and next to the baseline. A blended score can sit underneath, with its weighting stated in plain words.
Read the per-engine numbers before the blend. A blend can rise while one engine falls, and the engine that fell may be the one your buyers use most. Then look at direction over at least two periods. One month up is a hint. Two months up on the same engine, concentrated on prompts you worked on, is a signal.
Prompt movement: the most useful page
Percentages hide which questions changed. A good report has a page that sorts every prompt into four groups:
- Gained: named this month, not named last month
- Lost: named last month, not named this month
- Holding: named in both
- Never named: not named in either
As an invented example, a report might show 6 prompts gained and 2 lost on a 50-prompt set. That is a net gain of 4 prompts, which tells you more than "up 3 points". Then read the names. If the 2 lost prompts are your highest-intent comparison questions, a net gain is less comforting than it looks.
The never-named group is the work plan. Each prompt in it should have a reason attached by the time the next report arrives: a missing page, a crawler problem, a third-party source you are absent from.
Competitors: who took what
The competitor section shows share of voice against the agreed competitor set. The useful question here is specific: for each prompt you lost, who was named instead? If one competitor keeps appearing on your lost prompts, their cited pages are worth reading closely.
Also check that the competitor set has not quietly changed. A new name in that list changes share of voice without changing a single answer.
Citations: where the answers came from
This section shows which pages and domains the engines cited for your prompts. Look for three things.
Your own citation rate, and which of your pages get cited. If the same two pages carry everything, the rest of the site is not pulling its weight.
The third-party domains cited on your never-named prompts. These are the review sites, roundups, communities and videos where your absence is costing you. A domain that appears across many of those prompts is a priority.
Changes in cited pages. A competitor page that suddenly starts getting cited across your prompts is often the explanation for a lost prompt.
Crawler access: can the engines read you
This part is easy to skip and should not be. It shows AI crawler requests to your important pages and the status codes they received. Two things to check: whether pages shipped last month have been fetched yet, and whether error rates on key pages changed. A page that no AI crawler has requested cannot be the reason the score moved, in either direction.
Traffic and conversions
The last numbers are AI-referred visits and what those visitors did. Read them as a floor, not a total. Attribution only covers visits that arrive with a recognisable referral, and some people read an answer and then search for you directly.
What you are looking for is the chain holding together: prompts gained, pages cited, visitors arriving on those pages, some of them converting. When that chain is visible for even a few prompts, it is the best evidence the work is commercially useful.
The work log and next month
A report should end with what was shipped, mapped to the prompts it targeted, and what is planned next, mapped the same way. If a piece of work cannot be tied to a prompt in the set, ask why it was done.
Questions worth asking
Whether you are reading our report or someone else's, these questions surface problems quickly:
- Did the prompt set, competitor set or engine list change this month?
- Which prompts moved because of work we did, and which moved on their own?
- Did the untouched prompts move by the same amount as the worked ones?
- Has everything we shipped last month been crawled yet?
- What is the stated reason for each never-named prompt?
Red flags
A report with only a blended number. Screenshots of individual answers used as evidence. A prompt set that grows every month. Brand-name prompts counted inside the headline score. A proprietary "visibility index" with no stated formula. Any of these makes the number harder to trust, even when it is going up.
How we build our reports in Promptwatch
Every report we send is built from Promptwatch data. Prompt tracking and prompt trends supply the per-engine scores and the gained, lost and never-named groups. Competitive benchmarking and share of voice come from the competitor set we configure. Citation analytics gives page-level and domain-level citations, including Reddit and YouTube, and citation trends show which sources gained or lost ground over the month. Agent Analytics supplies the crawler section, with crawl errors and the crawl-to-citation path for new pages. Visitor analytics, installed with a small script or a GTM template, covers AI-referred visits and conversions.
On agency plans, custom reports and the Looker Studio connector let us put all of that into one document per client. Unified Actions, the AI-generated to-do list, is a useful cross-check for the next-month plan, though we still decide the priorities ourselves.