Agentic Web Census

August 2026by Omitsis

English

Technology · August 2026

Is Performance Lab ready for AI agents?

Performance Lab scores 14.3%, indistinguishable from the web overall once the confidence margin of ±7.99 is taken into account.

Poor14.3%

AWCS

What the AWCS is, and how it is measured

The figures the index is made of, with their own series: a technology whose index is flat can still have moved on one dimension.

Measured on 44 pages, 93.6% of which carried the agentic category, with a confidence margin of ±7.99.

= +0.5
AWCS over 4 months

14.7%14.3%

Read

What Read measures

Lighthouse builds the accessibility tree an agent would traverse and scores the page pass or fail. The denominator is only the pages the audit could evaluate: a notApplicable or an error is excluded rather than counted as a failure, and the fraction that could be evaluated is published beside it as the coverage.

n=44 · coverage 93.6% · margin ±7.99 pp.

53% of the score
Read on its full scale

Poor15.9%

Read: 4 months, no significant change= −5.0 pp
Guide

What Guide measures

The llms-txt audit passing: a file is served AND Lighthouse considers it valid. The denominator is every sampled page, so this is a fraction of the whole web and not of the sites that have a file. This is the figure the index uses, because a broken file does not guide an agent.

n=47 · coverage 100.0% · margin ±7.73 pp.

27% of the score
Guide on its full scale

Poor22.7%

Guide: 4 months, up▲ +11.1 pp
Act

What Act measures

A count, not a percentage: how many sampled sites register WebMCP tools that an agent can call. It is the audit that defines the Act dimension — the other two WebMCP audits measure how good those tools are, not whether they exist.

n=44 · coverage 93.6% · margin ±7.99 pp.

20% of the score
Act on its full scale

Very poor0.0%

Act: 4 months, no significant change= 0.0 pp
Reachable

What Reachable measures

Read from the robots.txt of the sampled sites: of the AI agents we track, how many are NOT disallowed. Gradual and not a yes or no — turning away one agent out of seventeen is not the same as turning away all of them. A site with no robots.txt blocks nobody, so it counts as fully reachable: that is a measurement, not a gap. This is the only figure that MULTIPLIES the index instead of adding to it, because it is a precondition: if an agent cannot fetch the page, nothing else about the page matters. One caveat we would rather state than hide: a blanket Disallow under User-agent: * is not counted, because the corpus-wide measurement cannot see it either, and we would rather be consistent than clever.

n=47 · coverage 100.0% · margin ±7.73 pp.

multiplies the rest
Reachable on its full scale

Excellent98.0%

Reachable: 4 months, no significant change= +1.4 pp

By major version

Does upgrading help? 2 majors of Performance Lab clear the sample gate, oldest first. Each one is measured on its own sites, so a version with few of them carries a wide margin — the tooltip has it.

Performance Lab 3

What Performance Lab 3 measures

The AWCS of the sites where Performance Lab 3 was detected, on its own sample. Swapping one version for another does not change a site by itself — this says what sites on that version score, which is what makes «does upgrading help?» answerable at all.

n=3 · coverage 100.0% · margin ±30.60.

Performance Lab 3 on its full scale

Poor17.8%

Performance Lab 3: 4 months, no significant change= −2.6
Performance Lab 4

What Performance Lab 4 measures

The AWCS of the sites where Performance Lab 4 was detected, on its own sample. Swapping one version for another does not change a site by itself — this says what sites on that version score, which is what makes «does upgrading help?» answerable at all.

n=41 · coverage 93.2% · margin ±8.28.

Performance Lab 4 on its full scale

Poor14.0%

Performance Lab 4: 4 months, no significant change= +0.0

Measured, and not part of the index

These are signals, not scores. They are kept apart on purpose: mixed in with the dimensions, a low one reads as a bad grade for something the index never counted.

53.9%
Google's category score

Google's category score

Sample
n=44
Coverage
93.6%

The score Lighthouse gives the agentic-browsing category, read from the report as published. We republish it as a declared reference and never as a figure of ours: it includes cumulative-layout-shift — a rendering metric — at half the weight, and its weights are conditional and undocumented. A number whose calculation we do not control is not a number we cite as our own.

Google's category score: 4 months, no significant change= −1.6
8.5%
Sites with an AI crawler policy

Sites with an AI crawler policy

Sample
n=47
Coverage
100.0%

Not a Lighthouse audit: it is read from the robots.txt of the sampled sites, checking whether the file names any of the AI crawlers we track. It measures HAVING AN EXPLICIT POLICY and not permitting: blocking AI crawlers is a legitimate decision, and penalising it would be a value judgement dressed up as a measurement. What is a sign of maturity is having decided.

Sites with an AI crawler policy: 4 months, no significant change= −2.1 pp
45.5%
Sites serving an llms.txt

Sites serving an llms.txt

Sample
n=47
Coverage
100.0%

The same llms-txt audit as the figure below, read differently: this one counts SOMETHING being served at /llms.txt, valid or not. Read that literally, because the distance between the two figures is large and it is the interesting part: of the pages that serve something, fewer than two in five serve a file that parses. Our reading — and we cannot verify it without fetching those files ourselves, which we do not do — is that a good share of the rest are soft 404s: a server answering 200 with an HTML error page. So the figure to cite is the valid one below, and this one is best read as an upper bound. Pages whose audit errored are excluded from both, and the coverage says how many that was.

Sites serving an llms.txt: 4 months, up▲ +12.9 pp

Against the rest of Performance

The 28 technologies in this category with a published index, and the web overall for scale. Performance Lab is the highlighted bar.

Booster Page Speed Optimizer
Booster Page Speed Optimizer: 57.3%
57.3%
PerfectApps Swift
PerfectApps Swift: 46.7%
46.7%
Sections.design Shopify App Optimization
Sections.design Shopify App Optimization: 46.7%
46.7%
FlyingPress
FlyingPress: 35.6%
35.6%
Flying Pages
Flying Pages: 34.9%
34.9%
Performant Translations
Performant Translations: 26.7%
26.7%
Priority Hints
Priority Hints: 25.9%
25.9%
Cloudflare Rocket Loader
Cloudflare Rocket Loader: 24.9%
24.9%
Image Placeholders
Image Placeholders: 21.1%
21.1%
a3 Lazy Load
a3 Lazy Load: 20.0%
20.0%
All sites
All sites: 19.9%
19.9%
Modern Image Formats
Modern Image Formats: 18.4%
18.4%
Google Cloud Trace
Google Cloud Trace: 17.3%
17.3%
Cloudflare Zaraz
Cloudflare Zaraz: 17.1%
17.1%
Partytown
Partytown: 16.5%
16.5%
Turbolinks
Turbolinks: 15.5%
15.5%
Perfmatters
Perfmatters: 15.3%
15.3%
Enhanced Responsive Images
Enhanced Responsive Images: 15.0%
15.0%
WP Fastest Cache
WP Fastest Cache: 14.8%
14.8%
Performance Lab
Performance Lab: 14.3%
14.3%
Optimization Detective
Optimization Detective: 13.9%
13.9%
WP-Optimize
WP-Optimize: 13.6%
13.6%
Autoptimize
Autoptimize: 13.2%
13.2%
Turbo
Turbo: 12.2%
12.2%
EWWW Image Optimizer
EWWW Image Optimizer: 11.3%
11.3%
Image Prioritizer
Image Prioritizer: 9.5%
9.5%
Embed Optimizer
Embed Optimizer: 8.9%
8.9%
Web Worker Offloading
Web Worker Offloading: 8.9%
8.9%
Speculative Loading
Speculative Loading: 8.3%
8.3%
00.250.500.751.0
Every Performance technology the census can publish, against the web overall (19.9%). The scale is always 0–1.

What else we measure here

53.9%
Google's category score

Google's category score

Sample
n=44
Coverage
93.6%
8.5%
Sites with an AI crawler policy

Sites with an AI crawler policy

Sample
n=47
Coverage
100.0%
98.0%
Agents that can reach the site

Agents that can reach the site

Sample
n=47
Coverage
100.0%
15.9%
Pages an agent can parse

Pages an agent can parse

Sample
n=44
Coverage
93.6%
45.5%
Sites serving an llms.txt

Sites serving an llms.txt

Sample
n=47
Coverage
100.0%
22.7%
Sites with a valid llms.txt

Sites with a valid llms.txt

Sample
n=47
Coverage
100.0%
0
Sites exposing tools

Sites exposing tools

Sample
n=44
Coverage
93.6%

How this was measured