The AI Visibility Index

One document a month, written rather than generated. It tells you how often 6 AI surfaces named your brand, how that compares with 3 competitors you chose, which specific questions you win and lose, and which page or publication the models actually drew on.

What is an AI visibility index?

Each cycle produces 1,800 observations per market: 100 prompts, 3 runs each, across 6 AI surfaces. Three figures come out of that and are never averaged together. In our own measurement 57.8% of ChatGPT runs never searched the live web at all, which is reported as a separate line because a brand invisible to a model that did not search has a different problem from one invisible to a model that did.

What the AI Visibility Index report contains

• /Three figures per surface — retrieval, citation and share — rather than one blended score, because a single percentage hides whether the model even went looking.
• /6 AI surfaces reported apart: ChatGPT, Gemini, Perplexity, Copilot, Google AI Overviews and AI Mode. Overviews and AI Mode are counted separately despite both being Google.
• /Three named competitors on the identical prompt set, so the comparison is like for like, not each brand judged on questions that flatter it.
• /Prompt-level detail: which questions you win, which you never appear in, and which flipped this cycle.
• /Source tracing — the URL a model drew on. Roughly 68% of AI citations point at third-party sources, so this is usually where the work turns out to be.
• /A written read from Robert on what moved, what did not, and which of the two needs a change of approach.

How to read it without fooling yourself

The most common misreading is treating a small monthly change as a result. Model output varies between identical runs, so the report labels movements that sit inside the variance band as exactly that, instead of presenting them as progress. It occasionally makes for a dull report and always for an honest one.

The second is reading the three figures as one. A brand can be flat on share while its retrieval rate doubles, which is real progress that a headline number would erase. Read them in order: did the model look, did it name anyone, did it name you.

The third is comparing surfaces to each other. They disagree by design — citations concentrate on a small pool of outlets and different engines pick different ones — so a strong Perplexity figure tells you very little about ChatGPT.

What a good score looks like

There is no published industry benchmark for this and anyone quoting one has invented it. What we can offer is shape: in most regulated markets a handful of brands hold the majority of mentions, a long tail appears occasionally, and a large group never appears at all.

That last group is bigger than operators expect. One test across three queries found 55% of live casino providers invisible to AI entirely — not ranked poorly, absent. If your first reading puts you there, it is a common starting point, not an unusual failure.

So you are graded against two things only: the 3 competitors you named, on identical prompts, and your own previous months. Those are currently the only comparisons that mean anything.

Turning the report into work

Source tracing is the part that makes it actionable. If 3 competitors appear in Ontario prompts because one comparison site keeps being quoted, the work is getting accurate information onto that site — third-party presence, not another page on your own domain.

Where the decisive source is a thin page you own, the answer is entity and content work instead. Knowing which of the two you are dealing with is what the report buys you, and it is the thing guesswork gets wrong most expensively.

How to get it

What it is not

• /An AI ranking. No fixed position exists and answers vary between identical runs.
• /A traffic forecast. AI platforms sent 1.13 billion referrals in June 2025 against roughly 191 billion from Google in the same month.
• /A substitute for rank tracking. It sits alongside conventional rankings, because the two now move independently.
• /A shortcut. Models favour sources already heavily cited, so a new brand climbs slowly and the report shows the climb instead of shortening it.

One structural note. Our figures come out lower than several dashboards on the market, and the reason is in the method: we include every run in the denominator, including the ones where the model never searched at all. Colder numbers, closer to reality.

A note on how the report is delivered, because clients ask before they ask about the method. It arrives as a written document with the figures in tables and the raw export attached and not as a dashboard login. That is deliberate: sections of it get forwarded internally to people who will never have an account, and a portal makes that unnecessarily difficult for no benefit to anybody except the vendor.

One more note on comparison, because it is the question that follows the first report. There is no published industry benchmark for AI visibility in iGaming, and any vendor quoting one has invented it. You are graded against the 3 competitors you named on identical prompts, and against your own previous months. Those are currently the only two comparisons that carry information, and we would rather say so than manufacture a percentile.

A closing note on what this replaces, never adds. Most operators already pay for rank tracking, and the Index is not a second version of it. Rankings tell you where your URLs sit; the Index tells you whether a model names your brand when somebody asks a question in natural language. Those two numbers used to move together and now do not, which is the entire reason for running both. If you can only run one, run the rank tracker — it is cheaper and better understood. If you can run two, this is the one that explains why the first one stopped predicting revenue.

One question that comes up in almost every first conversation: what happens when the models change? They will, and often. Retrieval behaviour shifts with model versions, citation patterns move when a platform changes its sourcing, and a surface that mattered last quarter can shrink. When that happens we say so in the report and explain what it means for your series instead of silently rebasing the numbers. A measurement that survives a model update only because nobody mentioned the update is not a measurement worth paying for, and this field has produced a great deal of that already.

What the Index will not do for you

It will not move the number by itself. Measurement is diagnosis and not treatment, and a report that nobody acts on is an expensive subscription. It will not give you an industry benchmark, because none exists and anyone quoting one has invented it. And it will not shorten the climb — models favour sources already heavily cited, so a new brand rises slowly and the Index shows the climb and not accelerating it.

One more thing worth setting out before you subscribe to anything. The Index is most useful to brands that already rank reasonably well and cannot understand why the commercial results no longer follow. If you are not being indexed, or your content has no named author, those problems come first and measuring them more precisely will not help. We will tell you that after the first reading, never sell you twelve months of monitoring on top of an unfixed foundation.

Where the Index stops being useful

The Index is one measurement, taken carefully. It is not a strategy, and there are four things it does not settle.

• /It does not rank you against the market. It measures you against the three competitors you named, on the prompts we agreed. Change either input and the number changes with it.
• /It does not survive a model update unchanged. When a surface shifts how it retrieves, the series breaks and we say so on the report instead of quietly rebasing the history.
• /It does not tell you what to fix. It tells you where you are absent and which sources are being cited instead. Turning that into work is a separate engagement.
• /It is not a substitute for rank tracking. Roughly a third of AI citations also hold a top-ten position, so the two measurements answer different questions and you need both.

We publish the prompt set with every report. If the method cannot be inspected, the number is just a claim with a decimal point.

The complete method sits on its own page: how the prompt set is built, how runs are sampled, what counts as a mention, and where the measurement breaks. It is published so the figures can be argued with, or replicated in-house if you would rather not buy them.

Related on this site

• /What we do — Nine iGaming SEO services for casino, sportsbook and affiliate brands. Technical, content,.
• /Who we work with — for operators, affiliates, B2B suppliers, crypto casinos and land-based venues. Six.
• /gambling SEO compliance — Our gambling SEO compliance policy. Licensed operators only, verified at onboarding, prohibited.
• /Glossary of terms — an glossary of 51 terms across AI search, technical SEO, links, commercial models.

Sources for the AI-surface figures here: the 57.8% ChatGPT non-retrieval measurement is our own, taken across 1,800 observations; the AI Overview citation overlap comes from Semrush and BrightEdge, early 2026; and the 68% third-party citation share is drawn from Seer Interactive analysis published in 2026.

The Index reports retrieval separately for a reason set out in this analysis: a brand invisible to a model that did not search has a different problem from one invisible to a model that did.

Why the prompt set is written per market rather than once in English: unlicensed recommendation rates vary from fourteen to eighteen out of roughly twenty prompts between countries.

Common questions