On this page
The short version
An AI visibility score squeezes five AI systems, many places and many questions into one number. That's handy for a dashboard and dangerous for decisions. Using one business's real numbers, here are six ways it misleads:
- It can move the wrong way. The score fell while the business was named more often. (Direction)
- It hides places. 16.6% overall was really 62% in one suburb and 0% in two others. (Places)
- It hides AI systems. 23% in one, 11% in another. (AI systems)
- It depends on the questions you choose. The same business scored 58% or 5% depending on the question set. (Questions)
- It rarely shows its own range. (Range)
- It breaks when the questions change. (Question changes)
Then: how to read any AI visibility score, including ours.
What an AI visibility score is made of
The short answer: usually some mix of how often AI names you and how often it links to your site, across every question and AI system you track.
Our own tool shows one too, so I'll use it as the example. LLM Visibility Lens calculates it as 70% "named" (the share of AI answers that recommend the business by name) plus 30% "cited" (the share that link to its website), across ChatGPT, Gemini, Google AI Mode, Google AI Overviews and Perplexity.
Other tools weight things differently, add "sentiment" or "position", or don't say how they calculate it at all. Whatever the recipe, every single score has the same problem: it averages things that behave very differently.
Everything below comes from Kato Electrical, a Vancouver-area electrician and our public demo. It covers two back-to-back periods on the same 16 questions: 24 September to 3 October, and the 12 days from 4 October. I sell the tool behind these numbers, so weigh my view. You can check them all in the demo.
It can move the wrong way
The short answer: a combined score can fall while the part you care about rises.
| 24 Sep–3 Oct | From 4 Oct | |
|---|---|---|
| Score | 15.2 | 14.3 (down) |
| Named rate (recommended by name) | 16.3% | 16.6% (up) |
| Cited rate (linked to his site) | 12.7% | 8.9% (down) |
Kato was recommended slightly more often in the second period. His score still went down, because AI linked to his website less, mostly Perplexity. Someone watching only the score would think things got worse.
The small rise in the named rate wasn't a real change: both periods sit within each other's range. The fall in citations is bigger and worth a look. But the score pointed down while the more important number didn't.
What to do: always look at the parts. "Named" is what wins customers. "Cited" is a useful side signal, and we found in when your own web page helps a competitor that a citation without your name often helps someone else.
It hides places
The short answer: for a local business, one number averages suburbs where it dominates with suburbs where it doesn't exist.
Kato's 16.6% in the second period broke down like this:
| Place | Named |
|---|---|
| Burnaby | 62% (31 of 50) |
| Vancouver | 28% (102 of 358) |
| North Vancouver | 20% (12 of 60) |
| Coquitlam | 4% (11 of 295) |
| Surrey | 0% (0 of 119) |
| Squamish | 0% (0 of 60) |
Not one place is anywhere near 16.6%. The single number describes nowhere Kato actually works. It also hides the most useful fact in the data: Coquitlam is where he's losing. We've traced why in Google Business Profile and AI answers.
What to do: for local businesses, read the results place by place.
It hides AI systems
The short answer: the five AI systems behave so differently that averaging them hides who's actually recommending you.
In the same period, Kato was named in:
- Google AI Mode: 22.9% of answers
- Perplexity: 21.4%
- ChatGPT: 14.1%
- Gemini: 13.0%
- Google AI Overviews: 10.9%
These systems read different sources and agree with each other surprisingly rarely. On the same morning, five AI systems agreed only about 1 time in 5, as we showed in 5 AIs, same questions, 1,814 answers. A rise in one can cancel a fall in another and leave the score flat.
There's a subtler trap too. Google shows an AI Overview only for some searches. In this period it showed none on 18 of Kato's checks. If a tool counts those as "not named", the score drops for reasons that have nothing to do with the business. Google's own page on AI features and your website explains that AI Overviews only appear when Google thinks they help. More in AI Overview tracking for near me searches.
What to do: read each AI system separately, and check how the tool treats searches with no AI answer.
It depends on the questions you choose
The short answer: the same business can score high or low depending only on which questions are tracked.
For Kato, in the same 12 days:
- The three "best electrician in…" questions: named in 58% of answers.
- All 16 tracked questions: 16.6%.
- The six long spoken questions ("can you find me an electrician in Surrey who can quote this week…"): 5%.
It gets more extreme with brand names. In an earlier test, a well-known mobility brand was named in 88% of answers to questions containing its name, and 0% of ready-to-buy questions. A tool, or an agency, that tracks mostly brand questions will report a flattering score that means almost nothing for sales.
What to do: ask which questions are behind any score. They should be the questions real buyers ask before they know your name. Our free question finder writes those from your website.
It hides its own range
The short answer: every AI visibility score is an estimate from a sample, and small samples swing.
AI answers change from run to run. The same check, run four times in one day, scored 13, 25, 18 and 20 in our data. SparkToro's research found the same brand list came back less than 1% of the time, as Search Engine Journal reported. The randomness is built into how these models generate text, as Thinking Machines Lab explains.
So a score from 900 answers is roughly ±2 or 3 points. From 50 answers, it's more like ±13. A score with no range or answer count can't tell you whether a 5-point move is real or noise. Our guide on what one free check can and can't do shows the ranges at different sample sizes.
What to do: ask for the number of answers behind the score, and its range. If the range isn't shown, treat small changes as noise.
It breaks when the questions change
The short answer: add or remove questions and the score changes even if nothing else did.
In late September, Kato's tracked questions grew from 10 to 16, adding long spoken questions he's rarely named in. That alone pulls an overall score down, with no change in how AI treats him. The test changed, not his visibility.
What to do: only compare periods with the same question set. A good tool marks the date the set changed and refuses to compare across it. This is also why there's no meaningful "rank #1 in ChatGPT"; see why there's no rank #1 in ChatGPT.
How to read any AI visibility score
Seven questions to ask of any score, ours included:
- What's it made of? Named, cited, sentiment? In what proportions?
- How many answers is it based on, and what's the range?
- What does it look like for each AI system?
- What does it look like for each place, or product line?
- Which questions are behind it, and are they questions buyers ask before they know your name?
- Did the question set change during the period you're comparing?
- How are missing answers counted, such as Google showing no AI Overview, or a failed check?
If a tool can't answer these, use its score as a rough mood indicator and nothing more.
How we handle it in our own tool
We still show a score, because people want one number for a dashboard. But we try not to let it stand alone.
- Every score comes with a range and an answer count. We use a standard 95% range for the named rate.
- Named and cited are shown separately beside the score.
- Each AI system has its own rate, and questions are listed one by one, with places shown in the question text.
- Days with no AI Overview are excluded, not counted as zeros. Failed checks are excluded too.
- Comparisons only run on the same question set. When the set changes, we say "not comparable" from that date.
The full method is on the how it works page. When we report to clients, the score is one line near the top, and the rest of the report explains it. We wrote up that format in how to report AI visibility to clients.
If you're comparing tools, there's a plain comparison with Semrush's AI visibility tool. Plans start free, and agencies can send reports under their own name; see the agency plan.
More reading:
- How to rank in ChatGPT
- How to get recommended by ChatGPT
- Invisible on Google, recommended by AI
- Which websites ChatGPT trusts
- ChatGPT citations
- How to rank in AI Overviews
- Google AI Mode for local businesses
- Perplexity vs ChatGPT
- Spoken questions
- 5 AI systems, 35 questions
- We asked AI who to hire 80 times
There's also the changelog, the FAQ page and the security page. If you'd like me to look at a score you've been given and what's behind it, get in touch.
Questions people ask
What is an AI visibility score?
It's a single number summarising how visible a business is in AI answers, usually combining how often AI names the business and how often it links to its website, across several AI systems and questions. Each tool calculates it differently, so always check what it's made of.
Is an AI visibility score accurate?
It's an estimate from a sample of AI answers, and AI answers change from day to day. A score based on hundreds of answers is usually accurate to within a few points. One based on a few dozen can swing by ten points or more. Ask for the number of answers and the range.
Why did my AI visibility score go down?
Check the parts before worrying. In our example, the score fell while the business was named slightly more often, because AI linked to its website less. Scores also move when the question set changes, when one AI system shifts, or because of normal day-to-day variation.
What is a good AI visibility score?
There's no universal benchmark, because scores depend on the questions tracked and how the tool calculates them. The same business scored 58% on its "best electrician" questions and 5% on long spoken questions. Compare a business with its own past, on the same questions, and with its competitors.
Should I report an AI visibility score to clients?
You can, as one line, but not on its own. Show the named rate with its range, results for each AI system and place, the questions won and lost, and what you'll do next. A single score with no breakdown invites the wrong conclusions.
Harjeet Singh Bedi Founder, LLM Visibility Lens · SEO consultant, Bedi SEO
Harjeet has worked in SEO since 2010, first in agencies and now running Bedi SEO for local businesses, online stores and agencies. He built LLM Visibility Lens after clients kept asking whether they showed up in ChatGPT and other AI tools and nobody could give them an honest, measured answer. Every figure in this article comes from answers his tool collected.