Labs / Experiments · Updated 2026-10-09
Audit Lab
Audit Lab is a website check that ties every finding to something it measured, and says could not check when it could not. In October 2026 one report broke that rule.
Running experimentBuilt and running, being proven.
- Web address
- Probe
- Findings
- Opportunities
- Priority
What we asked and what we found
Every entry answers the same eight questions, in this order.
Question
What can software measure about a website, and how well AI can read it, from a web address alone, and how should it report what it could not check?
Why it matters
Many website audits show a spinner and a score without saying what produced it. We wanted each finding to point at something the check saw, and a check that did not run to read could not check, never pass.
State
Running experiment Built and running, being proven.
What we built or tested
- Page-by-page checks of a site, each finding tied to something measured.
- Checks of how a site describes itself to search engines and how easily AI can read it.
- Improvement opportunities drawn from what the check saw, in a shareable report and a PDF.
- A source rule: a guess is never presented as a measurement.
Evidence
- How a check runs
- Web address, measured findings, gaps and opportunities, priority
- Refusals
- No fake progress, no guessed rankings or demand, no AI score without a measured basis
- Failure, October 2026
- A report told a business its website did not load. The site was up. The check tried one address, and the other answered our crawler with a bot check.
What we learned
- An honest could not check earns more trust than a confident guess.
- A blocked crawler and a dead website look the same unless the check is built to tell them apart.
Limitations
- The source rule failed in that report: a blocked crawl was filed as a measured finding. The fix is written and under review, not yet released.
- Since October 9, 2026 the check has no public entry point on lehvel.com. Existing report links still work.
What changes next
Publish what each check measures and what it cannot see, including blocked and unreachable sites.
Failed tests are published as results. A state on this page describes today, not plans.
Next in Labs
- Next: The AI Observatory
We would rather show no number than relabel ordinary search data as an AI score.
Research
Get Upstream
One email a week: what we asked the AI, what it got wrong, and how we checked.
Our research reaches 17,000+ followers across our channels. Upstream is the weekly version.
Weekly. Unsubscribe anytime. Privacy