Reading Your AI Visibility

What Is on the AI Visibility Screen

Open AI Visibility from the menu down the left of your dashboard. The screen is five blocks, in this order down the page:

1. Four numbers across the top. One week’s answers, counted four different ways.
2. AI Mentions. A chart of the share of answers that named you, week by week.
3. Question by Question. A table of every question against every assistant, which is where the detail is.
4. Who Got Named Instead. The other businesses those answers named.
5. Your Questions. The list of questions itself, where you add and remove them — covered in Writing the Questions We Ask AI Assistants.

The rest of this article takes the first four in turn. If you have just subscribed and the screen is mostly empty, nothing has gone wrong: nothing is asked until you have written a question, and the first answers arrive on your site’s own asking day.

How the Questions Are Asked

ChatGPT, Claude, Gemini, and Perplexity are asked directly, through the connection each company provides. For Google's AI Overviews — the AI answer Google puts above its own results — we run each of your questions as a Google search and read the answer it shows. That is five surfaces in total, asked every week.

We ask from outside any account, from the place you set, so your own results may differ — an assistant you have talked to before answers you differently from how it answers a stranger. Every answer is kept in full, and clicking any cell shows you what was actually said rather than our summary of it.

What “Mentioned in 3 of 12” Counts

The number on the right is answers we got back, not questions we sent.

Assistants refuse, time out, and occasionally change what they will answer. A question nothing answered this week is not a question you failed — so it is left out of the count entirely, and the screen says how many those were. A rate built on questions sent would quietly punish you for somebody else’s outage.

The Four Numbers at the Top

Four small boxes sit across the top of the screen, the figure above its label. All four count one week — the same week the table below is showing — and each counts it a different way:

Mentioned in answers

How many of the answers we got back named your business, out of how many came back.

Share of answers

The same thing as a percentage, which is the figure worth watching over months.

Linked to you

Answers that did not only name you but pointed at your website. A link is worth more than a mention, because it can be clicked.

Competitors named

How many other businesses came up across the week's answers.

None of these carries a week-over-week change, deliberately. A chip reading “down two since last week” would assert exactly what this screen’s own first rule denies — that one week’s reading is a verdict. Movement lives in the chart, over weeks.

Question by Question: Every Question Against Every Assistant

The panel headed Question by Question is a table: one row per question you have written, one column per assistant. Each cell says in words what that assistant did with that question, and there are three things it can say:

Mentioned

We got an answer and it named your business. The words · linked appear beside it when the answer pointed at your website as well as naming you.

Not mentioned

We got an answer and your business was not in it. That is a real answer rather than missing data.

Couldn't ask

We got nothing usable back from that assistant that week. Nothing is being claimed either way.

Clicking a cell with an answer behind it opens that answer in full. Clicking the question itself, in the left column, opens Week by week for that one question: every week we have asked it, assistant by assistant.

This is the most useful part of the screen, because it turns one rate into a diagnosis. Mentioned for “who does emergency callouts in Leeds” but Not mentioned for “what does a callout cost” tells you which page you are missing.

Looking Back at Earlier Weeks

The screen opens on this week, and there are two ways to look further back — one for a whole week, one for a single question.

The week picker, above the grid

Choose a date and the whole screen moves to that week — the four figures, the grid, and the list of competitors below it all change together. What you are reading is always one week, and the date says which.

A question's own history

Open any question in the left column and it shows every week we have asked it, assistant by assistant, so you can see whether you are being named more often or less.

Your site has one weekday of its own for asking, named on the AI Visibility screen, and it keeps that day so the weeks line up with each other. Those are the only dates you can pick, because they are the only days anything was asked.

Who Got Named Instead

The list below counts every other business named, across every answer that came back. It is who the assistants reach for when they answer your customers’ questions.

We can tell you who was named. We cannot tell you why — assistants do not publish their reasoning, and inventing one would be exactly the guessing this product exists to refuse. What you can do with the list is read their sites — and when one of them also comes up in the top results for a keyword you track, a competitor page analysis on the Competitors screen will read that page beside yours.

What This Measurement Cannot Tell You

• Whether an assistant recommends you to a real customer. We ask our questions from outside; a customer asks theirs from inside their own account, with their own history behind it.
• Why the answer changed. These products update without notice. We record which version answered, so at least a change of model is visible.
• How many people asked. Nobody publishes that. This is a measurement of what the answer says, not of how often it is asked for.

The full method, including what it cannot establish, is published on How It Works — the same page a stranger can read before buying anything.