Making your company readable by AI. Outside, and in. The speed of the market’s websites, measured
The same twenty players as the market register, run through the same public instrument: Google’s PageSpeed Insights, homepage, mobile, readings of 18 and 19 August 2026. The register says what each player declares; this page measures what each player does. Anyone can reproduce any line by pasting the URL into the tool.
The method, before the numbers
Two kinds of numbers, not to be confused. Field Web Vitals come from 28 days of real Chrome users: they are stable and identical for anyone who queries the tool. “Not enough field data” means Google does not publish these measures for this website, most often for lack of sufficient traffic, sometimes because another condition of its Chrome panel is not met: that is a fact, not a criticism, and our own website is in that case. The lab scores (performance, accessibility, best practices, SEO) are estimated by a simulated pass: they vary from one pass to the next, usually by a few points.
Agentic browsing is the tool’s newest category, still flagged as experimental by Google. It evaluates, through deterministic checks, how the website is built for machine interaction: agent-readable accessibility, display stability, the presence of an llms.txt file, and support for the proposed WebMCP standard. The score is a fraction of passed checks, whose denominator varies with the checks applicable to each site; we publish the displayed value as is.
One pass per site, “n. m.” for not measured when the tool did not deliver its result, and the dated screenshot published below the table, line by line: the evidence is there to look at. A player who spots a discrepancy can write to us: the line is re-measured and the correction published, dated. The page is re-read at every quarterly reading, like the register.
The readings of 18 and 19 August 2026
| Player | Field Web Vitals | Performance | Accessibility | Best practices | SEO | Agentic browsing |
|---|---|---|---|---|---|---|
| Eskimoz | Failed (INP 222 ms) | n. m. | n. m. | n. m. | n. m. | n. m. |
| L’Agence WAM | Passed | 37 | 77 | 96 | 92 | 2/3 |
| datashake | Passed | 92 | 93 | 96 | 92 | 1/2 |
| Noiise | Passed | 96 | 89 | 100 | 100 | 2/3 |
| Yumens | Failed (LCP 3,5 s, CLS 0,1) | 56 | 87 | 96 | 100 | 3/3 |
| Webconversion | Passed | 63 | 85 | 100 | 92 | 2/3 |
| SEO.fr | Passed | 56 | 93 | 96 | 100 | 1/2 |
| Natural-Net | Passed | 60 | 86 | 96 | 100 | 1/2 |
| Optimize 360 | Not enough field data | 90 | 91 | 100 | 92 | 3/3 |
| Promoovoir | Failed (LCP 2,5 s) | 82 | 100 | 73 | 100 | 3/3 |
| Uplix | Failed (LCP 4,4 s) | 95 | 82 | 96 | 92 | 2/3 |
| iaba | Not enough field data | 88 | 90 | 100 | 92 | 2/3 |
| SEO-GEO | Not enough field data | 62 | 82 | 96 | 100 | 2/3 |
| Growth Room | Passed | 51 | 85 | 96 | 85 | 1/2 |
| IA Agency | Not enough field data | 58 | 85 | 96 | 100 | 2/3 |
| JAIKIN | Not enough field data | 73 | 100 | 96 | 100 | 3/3 |
| Paul Vengeons | Not enough field data | 87 | 88 | 100 | 100 | 1/2 |
| Breizh AI | Not enough field data | 82 | 96 | 100 | 92 | 2/3 |
| ADN IA | Not enough field data | 86 | 100 | 100 | 100 | 3/3 |
| HENUSSE | Not enough field data | 100 | 100 | 100 | 100 | 3/3 |
Our acceptance threshold for the websites we deliver is public: a minimum median of 90/100 across the four scores at delivery. This page applies the same instrument to us as to everyone else.
The captures, line by line
Eskimoz · capture of 2026-08-18
Reading of 2026-08-18, mobile. Reproduce this measurement in the tool.
The tool aborted on every attempt (RPC_DEADLINE_EXCEEDED), 4 tries on 18 August 2026.
L’Agence WAM · capture of 2026-08-19
Reading of 2026-08-19, mobile. Reproduce this measurement in the tool.
datashake · capture of 2026-08-19
Reading of 2026-08-19, mobile. Reproduce this measurement in the tool.
Noiise · capture of 2026-08-18
Reading of 2026-08-18, mobile. Reproduce this measurement in the tool.
Yumens · capture of 2026-08-18
Reading of 2026-08-18, mobile. Reproduce this measurement in the tool.
Webconversion · capture of 2026-08-18
Reading of 2026-08-18, mobile. Reproduce this measurement in the tool.
SEO.fr · capture of 2026-08-18
Reading of 2026-08-18, mobile. Reproduce this measurement in the tool.
Natural-Net · capture of 2026-08-18
Reading of 2026-08-18, mobile. Reproduce this measurement in the tool.
Optimize 360 · capture of 2026-08-18
Reading of 2026-08-18, mobile. Reproduce this measurement in the tool.
Promoovoir · capture of 2026-08-18
Reading of 2026-08-18, mobile. Reproduce this measurement in the tool.
Uplix · capture of 2026-08-18
Reading of 2026-08-18, mobile. Reproduce this measurement in the tool.
SEO-GEO · capture of 2026-08-18
Reading of 2026-08-18, mobile. Reproduce this measurement in the tool.
Growth Room · capture of 2026-08-18
Reading of 2026-08-18, mobile. Reproduce this measurement in the tool.
IA Agency · capture of 2026-08-18
Reading of 2026-08-18, mobile. Reproduce this measurement in the tool.
JAIKIN · capture of 2026-08-18
Reading of 2026-08-18, mobile. Reproduce this measurement in the tool.
Paul Vengeons · capture of 2026-08-18
Reading of 2026-08-18, mobile. Reproduce this measurement in the tool.
Breizh AI · capture of 2026-08-18
Reading of 2026-08-18, mobile. Reproduce this measurement in the tool.
ADN IA · capture of 2026-08-18
Reading of 2026-08-18, mobile. Reproduce this measurement in the tool.
HENUSSE · capture of 2026-08-18
Reading of 2026-08-18, mobile. Reproduce this measurement in the tool.
Why speed matters here
A slow or hard-to-crawl website may be retrieved less well, by search engines and answer engines alike; that is a reasonable working hypothesis, not a demonstrated rule. Agentic browsing, for its part, measures three machine-readability checks in Chrome: part of what this consultancy sells, never the citation itself. For the rest of the decision, see how to choose and the market register.