FrontStat data

What is a good AI visibility score? Real numbers from 347 new products

By Alex, maker of FrontStat · Updated 8 October 2026 · 7 min read

A good AI visibility score depends on who built the score. AI search visibility, brand visibility in AI answers, AI share of voice: every AI visibility tool reports a number between 0 and 100 under one of these names, and every tool builds it differently, so "we are at 38" means nothing until you know what was asked, of which AI models, and how an answer was counted. This guide explains the three ways scores are built, gives the real distribution for new products, and says what to compare your number against.

How an AI visibility score is built

Underneath every score is the same unit: one buyer question, put to one AI assistant, on one day, with a yes or no answer to "did it name the product?". Tools differ in three choices.

1. The share. The simplest score is answers that name you divided by answers checked. Ten questions on three assistants is thirty answers; named in nine of them is 30%. This is the honest baseline, and it is what FrontStat shows as "AI visibility".
2. The weighting. Some tools weight a mention by position (first in the list counts more than fourth), by sentiment, or by whether the assistant linked to you. A weighted score of 38 and an unweighted 38 are different animals.
3. The prompt set. Brand-scale tools run 50 to 500 prompts across a category. A founder tool runs the 5 to 10 questions a buyer of that product would ask. A product can score 60 on its own buyer questions and 4 on a category-wide set, and both numbers are correct.

So the first question to ask about any score is "a share of what?". The second is "weighted how?". Only then does "is it good?" have an answer.

The real distribution for new products

FrontStat checks every product that makes its weekly launch board. Each gets its main buyer question put to ChatGPT, Claude and Gemini with web search on, in its launch week. Across the 347 products checked between 17 September and 7 October 2026, the share recommended by each assistant was:

Put together: 73% of new products were recommended by none of the three on their main question, 27% by at least one, and 2% by all three. On the unweighted three-answer share, that means the median new product scores 0%, a product at 33% is in the top quarter, and 100% is rare.

FrontStat's overall launch score, which adds listings, links that count and search for the product's name to the AI answers, ran from 11 to 76 across the same products, with a median of 45. Nobody scored above 80. The AI part of that score is where almost everyone has room.

So what is a good score?

Three answers, for three situations.

You launched this month. One assistant naming you on your main buyer question is a good score. Two is excellent. Zero is normal, and it is the starting point for most products, including ones that go on to do well. What matters is the direction over the next 30 days.

You have been live for a year or more. The "above 50" rule of thumb that circulates comes from brand benchmarks: category-leading brands tend to land between 50 and 70 on broad prompt sets. If you are the known name in a niche and you score under 30 on your own buyer questions, something specific is wrong, usually a page the assistants cannot read or a description they get wrong.

You are comparing two tools' scores. Do not. Compare each tool against itself over time. A score that went from 20 to 40 on the same questions, same assistants, same weighting, is a result. A score of 40 in one tool against 55 in another is two methods.

Free AI visibility checkers, and what their scores mean

Several SEO platforms offer a free AI visibility checker or a free AI visibility report: enter a brand, get a score across ChatGPT, Gemini, Perplexity, Google AI Overviews and AI Mode. These run against the platform's own index of prompts, so an established brand gets a useful number and a product launched this month usually gets zero or "no data", because no indexed prompt names it yet. That is a fact about the index, not about the product. A checker that writes a buyer question for your product and asks it, as FrontStat's does, gives a new product a real score on day one. The two kinds of free tool answer different questions: "how visible is this brand across AI search?" against "would an AI answer name this product for the question its buyers ask?".

Why the score moves without you doing anything

Assistants that search before they answer read whatever is current. A single answer varies from one check to the next, which is why FrontStat confirms a change on ChatGPT or Gemini with two more samples before telling anyone, and why a score built on one check a week is noisier than one built on a check a day. Expect the main-question share to flip on one assistant every week or two for a new product; expect it to settle as the assistants gather more pages about you.

A product named in 100% of answers on day one is usually a product with a very specific question that only it answers. That is a fine place to be, but the useful scores are the ones that leave room to improve.

What correlates with a higher score

From the same 347 products, the share recommended by at least one assistant, split by one fact at a time:

The first row is the one to act on. If you want a higher score, make your site a page the assistants cite. The steps are in How to improve AI visibility.

Check your own score

The free checker gives you the three-answer share for your main buyer question, who each assistant named instead, and the pages it cited. FrontStat Pro tracks up to 10 questions every day, so the score you watch is built the same way every morning and the trend means something.

Questions