How to read your AI visibility score
The AEO overview shows four tiles, two cards and a line of provenance underneath. This article says what each number counts, what it is divided by, and when two runs can be compared. If you have no run on the screen yet, start with See whether AI assistants recommend you.
The four tiles
Each tile carries the count it rests on, and shows "Not measured" rather than a zero when nothing came back. A zero would mean the engines answered and left you out. "Not measured" means they did not answer.
| Tile | What it counts | What it is divided by |
|---|---|---|
| AI visibility | Answers that name you, frozen when the run finished. | The answers that came back for prompts that do not name your brand. |
| Prompt coverage | Prompts where at least one engine named you. | The scored prompts the run asked — not the ones that happened to answer. |
| Sentiment | How favourably the passages naming you read, on a −100 to +100 scale. | A sample of at most 32 excerpts spread across engines. A number appears only from three rated answers up, always with its sample and the model that read it. |
| Open recommendations | Recommendations not started or in progress. | Needs AEO Advanced. Below it the tile is headed Recommendations and reads "Enterprise plan" in place of a count, and clicking through offers the 7-day trial. |
A visibility figure is never recomputed from surviving data, so deleting a prompt later cannot restate a past run's score.
Share of voice, and by engine
Below the tiles are two cards.
- Share of voice — your mentions against each listed competitor across the run's scored answers, counted once per answer per name. The card shows the top eight rows and your own brand is one of them, so with all eight competitor slots used the lowest-scoring rival falls off the chart. With no competitors listed it shows your mention count and says there is nothing to compare against.
- By engine — mentions out of answers for every engine the run asked, with the model that answered. An amber note appears when the model differs from the previous run, because the trend then compares two different models.
Tip
Read By engine before you read the headline. One engine at zero and another at half pulls the average to a number neither engine produced, and the two need different work.
Where the number came from
The line under the tiles is the run's provenance: how long ago it was measured, its date and time, how many prompts it asked, how many were aided, and its health — "complete: 96 of 96 answers", or "partial" with the counts. The same health line is shown wherever the score is shown.
Once a run is more than 30 days old, a banner gives its date and age and says AI answers move quickly. It carries a Run again button.
What one run does and does not tell you
A single run tells you where you stood on one day. Run a second check before you draw any conclusion from it.
- A change since the last check is shown only when both runs asked the same scored prompts on the same engines and both got every answer back. When they differ, MarqIQ names which condition failed instead of printing a change.
- Adding or retiring a question breaks the first of those conditions, so keep the prompt set the same across the runs you want to compare.
- This is not a search ranking. There is no index to place you in and no query volume behind the figure. It is the share of the answers your prompts actually got that named you. A Google position is measured on something else, so the two cannot be compared.
Note
Prompts that name your own brand are labelled aided. They are asked on every run and their answers are shown, but no score counts them. If every active prompt names your brand, nothing is scored at all, and the overview says so and links to the Prompts tab.
If it did not work
A tile reads "Not measured" instead of a percentage
Why: No answer came back for that figure. A partial run with a silent engine is not quoted as your visibility anywhere, and a run with no answers at all is not either.
Fix: Read the amber banner above the tiles. Why one AI engine returned nothing, or half your prompts were not asked covers each reason and says whether to run again now or fix billing first.
Sentiment shows words rather than a number
Why: A reading needs at least three rated answers. Below that, and where the reader failed, ran out of time or declined, the tile says which in words rather than showing a figure that rests on one excerpt.
Fix: Nothing to fix on your side. Add more unaided prompts, or connect another engine, so a run produces more passages naming you.
No change is shown against the last check
Why: The two runs did not ask the same scored prompts on the same engines, or one of them did not get every answer back.
Fix: Read the line under the tiles — it names the condition that failed. To keep the trend, choose Keep the same prompts and run on the panel before each check.
The score did not move although an engine clearly failed
Why: The visibility figure is frozen when a run finishes and is never recomputed. You are reading the last finished run, not the one that failed.
Fix: Check the date on the line under the tiles. An unfinished or discarded run never replaces the figure above it.
Next steps
Related articles
Was this article helpful?