My agent tagged 261 businesses "no website". One random check said otherwise.

Description and use

The author used AI to compile a list of 261 businesses claiming no website. A single random check revealed the status column was misleading, leading to a shift to per‑row verification before outreach.

Problem

An AI‑generated list of businesses claimed to have no website, but a random verification revealed that the status column was misleading, undermining trust in the entire dataset.

Outcome

The author discovered that a single random verification can invalidate an entire AI‑generated dataset, prompting a shift to per‑row verification at the point of use and marking the status column as weak.

Workflow

• 1. AI assistant compiles list of 261 businesses with status column. 2. Author performs a random check on one row using a separate AI assistant. 3. Check finds a WhatsApp link in the website slot, contradicting the "no website" status. 4. Author concludes the status column is unreliable and marks the entire list as weak. 5. Future outreach messages are generated only after re‑checking each row immediately before sending.
• 6. The list becomes a queue where each row is verified at the point of use.

Evidence

The agent was not exactly wrong about the thing I cared about. The clinic still has no real website; a WhatsApp link is not a page that lists services and hours.

Source

That was enough. The check failed, so the column failed, for all 261 rows.

Source

The header of the file says the status column is weak. Each row is measured again right before it is used...

Source