Do not burn tokens watching. Wake your agent only when it matters.
Access signals vs. answer evidence

Readable is not recommended.
Desvela measures both.

The AI-readable Web Census maps how sites expose themselves to AI agents. Check tests your domain. Brand Watch measures what ChatGPT, Claude, Perplexity and Gemini actually cite and recommend.

Check your site → Explore the Census
For teams building agents — and for marketers who need to know what AI says about them.
The census, August 2026 edition

We crawled the top 100,000 domains. Here is what agents actually find.

13
valid ai-catalog.json manifests with entries, across the entire top 100,000 domains checked
See the census →
79.5%
of sites naming GPTBot in robots.txt block it outright — 8,174 of 10,288
See the census →
6.9%
of the top 1,000 sites publish an llms.txt — 69 of 1,000
See the census →
Full methodology, denominators and downloadable data: how we count.
Two different questions

Can it be read? Is it recommended? We never let one answer for the other.

Access signals
Deterministic. A file exists or it doesn't.
  • robots.txt and AI-agent directives
  • llms.txt
  • Machine-readable catalogs (ai-catalog.json, agents.md)
  • Reproducible HTTP checks, re-run any time
Powers: the Census & Check
Answer evidence
Non-deterministic. Ask twice, get different answers.
  • Mentions and position in AI answers
  • Share of voice against named competitors
  • The sources each engine actually cites
  • Repeated sampling and change over time
Powers: Brand Watch
Access signals do not guarantee discovery, citation or recommendation. A domain can score perfectly on the Check and still be invisible in every AI answer — that is exactly why Brand Watch measures the second layer separately.
One path, one product at the end

Explore. Check. Audit. Watch.

01
Explore the Census
See what the top 100,000 domains publish for agents — the baseline every Check result compares against.
02
Check your domain
The Census turned into a diagnostic for one domain: yours. Free, no signup, about ten seconds.
03
Run a brand audit
Being readable is not the same as being recommended. A Brand Watch run asks ChatGPT, Claude, Perplexity and Gemini directly, and reports what they actually say.
04
Watch what changes
Same product, scheduled: the first run is the audit and its baseline, every later run is the watch — a signed alert only when the answer actually moves.
The commercial product

Know the day AI stops recommending you.

AI Brand Watch asks the questions that define your category to ChatGPT, Claude, Perplexity and Gemini, sampled repeatedly so a single lucky answer never counts as the truth, and reports the mention rate, position, sources and competitors each engine returns.

Live now · pay per check, no subscription
AI Brand Watch — track your brand in ChatGPT, Claude, Perplexity & Gemini.
A brand name is enough — the first run audits your baseline, a schedule turns it into a watch. Runs on Apify.
Open Brand Watch on Apify →
Honest measurement

We refuse to sell you noise.

AI answers vary between runs. That is why Brand Watch reports a mention rate across N samples and only alerts past defined thresholds. We call the engines' APIs with web search enabled and always measure the same way, so diffs stay valid.

Alerts, not dashboards.
Built for machines, readable by humans

An alert should lead somewhere.

Every Brand Watch change event is a signed, structured payload: what changed, before and after, which engine, which query, which source. Verifiable with an HMAC signature your systems can check.

A human reads it in Slack. An agent consumes it from a webhook and acts. Same event, same evidence — no dashboard in between.

Read the payload docs →
DESVELA
SIGNED
HMAC-SHA256
CHANGE EVENT
Event № 0117 · HMAC signed
Acme CRM disappeared from Perplexity's answer.
Query"best CRM for small business in Spain" · Perplexity
Mention rate67% → 0% · new source cited: reddit.com/r/…
Delivered07 · 07 · 2026 — webhook · 210 ms
See the webhook payload →

Find out what's true before you pay to find out what's said.

Readable is not recommended. Desvela measures both.
Check your site → Explore the Census