What we measured, and what it means
Each entry is a study, not a blog post: a number with its sample, the method that produced it, and the limitation that qualifies it. If a finding cannot survive those three, it does not get published here.
The first study is not published yet
We publish studies from live engagements, and the first engagement has not closed. Rather than fill this page with placeholder articles, it stays empty until there is a number we measured ourselves and a client we can describe without naming.
What is already published and needs no client: the exact twenty prompts and ten checks behind every audit.
The format
What every study will contain
Written down in advance so it cannot be quietly relaxed when a result is disappointing.
A number, with its sample
Every headline figure is printed next to what it was measured on — how many businesses, over what window. A percentage with no denominator is not published.
The method, written out
How the data was collected, what was excluded and why. Close enough that another person could repeat it and check us.
The limitation
What the finding cannot tell you. Studies about AI answers are especially prone to over-reading: models change version, and a sample is a sample.
The date
An answer-engine result has a shelf life measured in weeks. Every entry carries the month it was collected, and is not silently refreshed.
Want the data before the write-up? Every audit hands over the raw answers, dated and version-stamped, whether or not the engagement continues.
See what an audit costs