The 5 core AI visibility metrics that actually matter

Most brands know exactly how many visitors they get through Google. But almost nobody knows how visible they are in AI systems like ChatGPT, Perplexity or Claude.

At Customer Impact, we believe visibility no longer starts with clicks, but with understanding.

Because in the AI era, it is no longer about who gets found most often, but about who is understood best.

Check your own page: get a GEO score with concrete tips through our free GEO check.

What we measure

To know whether a brand is properly “understandable” to AI, we measure five core indicators. Together they form what we call the Generative Visibility Score (GVS), a new way to determine how strongly your brand is present in the cognitive space of language models.

These indicators sit within the broader framework we describe in our guide to Generative Engine Optimization. Where classic SEO measures whether a page can be found, GEO measures whether a model knows your brand, places it correctly and recommends it. The five indicators below turn the abstract notion of “AI visibility” into something concrete and steerable. For each indicator you will read what it means and what you can do today to score better on it.

1. PRR: Prompt Recall Rate

What it measures: how often your brand appears spontaneously in AI answers within your domain. If someone asks ChatGPT for “the best fleet management software in Belgium”, does your name come up, and how high in the answer? PRR is the AI counterpart of a top position in Google: it is the difference between being named and staying invisible at the exact moment a buyer asks for a recommendation.

How to improve it: start with the prompts your buyers actually ask, not with abstract keywords. Make sure clear, citable content exists that links your brand to that specific question, category and region. Models retrieve brand names from repeated, consistent associations. The more often trustworthy sources mention your name in the same context (“fleet management”, “Belgium”, “for transport companies”), the greater the chance the model returns you. Avoid vague self-descriptions: a model can only recommend you for what it understands unambiguously.

2. SAI: Semantic Authority Index

What it measures: how accurately a model understands who you are and what you stand for. Does it call you a “trucking app” or a “fleet intelligence platform”? SAI weighs whether AI reflects your official positioning correctly or paints an outdated, oversimplified or plainly wrong picture.

How to improve it: pick one clear category description and use it identically everywhere: on your homepage, in your about page, in your metadata, in your press releases and on external profiles. Inconsistency is the biggest enemy of a high SAI. If one source says “marketing agency” and another says “growth partner”, the model averages that noise into something vague. On top of that, back your positioning with evidence (cases, figures, expertise) so the model does not just know what you claim, but also has the context to confirm that claim. Structured data and clear headings help models extract that meaning reliably.

3. SDR: Source Diversity Ratio

What it measures: how widely your mentions are spread across independent, trustworthy domains such as Wikipedia, trade media, review platforms, press articles and company registers. AI trusts brands that are confirmed by multiple sources, not brands that only talk about themselves on their own site.

How to improve it: build presence outside your own domain. One strong mention on your website carries little weight; ten consistent mentions on authoritative external sources make your story hard to ignore. Think guest articles in trade media, mentions in independent comparisons, profiles on relevant platforms and citations in research or news. Important: make all those sources carry the same core message. The more independent voices say the same thing, the harder it becomes to “train your brand out” of the model at the next update.

4. TAS: Trust Aggregation Score

What it measures: not all sources are equal. TAS weighs both the authority and the sentiment of every source writing about your brand. A positive mention on an authoritative domain counts for more than a random blog post, and a negative or contradictory source can actively cloud your image.

How to improve it: focus on quality over quantity. One citation in a respected trade publication does more for your TAS than dozens of low-authority mentions. Work actively on reputation in the places where AI models source their trust: recognised media, strong review profiles, authoritative reference works. Keep an eye on sentiment too. Unresolved complaints or outdated, negative information colour how a model summarises your brand. Trust is cumulative: it builds slowly, but afterwards it makes your position stable and defensible.

5. MDI: Model Drift Index

What it measures: AI models change constantly. MDI measures how stable your brand image stays across different model versions and updates. A healthy MDI means your story holds up consistently, even when a model is retrained or updated.

How to improve it: stability comes from consistency over time. Models that encounter your brand in the same clear context every time remember that image more robustly. So do not change your positioning, name or core message every other month, and clear out outdated information before a new training round locks it in. Periodically monitor how the large models describe your brand, so you spot drift early and can correct course before a wrong picture sets in. AI visibility is not a one-off project but an ongoing discipline.

Why this is different from SEO

SEO measures behaviour: clicks, click-through rates, positions. GEO measures understanding: how well a machine grasps your meaning. That difference becomes sharper now that search engines show AI answers, as Google explains about AI in search.

• SEO asks: how many people see us?
• GEO asks: how well does AI understand us?

In a world where AI gives the answer, that understanding is your real brand value. A top position helps little if the model misjudges your category or simply does not mention you in its summary. That is why we do not steer on vanity rankings, but on leads, revenue and correct AI mentions.

How the five indicators work together

The five indicators are not separate gauges but a chain. PRR is the visible end result: are you mentioned or not? But that result rests on the other four. SAI determines whether you get mentioned for the right questions, because a model that misunderstands your category recommends you at the wrong moments or not at all. SDR and TAS determine whether the model trusts your image in the first place: without independent, authoritative sources, your story stays a lonely claim that is easily overwritten. And MDI determines whether that hard-won position stays standing when models get updated.

In practice this means you cannot force a single indicator. A brand that publishes a lot about itself but is confirmed nowhere externally scores high on intent and low on trust. A brand with strong press coverage but unclear positioning does get mentioned, but in the wrong context. Real AI visibility emerges when the five reinforce each other: a clear message, broadly and reliably confirmed, kept consistent over time.

How we measure GEO for our clients

At Customer Impact we use a mix of simulations, model testing and data verification. This is exactly the work we deliver within our generative engine optimization (GEO).

1. Prompt testing
   We test hundreds of prompts in different models (ChatGPT, Claude, Gemini, Perplexity) to see how often and how correctly your brand is mentioned. This mainly feeds your PRR and SAI.

2. Semantic audit
   We analyse the description of your brand: does it match your official positioning? If not, we rewrite your digital foundations (metadata, structure, context) so models pick up your category unambiguously.

3. Source mapping
   We map out everywhere your brand is mentioned, and how consistent those sources are in their language, tone and content. This is where we specifically improve your SDR and TAS.

4. GEO dashboard
   Everything comes together in one overview where you can follow your progress quarter by quarter, including the MDI across model updates. Just like Google Analytics used to be, but for the AI era.

The result

GEO does not change what you do, but how AI sees you.

Companies with their semantic foundation in order get picked up faster by AI systems, and stay visible even as classic search engines lose importance. You build that foundation together with your SEO, so classic findability and AI visibility reinforce each other instead of competing.

The five indicators are not separate scores to show off. They are a steering instrument: they show where AI misunderstands your brand, and therefore where improvement pays off most. Anyone who tracks them structurally builds a lead that is hard to catch up with, because trust and consistency stack up over time.

Want to know how AI sees your brand and which of the five indicators are holding you back today? Discuss it with us through our contact page. Within a first analysis you will see how strongly your brand is present in the mental world of AI, and which steps make the difference.