AI visibility score: definition and what it hides

An AI visibility score is a composite metric that summarizes how often, and how well, AI platforms such as ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews mention a brand when answering a fixed set of prompts.

The typical inputs are:

• Mention rate: the share of answers that name the brand.
• Citation rate: the share of answers that link to or cite the brand's pages.
• Sentiment: whether the answers describe the brand positively, neutrally or negatively.
• Accuracy: whether the facts stated about the brand are correct.

Each tool chooses its own components and weights. The number is the tool's model of visibility, not a figure any AI platform publishes. For the full breakdown of what these scores average, see the guide to AI visibility scores.

A tool sends a set of prompts to each engine, records the answers, scores each answer on each component, and then combines the results into one number. The prompt set, the engines, the weights and the scoring rules all belong to the tool. That's why the calculation differs from vendor to vendor.

Two things sit outside the tool's control. First, retrieval is per query and discretionary. As of October 2026, ChatGPT decides for itself when to search, Claude.ai search is a user toggle, and Google AI Overviews often do not trigger at all. Second, engines rewrite prompts into their own search queries, so the prompt a tool sends is not the query the engine runs.

A score is also a snapshot of the prompts asked on the day they were asked. The free checker on this site, for example, scores up to three buyer questions and shows one point in time, not a trend.

Retrieval-on and retrieval-off answers

An answer built from a live web search reflects what the engine could fetch today. An answer without retrieval reflects only what the model learned before its knowledge cutoff. One number averages both, and the tool usually cannot tell which kind each answer was unless citations appear.

Citation units that don't match

Anthropic's web search tool cites short text spans, while Perplexity cites whole sources by index. Google states that AI Overviews and AI Mode "may use different models and techniques" and can return different links. Adding these into one citation figure combines things measured in different units. The difference between citation rate and citation share adds another layer.

Citation is not use

A citation is a weak proxy for influence. Mechanistic research (arXiv 2606.28358) repaired over 90% of missed citations by adjusting internal model components without changing the sources, so whether a source gets cited is partly separate from whether it shaped the answer.

Engines that behave differently

In Otterly's 2026 experiment, Microsoft Copilot produced 2,718 of the 3,934 citations tracked for one site, while several other engines almost never cited it. A blended score would have shown a modest total and hidden that split. This is the case for per-platform scores over one blended number.

• "It's my rank in AI." It isn't. No engine publishes position data for generative answers, so no score can be checked against a real rank.
• "Two tools should agree." Different prompts, engines, weights and dates produce different numbers. That's normal, and it's why AI visibility tools disagree.
• "There's a universal good score." A benchmark only means something against like-sized brands in the same category. See what a good AI visibility score is.
• "A GEO score is the same thing." "GEO score" often means a readiness quiz about your site's setup, not a measurement of AI answers.
• "Content tweaks will raise it." A 2026 critical survey of 45 GEO studies (arXiv 2607.14035) found no technique with a stable, cross-platform causal effect on discoverability. What you control is access: crawler permissions, server-rendered content and index eligibility.

If you report the number upward, use this playbook for explaining score limits to stakeholders.

• Mention rate: share of answers that name the brand.
• Citation rate: share of answers that cite the brand's pages.
• Citation share: the brand's citations as a share of all citations for a query set.
• Retrieval: the step where an engine fetches web pages before answering. Without it, the answer comes from training data alone.
• Knowledge cutoff: the date after which a model learned nothing in training.

Should I track the total score or the components? Track both, but act on the components. The total tells you whether something moved. The components tell you what moved and on which engine. A falling total could mean fewer mentions, a wrong fact spreading, or one platform that stopped searching for your category. Each cause has a different fix, and only the per-engine, per-component view shows which one you're dealing with.

Why does my score change when I haven't changed anything? Engines change models, retrieval behavior and defaults without notice, and they decide per query whether to search the web. Infrastructure can change too: from 2026-09-15, Cloudflare blocks some AI crawler categories by default on certain accounts. Any of these can move a score while your site stays the same. Check crawler access first, then compare per-engine results across dates.

Can a high AI visibility score hurt me? The number itself can't, but it can mislead you. A high mention rate paired with wrong facts means AI platforms are spreading errors about your brand widely. That is why accuracy belongs in any score you rely on. Look at what the answers actually say about your pricing, products and company, not only how often they name you.

Does a better score mean more traffic? Not necessarily. Scores count mentions and citations in sampled answers, not clicks. Some AI surfaces send no referral data, so traffic from them is hard to attribute. Treat the score as a view of how engines describe you, and measure traffic separately in your analytics.