Lehvel Observatory

The AI platforms your customers use are describing your business. We check what they say.

The Lehvel Observatory is our own software. It asks those platforms the questions your customers ask; which ones answered on the latest run, and any it could not reach, are shown on /live. It keeps each answer as given and checks every claim against what the business has confirmed. You see where the AI is right, where it is wrong and where it has nothing to go on.

Lehvel's own software

A barcode label on a carton being read by a red scanner line.

Of 135 statements, 64 could not be checked.

From 25 answers about 4 businesses, September 2026. Telling missing evidence apart from a confirmed contradiction is the core of the method.

135claims

each checked against the record

  • 69SupportedMatched the working record.
  • 2In conflictDisagreed with the working record.
  • 64UnknownNo recorded fact covered it. Unknown is never counted as wrong.

Was the business named in the answer?

  • 13 of 13Asked about the business by name
  • 6 of 12Asked for the best provider of its service

Collected 2026-09-25Published 2026-09-29

Read the method
How answers are checked, and the limits

  1. The facts

    What the business has confirmed, each with its source.

  2. The question

    The same question goes to each AI. The answer is saved as given, with the time it was asked.

  3. The check

    Each claim is marked supported, in conflict or unknown.

  4. The next look

    Asked again later, to see what changed.

  • Each answer's factual claims were pulled out and checked one by one against the record.
  • Results are pooled across the four businesses. No business is named and no per-business figure is shown.
  • The record was a working one. We built it from each business's own site, listings and registry. The owners did not confirm it.
  • A small sample: 4 businesses and 25 answers. It describes this study, not businesses in general.
  • Gemini failed on 7 of 8 calls. Failed calls are excluded, so the results come almost entirely from ChatGPT, Claude and Perplexity.
  • One run, on one day.

Lehvel does not control the AI companies. The Observatory records what they say and where it came from. It gives no scores and promises no rankings or citations. It runs today as a private pilot and is not generally available.

What you get in Business Truth Diagnostic.

  • Every answer, word for word, with the date and the AI that gave it.
  • A claim-by-claim check against your confirmed facts.
  • A list of the wrong answers and the sources they most likely came from.

What runs today.

  • Asks the AI platforms your customers use about a business and keeps every answer as it was given.

    In use

  • Checks each claim in an answer against the business's record: supported, in conflict or unknown.

    In use

  • The September 2026 study: four businesses, 25 answers, 135 claims, with its method and limits.

    Published (Published research)

  • A fix counts only when a later answer confirms it. Proven on a public listing.

    Experiment (Running experiment)

  • Scheduled checks that ask again on their own.

    Planned

Want this run on your business?

Business Truth Diagnostic

Get Upstream

One note a week from me: one job I took off an owner's plate, and what came of it.

My posts reach 17,000+ followers across our channels. The weekly email is called Upstream.

Weekly. Unsubscribe anytime. Privacy