What an AI Visibility Score Actually Measures

Quick read: Multiple vendors now publish AI-search benchmark data and proprietary "visibility scores." Conductor's 2026 AEO/GEO Benchmarks Report, released in November 2025, aggregated 3.3 billion sessions across 13,770 enterprise domains in 10 industries and found AI referral traffic averaging 1.08% of all website traffic, growing roughly 1% month over month, with ChatGPT accounting for 87.4% of that AI referral traffic (Conductor). Separately, Profound sells a proprietary "AEO Content Score," a machine-learning model trained on Profound's own dataset that scores a page's likelihood of being cited by AI search tools (Profound). These reports and scores measure industry-wide or platform-wide patterns — none of them can tell an individual small business whether it personally shows up when someone asks an AI tool for a recommendation. That requires a direct check, which is free to do.

A version of the same pitch keeps landing in inboxes this year: a vendor offering to run a business through an "AI Visibility Audit" and hand back a score out of 100, no methodology attached, renewable monthly. I get asked about a version of this every few weeks now. The honest answer is that the score itself is usually the least useful part of what's being sold.

That's not a knock on the whole category — some of this data is genuinely good, and I'm about to use a chunk of it. The problem is narrower: a benchmark report built from thousands of enterprise domains, or a proprietary score built from one vendor's private training data, is answering a different question than "does my shop show up when someone asks ChatGPT where to get this done locally." Those two things get sold as the same product a lot.

What a "GEO benchmark" or "AI visibility score" actually is right now

SEO gets a page found, AEO and GEO get it cited — and "GEO benchmark" or "AI visibility score" is the label vendors have landed on for measuring that last part. The scores and reports being sold under that label right now are two different things wearing the same acronym, and it's worth separating them.

The first is an aggregate benchmark report — a vendor pools traffic or citation data across a large sample of domains and publishes industry-level findings. Conductor's is the clearest recent example: 13,770 enterprise domains, 3.3 billion sessions, split across 10 industries and 22 sub-industries, comparing traditional organic traffic against AI referral traffic and Google AI Overview appearances (Conductor). OtterlyAI has published something similar at a citation level — an analysis of more than a million citations across ChatGPT, Perplexity, and Google AI Overviews collected in January and February 2026 (OtterlyAI). These are real, useful, and tell you what's true across a market. They don't, and don't claim to, tell you what's true about one specific business.

The second is a proprietary per-page or per-brand score, sold as a product. Profound's AEO Content Score is the clean example: a machine-learning model trained on Profound's own dataset of pages it has seen get cited, evaluating a new page against patterns like structured data usage, heading and paragraph structure, and topical alignment, then outputting a number (Profound). That's a legitimate product for the customers it's built for — mostly larger marketing teams running Profound's platform across a big content library. But the model, the training data, and the scoring weights are Profound's, not published for anyone to check. There's no way for an outside business to independently verify what a "72" versus an "89" actually means, because nobody outside the vendor can see the math.

More of both are coming. OtterlyAI is headlining brightonSEO San Diego this September as an AI-search-visibility sponsor, with its own next benchmarks report expected around the event. None of that changes the underlying advice below — it just means the pitch in your inbox is going to get more crowded, not less, before it gets more standardized.

Neither category is fake. Both are also not what a solo owner checking "am I showing up in AI search" actually needs.

Why this matters if you run the business, not the marketing department

If you're a local service business, the number worth knowing from that Conductor data isn't the AI-referral-traffic figure alone — it's that alongside a small, slow-growing referral share, Google AI Overviews already appear on roughly a quarter of searches on average, a figure this site has cited before from the same Conductor research (Conductor). Put those together and you get the real shape of the problem: AI tools are already answering questions constantly, but most of those appearances never send a click at all. A benchmark report averaged across 13,770 enterprise domains can't tell you whether your specific answer is the one getting surfaced when someone in your town asks. Only checking your own visibility can tell you that, and it doesn't require buying a score to do it.

I ran a dog grooming salon for years before any of this, and I can tell you exactly how much appetite a business owner has for parsing a vendor's proprietary scoring methodology at the end of a working day: none. That's why a five-minute manual check beats a subscription to a black-box number nobody outside the vendor can explain.

The mechanics: a free check that tells you more than a score does

• Search the way a customer actually would, not your own business name. Open ChatGPT, Perplexity, and Google, and ask the question a customer would type — "best [what you do] near [your area]" — not your business name. Searching your own name tests whether the tool has heard of you, not whether it recommends you.
• Log what comes back, and who else does. Note which businesses get named, whether yours is one of them, and whether it's cited as a source versus actually recommended — those are two different outcomes decided by two different steps of the same system, and conflating them is exactly what buying a single "visibility score" tends to do.
• Check Google Search Console's AI Search report if you have UK-market access, and prioritize AI Overview appearances directly if you don't. Google shipped a dedicated report for this in June 2026, though it launched region-limited — in the meantime, note which of your queries in the regular Search Console Performance report show high impressions but unusually low click-through, a pattern that often means an AI Overview is answering the question before a click happens.
• Repeat it monthly, same queries, same order. One check tells you where you stand today. A repeated check tells you whether something you changed — a new page, updated hours, a review response — actually moved the needle, which is the only number that matters more than a vendor's score.

What won't help

• Paying for a one-time "AI visibility score" and treating the number as an action plan. A score without a public methodology can't be diagnosed — you can't fix what you can't see the inputs to.
• Assuming a high score from one platform means visibility everywhere. A score trained on one vendor's dataset reflects that vendor's citation patterns, not ChatGPT's, Perplexity's, and Google AI Overviews' patterns combined, and those three don't behave identically.
• Chasing the score instead of the underlying page quality. The signals these tools evaluate — clear structure, accurate and current information, real topical depth — are worth having regardless of whether a vendor ever scores them. The score is a proxy; the page is the actual asset.
• Treating a benchmark report's industry average as your business's grade. An average across 13,770 enterprise domains includes businesses with entirely different scale, content volume, and market position than a local service business. It's directional context, not a target to hit.

FAQ

What is an AI visibility score?

It's a number, usually proprietary to whichever vendor is selling it, meant to represent how likely a business or page is to be cited or recommended by AI search tools like ChatGPT, Perplexity, or Google AI Overviews. Methodology and training data are typically not published.

Are GEO benchmark reports trustworthy?

The aggregate ones from established research and analytics firms — measuring real traffic and citation data across large samples — are generally solid for understanding market-wide trends. They're not designed to tell you about one specific business, which is a different question than what they answer.

How much of my website traffic is actually coming from AI tools right now?

Industry-wide, it's still small. Conductor's 2026 benchmarks report put AI referral traffic at an average of 1.08% of total website traffic across 13,770 domains it tracked, growing about 1% a month — small today, but the trend line matters more than the current number.

Do I need to buy a tool to check my AI search visibility?

Not to start. A manual monthly check across ChatGPT, Perplexity, and Google — searching the way a customer would, not your own business name — gives you the same core information a paid tracking tool automates.

What's the difference between being cited and being recommended by AI search?

Being cited means an AI tool pulled your page as a source. Being recommended means it actually named your business as the answer to the person's question. A page can clear the first without clearing the second.

Is a high AI visibility score the same as ranking well on Google?

No. It's measuring a different system with different inputs. A page can rank well organically and still not get cited or recommended by an AI tool, and vice versa.

Should I wait for a vendor's benchmark report before doing anything about AI search visibility?

No — the manual check above works today, doesn't require waiting for anyone's report, and tells you where your business actually stands rather than where an industry average sits.

---

Checking your own AI search visibility by hand is free and takes fifteen minutes. Turning what you find into pages that consistently get you cited — and recommended — is the slower, ongoing work Care exists for.