The 17.2% brand-mention benchmark: how visible is your store in AI answers, really?

There is no single "17.2%" brand-mention benchmark that your store can be measured against — the honest answer is that published brand-mention rates swing from about 21% to 84% depending only on which AI engine you ask, according to Semrush's June 2026 Ghost Citations study. Any tidy universal figure is a mirage; the only mention rate that means anything for your store is your own, measured on the prompts your buyers actually type.

Watching a store's real numbers move is a daily thing for us, because we build Contexta, a WooCommerce SEO plugin driven by the site's own Google Search Console data — so the "what's my AI number, and is it good?" question lands on our desk constantly. The reason we distrust any headline benchmark is simple: we have re-run the same prompt set a week apart on the same store and watched the result move by double digits with nothing changed. Here is what the real benchmarks say, why no single figure travels between stores, and a DIY way to get the one number that matters — yours.

Is there really a single brand-mention benchmark for AI answers?

No — there is no universal brand-mention benchmark, because the number depends more on which engine you query and how big your brand is than on anything you wrote. Semrush's Ghost Citations study (June 9, 2026), built on 3,981 domain appearances across 115 prompts in 14 countries, found brands were actually named in only 38.3% of appearances overall — but the per-engine split is the part that kills any single benchmark. ChatGPT names a brand in just 20.7% of its answers, while Gemini names one in 83.7%. Same web, same brands, four times the mention rate depending purely on which assistant answered.

Gemini names a brand in its answer84%

Cross-engine average38%

ChatGPT names a brand in its answer21%

So a benchmark quoted as a bare percentage, with no engine attached, is not a benchmark at all. A figure like "17.2%" could describe a strong performer on ChatGPT and a weak one on Gemini in the very same afternoon. Before you accept any number as a target, the first question is always: measured on which engine?

What is a "normal" brand-mention rate for a store like mine?

For a typical small or mid-sized store, a measured citation rate in AI answers of low single digits is normal, not a failure. A 200-site cohort tracked between January 12 and April 28, 2026 by Attrifast found median citation rates of 1.4% for solo brands, 3.4% for 10–50-person companies, 6.2% for mid-market, and 11.4% for enterprises — with a cohort-wide median of just 4.1%. The category-leading 40%+ numbers that vendors love to quote describe the top of the market, not a five-person WooCommerce store.

1. 1

   Enterprise, 500+ staff — 11.4%

   Category leaders with heavy backlinks and PR

2. 2

   Mid-market, 50–500 staff — 6.2%

   Established brands with real search authority

3. 3

   Growth, 10–50 staff — 3.4%

   Building presence, patchy across engines

4. 4

   Solo, under 10 staff — 1.4%

   Where most independent stores actually sit

This is the context missing from almost every benchmark headline. If your store sits in the solo or growth band and a tool tells you the benchmark is 17% or 40%, it is comparing you to companies with a hundred times your link profile. The honest read: roughly 9 in 10 brands register no measurable AI mentions at all (Search Engine Journal, July 2026), so simply being named some of the time already puts you ahead of most of the field. Judge yourself against your own size band and your own engines — the volume side of that picture is covered in what percentage of traffic AI assistants actually send.

Why does the same store get a different number every time you check?

Because AI answers are non-deterministic and engine-specific, so the same prompt set genuinely returns different results run to run. Citation volume for a single brand can differ by up to 615x across platforms (Superlines, March 2026), and even within one engine, answers drift with session, personalization, date and geography. A benchmark snapshot is a weather reading, not a constant.

There is a second trap hiding inside the word "mention." A domain can be cited as a source link without the brand name ever appearing in the answer text — Semrush found 62% of citations are exactly these "ghost citations." So "mention rate" and "citation rate" measure different things, and a tool that blends them will hand you a number you can't act on. This is the same reason a mention rate makes a poor KPI: it is directional at best, which is why proving the channel earns money needs a different chain entirely, laid out in how to prove AI visibility actually pays.

How do I test my own store's category prompts?

Build a prompt set that mirrors how buyers actually phrase category questions, run each one several times across the engines that already send you traffic, and record "mentioned" and "cited" as separate columns — that gives you a dated, defensible mention rate for your store instead of a borrowed benchmark. The whole test fits in an afternoon.

Expect the engines to disagree with each other far more than they agree. ChatGPT and Perplexity overlap on sources only about 11% of the time, so a single-engine test tells you very little, and why AI engines cite different sources explains what drives the split.

1. 1

   Pull real category prompts

   Take how buyers phrase non-branded questions from Google Search Console — best-in-category, comparison, problem-solution, shortlist — not keywords you invented.

2. 2

   Pick engines by traffic

   Test the assistants your analytics already shows citing you first, rather than running all four equally.

3. 3

   Run each prompt three times

   Space the runs hours apart and log the date, country and model version, because answers drift between runs.

4. 4

   Score mentioned vs cited separately

   Mark when your brand name appears in the text, and separately when your domain is linked — the two rarely match.

5. 5

   Compute your rate and share of voice

   Your mention rate is named answers divided by total prompts; note which competitors were named alongside you.

The part people skip is the first step. A prompt set you dream up tests how you talk about your products, not how customers do — and AI answers reward the customer's phrasing. Pulling the actual non-branded queries from Search Console keeps the test honest. The four intent clusters that matter for a store are category (best running shoes for flat feet), comparison (Brooks vs Hoka for beginners), problem-solution (how do I stop heel slipping in trail shoes), and shortlist (recommend a waterproof hiking boot under $150). Run those, not is [your brand] good, which almost always returns a flattering, useless yes.

My mention rate is basically zero — is that a content problem?

Usually not — the most common cause of a flat zero is not weak writing, it is that the AI literally cannot read your store. Before you rewrite a single product page, confirm the named AI bots — GPTBot, OAI-SearchBot, PerplexityBot — can actually fetch a live URL and that your content is present in the raw HTML without JavaScript. A page that ranks fine in Google can still be invisible to these crawlers, which is the whole subject of why AI assistants won't cite your blog.

This is the check we built Contexta's AI Visibility test around: it fetches your pages as GPTBot, OAI-SearchBot and PerplexityBot, reads your robots.txt, flags a Cloudflare block in the way, and shows what the page looks like with JavaScript switched off — so a zero mention rate points you at the real cause instead of a guess. Ruling out reachability first saves you from rewriting content an assistant was never able to see. Once the bots can reach and read the page, the prompt-testing loop above becomes worth running again, because now the numbers can actually move.