Measured on 7 October 2026 · same protocol as the SME barometer
Can AI read brand websites? We measured 456 sites in cosmetics, SaaS and fashion.
Our barometer measures the websites of French SMEs every month. Here is the same diagnosis on another population: the websites of well-known brands, in three sectors where an AI recommendation already weighs on purchases.
The sample: brands, not SMEs
456 websites of brands present in France: 150 in cosmetics, 150 in online software (SaaS), 156 in fashion. RadarIA drew up the list from well-known brands in each sector: it is not a random draw, and the figures hold for this list only. Each site was read on 7 October 2026, by the same robot as our free tool.
What it shows
Deliberately blocking AI remains rare, even among brands: 1.1% of sites shut the door on a search robot (ChatGPT, Claude, Perplexity). That is not where visibility is lost.
The door closes elsewhere, without anyone deciding it: 11.6% of sites refuse or do not answer a polite, identified robot, and 19.2% in fashion. It is usually an anti-bot firewall. It does not target AI, but it can stop it too: that is checked with the hosting provider, not in robots.txt.
And information is missing: 71.6% of SaaS sites write no price on their homepage, and 88.5% of sites show no date. An AI that has to compare offers, or judge whether a page is up to date, goes without them.
Results, sector by sector
| What was found | Cosmetics | SaaS | Fashion | All | SMEs (September 2026) |
|---|---|---|---|---|---|
| Sites read | 150 | 150 | 156 | 456 | 1,200 |
| At least one search bot blocked — the site cannot be quoted | 0.0% | 1.3% | 1.9% | 1.1% | 1.1% |
| of which OAI-SearchBot (ChatGPT) | 0.0% | 0.0% | 1.3% | 0.4% | 0.9% |
| of which Claude-SearchBot | 0.0% | 0.0% | 1.3% | 0.4% | 0.6% |
| of which PerplexityBot | 0.0% | 0.7% | 1.9% | 0.9% | 0.9% |
| GPTBot blocked — training only, costs no citation | 2.0% | 1.3% | 1.9% | 1.8% | 3.5% |
| robots.txt unreadable: the server refuses or does not answer | 11.3% | 4.0% | 19.2% | 11.6% | 19.9% |
| noindex or nosnippet tag on the home page | 0.8% | 0.7% | 0.0% | 0.5% | 1.2% |
| Content that probably depends on JavaScript | 7.1% | 1.4% | 2.6% | 3.7% | 4.6% |
| No price written out in full | 27.8% | 71.6% | 26.7% | 43.6% | 71.2% |
| No readable update date | 96.0% | 73.0% | 99.1% | 88.5% | 74.9% |
Two denominators: what is read in robots.txt covers every site, what is read in the page covers only the pages actually read. The SME column recalls the latest monthly barometer: it is another population, to read side by side, not as a trend.
What it does not mean
That a site “unreadable” for our robot is necessarily unreadable for ChatGPT: we have not checked it, and we do not claim it. Nor that writing a price will get a brand cited: no before/after study shows that today. This page measures doors and information, not citations.
The limits, in plain words
A single reading, on one date: a site may change the next day. A chosen list, not a random sample. The homepage only: a price may appear on another page. No brand name is published: we aggregate, we do not point fingers.
And your client's site — where does it stand?
The same diagnostic, on any address. Free, no account, ten seconds.
Check a site →