A good score and no AI citations: what the gap tells you

A strong AI visibility score paired with zero citations in AI answers is a common outcome, and it is not a contradiction: a score measures what your site technically allows, while a citation depends on your reputation and on the question being asked. These are two separate measurements, and the gap between them is what tells you what to do next.

A tool that melts both measurements into a single number takes away the useful part. Knowing that your AI visibility sits at 68 out of 100 does not tell you whether the blocker is a missing file you can fix in an afternoon or an absence of reputation that takes months of content and third-party mentions. Both situations look identical inside a blended score, and they call for completely different budgets.

Why does an AI visibility score not predict your citations?

Because a score measures conditions, while a citation is a choice an answer engine makes at the moment the question reaches it. In August 2025, Ahrefs compared the URLs cited by AI assistants against Google's rankings for the same 15,000 long-tail prompts: on average 12% of cited URLs appeared in Google's top 10, with 28.6% for Perplexity and 8.00% for ChatGPT's in-text citations.

The same company measured, in July 2025 across 1.9 million citations drawn from one million AI Overviews, that 76.10% of pages cited in Google AI Overviews came from Google's own top 10. The two findings do not contradict each other: AI Overviews draw on Google's index, and ChatGPT largely does not. One number cannot describe both what your site permits and what five different engines decide to do with it.

What does an AI visibility score actually measure?

An AI visibility scan checks 19 signals across three axes: whether an AI crawler can read your site, whether your content is structured so a passage can be lifted into an answer, and whether an AI agent can understand what you sell. The full list of 19 signals is published on the Biencité method page at /en/methode. The scan reads your home page and a handful of public files, read-only, with no account and nothing installed on your site.

The verdict attached to that score follows a cascade rather than an average: a site an AI cannot read is reported as invisible, whatever the rest of the score says. A signal whose file could not be reached is marked not measured and leaves the calculation instead of counting as zero, and the report states how many of the 19 signals were genuinely verified. Shopify, Wix, Webflow and Framer lock certain files at the domain root, so those signals are marked impossible on your platform instead of counting against you.

What the score leaves out matters just as much: it holds no measure of your reputation, your inbound links, or what AI engines already say about you. The weight of each signal is not published, because it shifts as answer engines change their habits. A score is meant to be compared with its own history, not with another company's number.

What does citation tracking measure, and on which questions?

Citation tracking puts your own questions to five generative engines once a week and records whether your site appears in the answers. The five surfaces queried are ChatGPT, Claude, Perplexity, Gemini and Mistral's Le Chat. The questions are the ones your customers actually ask, up to fifteen per site, and results read question by question and surface by surface, naming the sites that get cited when you do not.

The weekly cadence is not a technical limit, it is what makes the measurement readable: two answers from the same engine to the same question do not necessarily cite the same sources, so a single test is a draw rather than a measurement. Our article "Asking ChatGPT whether it knows you measures nothing" sets out that variability and what has to be repeated before a number means anything.

A good score and no citations: what should you conclude?

A good AI visibility score with no citations means the technical obstacles are cleared and what remains is editorial and reputational. Your site is readable, structured and declared, and the answer engines still choose to cite somebody else. At that point, adding one more file changes nothing, and that is exactly the conclusion a blended score would never hand you.

What this diagnosis calls for is content that answers the tracked questions, and mentions of your company somewhere other than your own website. Our article "How do you get cited by ChatGPT when you are a small business?" describes that work. It is counted in months rather than days, and no tool can promise a citation at the end of it: the scan measures eligibility, the tracker measures an observed probability.

A weak score but existing citations: is a fix still worth it?

A poorly scored site that already gets cited is running on reputation, and that is the most fragile of the four positions. Answer engines swap models and selection methods without notice, and a site that crawlers struggle to read holds no technical reserve to absorb that change. Fixes are still worth making, in a different order of urgency:

• Signals on the Visible axis come first, because an unreadable site loses its citations all at once rather than gradually.
• Signals on the Citable axis come next, since they decide whether an answer can quote a specific passage rather than drop your name alone.
• Signals on the Agent axis come last, unless you sell online and agents are starting to place orders with your competitors.
• Signals marked impossible on your platform are not debt, and they should never appear in your action plan.

Why keep two measurements instead of one number?

Because the two do not move at the same speed and are not fixed with the same budget. An AI visibility score changes the day a file is deployed on your site, and the next scan confirms it. A citation curve is built over several weeks, one point at a time, and it can fall without a single line of your site changing, simply because an engine switched models.

Blending them would produce an indicator that moves for incomparable reasons. It would rise because a file was added, drop because Google shipped an update, and you would never know which of the two events concerned you. The contrast in the Ahrefs data makes the point: AI Overviews pulled 76.10% of their citations from Google's top 10 in July 2025, while ChatGPT's in-text citations overlapped with that same top 10 only 8.00% of the time in the August 2025 study. Your technical groundwork is one thing. Which engine chooses you, on which question, is another.

When do neither of these measurements help you?

Measuring is pointless when there is nothing to measure, and that happens more often than a software vendor cares to admit. Here are the situations where your budget belongs somewhere else:

• Your site is a brochure page with no content answering a buyer's question: write first, measure afterwards.
• Your company exists nowhere except on its own website: third-party mentions matter far more than one more file at your domain root.
• Your business generates no online search at all because it lives on local word of mouth or a closed address book: a citation tracker will measure a desert.
• Your platform locks part of the signals and you have no plans to move: your score will plateau, and that ceiling is not a failure.

One reason to measure survives even in those cases: knowing the cost of doing nothing. Ahrefs compared 300,000 keywords between December 2023 and December 2025 and found that the presence of an AI Overview comes with a clickthrough rate for position one that is 58% below the rate forecast without one. The traffic a strong ranking used to deliver does not come back on its own, and finding out where you stand costs less than discovering it in your revenue.