What changed, and what we corrected.
Changes to our checks that can change a result, and corrections to reports already delivered, newest first. Client names are never published.
Log
Repeated questions to AI assistants are now separate requests, each counted once. Before, a repeated answer could be counted more than once, which made a result look more precise than it was.
Brand mentions are shown as evidence and no longer scored, because the data cannot tell which business each mention is about.
A crawler refusal is reported only when it repeats while an ordinary browser request succeeds before and after.
Reading ease is scored for English pages only. Pages in other languages are shown as not measured.
When a data source returns an incomplete answer, the report says so and never treats it as a confident negative result.
Two client reports delivered in August 2026 were measured again and corrected editions prepared. The corrections included a brand-mention count that matched a common word, and crawler access that had been read from robots.txt without test requests. [To be provided by the founder: confirm this entry may be published]
Results measured before a change are not compared with results measured after it. How we work