Guide
What is an AI visibility score, and what does it miss?
Updated 21 October 2026 · 6 minute read
One app we track scored 15 out of 100 on AI visibility last month. In the same month it was the app AI assistants named first when merchants asked which one to install. Both numbers are right, and the gap between them is worth understanding before you act on either.
Key takeaways
- An AI visibility score measures how often assistants name your brand in the prompts that tool tracks, so the prompts decide the score as much as your visibility does.
- Most tools draw those prompts from a consumer search database, which means the score reflects shopper questions rather than anyone choosing software.
- A Shopify app can score 15 out of 100 on brand mentions and still be the app assistants name first when a merchant asks which one to install.
- Check which prompts a tool is scoring you on before you act on the number, because the fix for a low shopper score is not the fix for a low merchant one.
What is an AI visibility score?
An AI visibility score is a 0 to 100 rating of how often AI assistants mention your brand across a set of tracked prompts. Tools such as Semrush, Peec and Otterly run those prompts through ChatGPT, Gemini, Perplexity and others, count the answers that name you, and turn the rate into a single number you can watch over time.
Useful number. Incomplete number. The gap between those two is this whole page.
Here is one Shopify app, measured three ways inside a month.
What is the score built from?
The app runs a content campaign. Its monthly report is the kind any agency sends: traffic, keywords, backlinks, what moved.
August was soft. Site-wide organic traffic fell about 13%, ranking keywords about 9%. Then there was this.
By every normal measure, this is working.
Why can a good app score 15 out of 100?
Semrush also sells an AI Visibility Toolkit. Same domain, same week, it came back with 15 out of 100 and two ChatGPT mentions.
The cited-pages view gave the game away instantly. The pages assistants were reading were those same explainer articles.
So the content carrying 80% of the traffic is also the content the AI reads. It is just being read by people solving their own problem, not by anyone choosing a Shopify app.
Which measurement predicts installs?
AppNod asks a different question. Each category gets 50 written merchant questions, including problem-first ones naming no app at all, put to five assistants.
On those, same month, this app was named in 47% of answers and ranked first in its category for share of voice.
- Google rankings, from content
- 8,300
- keywords, 1,266 in the top three. August 2026.
- Semrush AI Visibility
- 15/100
- 2 ChatGPT mentions. Prompts are shoppers. October 2026.
- AppNod, merchant questions
- 47%
- Named in answers, first for share of voice. October 2026.
Why are all three numbers correct?
Because the three tools are not asking the same thing.
Those 8,300 keywords are almost entirely not why assistants recommend this app to merchants.
They are a real asset earning real traffic. They just sit beside the install decision rather than feeding it.
None of this means the content was wasted or that Semrush is measuring badly. The expensive mistake is reading a 15 as “we have an AI problem” and commissioning more of the content that produced it.
What do assistants read before recommending an app?
You can just look. With web search on, an assistant cites its sources, and those citations are recordable.
Here is the list for Shopify review apps. Published AppNod leaderboard, a different category from the app above, included because the pattern is the point and the data is already public.
Measured 12 October 2026. Check it against the published leaderboard.
Three things fall out of that list. Not one of them is an explainer article.
The App Store listing is doing most of the work
45% of searched answers. More than twice any vendor site. If it does not describe your app the way a merchant would, nothing else you publish closes that gap.
Your competitors’ own sites are being read as sources
judge.me at 34%, wiserreview.com at 22%. Assistants are repeating your rivals’ version of the category, including what their comparison pages say about you.
A short list of publishers decides the rest
charleagency.com and eevy.ai at 11% each. That is two pages. Getting named on them is an afternoon, not a quarter.
An AI visibility score tells you which pages cite you. The list worth having is the opposite: pages that decide your category and never mention you.
SEO calls these positions 11 to 20: a refresh beats a new article. That app has 2,400. The AI equivalent is the cited page that never names you, and most apps have under ten.
One check first. Assistants can only cite a page they are allowed to read, so if your firewall turns ChatGPT’s crawler away, none of this applies yet. The crawler log analyzer settles it from a week of logs; the free tools page has the rest.
What should you do about a low score?
Nothing about the existing content changes. It earns its keep and it compounds. August proved it holds when the rest of the site does not.
What gets added is a second, much smaller pile of work aimed at the other question.
The lesson from August is not that one number was low.
It is that an app can be winning the question it measures, losing the question it does not, and never find out, because nothing it looks at is pointed at the second one.
Common questions
Does a higher AI visibility score mean more installs?
Not necessarily. The score reflects mentions in the prompts being tracked. If those prompts are shopper questions, the score can move a long way without any merchant being told to install your app.
Can I just ask ChatGPT myself?
Once, as a sanity check. The problem is repeats: ask the same question twice and you can get two different answers, so a single check tells you very little. That is why the figures here come from asking each question several times across five assistants.
Should we stop writing shopper content?
No. In the case above it is carrying the site, and it held up in a month when everything else fell 41%. The point is that it is doing one job, not two, and the second job needs its own small pile of work.
Why is the app not named?
Its category leaderboard is published, so naming the category beside “ranked first” would identify it in one click. Two of the three figures come from reports that are not public. Every number here is unchanged.
See which apps get recommended in your category
AppNod runs 50 merchant questions across ChatGPT, Claude, Gemini, Google AI Mode, Google AI Overviews and Perplexity, stores every answer, and shows you which app won each question, what the assistants cited while deciding, and which of those pages never mention you.