Research
We Audited 151 Sites After the Google Update. The Average Score Was 4.4/10.
Google spent I/O telling everyone the AI answer box is the future. We took 151 real mid-market sites and checked whether any of them are ready for it. Almost none are.
By Aditya Nittoor, Co-founder · 2 June 2026 · 9 min read
The numbers
- 1. We audited 151 mid-market sites across automotive, supplements, and marine.
- 2. The average AI visibility score was 4.4 out of 10.
- 3. Five failures account for nearly every low score. All five are fixable.
- 4. Whoever fixes them first is the one the model names. Everyone else gets skipped.
At I/O in May 2026, Google called its AI search box the biggest upgrade in over 25 years and said AI Mode had crossed a billion monthly users. A model writes the answer now, and that model decides which businesses get a mention. If you sell anything online, here is what that comes down to. When the AI builds the answer, can it read your site well enough to put you in it? For most of the sites we looked at, the answer was no.
So we ran the numbers. Across 151 mid-market sites in automotive parts, supplements, and marine, the average AI visibility score came out to 4.4 out of 10. These aren't bad businesses. Most of them rank fine on Google. The problem is that their sites were built for two audiences, a human reader and a search crawler, and neither one is the model that now has to pull facts off the page, check them, and decide whether to cite you.
The five failures behind almost every low score
1. No machine-readable guide to the site
Nothing tells an AI crawler what the business sells or how the catalog is laid out. The model is left to guess, and when it guesses it usually guesses without you.
2. Specs and facts locked in images or scripts
The compatibility table, the dosage, the dimensions, all baked into an image or a script the model never reads as text. A fact it can't pull off the page is a fact it can't cite.
3. Descriptions written as marketing, not data
Lovely copy that gives a model nothing to match against a real query. No part numbers, no structured attributes, no plain statement of what fits what.
4. Weak or missing entity and location signals
When the model can't pin down who you are or where you operate, it plays safe and recommends someone it's sure about instead. You lose the specific, local, and technical questions, which are the ones worth winning.
5. Crawlers quietly blocked
A security plugin or some forgotten robots.txt rule slams the door on AI crawlers, and nobody on the team knows. Great content fixes nothing if the model can never fetch it.
A 25-facility hospital network that fixed all five of these saw 245,945 views in under 10 days, a 5.5x daily traffic uplift, and zero ad spend. (Full case: hospital network GEO case study.)
“Not one site scored low on the strength of its product. They scored low because nobody ever wrote the site to be read by a machine. I'll take that problem any day. It's plumbing, and plumbing is something you can go fix this quarter.”
— Aditya Nittoor, Co-founder, FlowBlinq
A 4.4 average is an opening, not a verdict
When a whole category averages 4.4, it means almost nobody in it has moved yet. Catching up in classic SEO is brutal. You're trying to out-spend competitors who have a decade of backlinks on you. AI visibility doesn't work that way. You just have to be the first site in your space the model can read cleanly and trust. One laptop repair shop in Australia had zero AI visibility. After fixing its structure, it was cited by name within 24 hours and picked up 655 AI-driven visits in 34 days — because the bar in most categories is still sitting on the floor.
The full methodology, including the scoring parameters, the score distribution, and the worked examples, lives in our deep-dive on the 151-merchant audit. The short version doesn't change. Most mid-market sites land at a 4 or 5 out of 10 for AI, and the ones that fix that first end up owning the category.
Google has been clear about where search is headed. What it hasn't decided yet is whether your site is one it can read when it writes the answer. That part is still on you, and right now most of your competitors haven't touched it either.
Get your own score across the same parameters — free, in about three minutes.
Run a free AI visibility audit →Want the full 17-pillar scored PDF with 250 pages crawled and a competitor benchmark? $10 report →