RESEARCH & METHODS

Evidence first.
Claims second.

Our reports make it clear what we observed, what we think it means, and what still needs testing. This page shows the method, one evidence record you can inspect, and the published work it is applied in.

THREE LABELS ON EVERY FINDING

01

Confirmed finding

Directly supported by a recorded answer, source, or test. The report states the conditions and limits of that observation.

02

Reasonable hypothesis

A possible explanation supported by some evidence. It remains a hypothesis until the relevant test supports it.

03

Unsupported claim

A statement without enough evidence. It should not be presented as an established ranking factor or a promised result.

AN EVIDENCE RECORD YOU CAN INSPECT

“Named” and “cited”
answer different questions.

TEACHING EXAMPLE: NOT A CLIENT RESULT
QUESTION
Which providers structure procurement documents for AI review?
OBSERVED ANSWER
“Provider A is one option to consider for document structuring. Provider B is also commonly listed.”
CITED PAGE
An independent review website. Not Provider A’s own site.
FINDING

CONFIRMED Provider A was mentioned, and recommended as one option.

CONFIRMED The AI answer did not cite Provider A’s own website. It cited a review site instead.

LIMITATION

This is one AI system, on one date, asked once. It does not tell us whether the same answer comes back next time, or why the AI system used the review site rather than Provider A’s own page.

HYPOTHESIS A likely reason is that the review site was readable to the AI system and Provider A’s page was not. That is a possible explanation, not a finding. A page-readiness test would confirm or rule it out.

Each row is labeled on its own. A confirmed finding in one row does not turn a hypothesis in another row into a finding.

A published record also carries the author, date, AI system, access method, prompt, and repeat count, so anyone can check the observation against the conditions.

See the full research record format

A RECORD THAT CAN BE CHECKED

What a useful test report includes.

01

Test conditions

Questions, dates, AI systems, access methods, repeat count, and relevant settings.

02

Observed output

The answer text, source links, brand mentions, and recommendation context.

03

Practical limits

What the sample can support, what it cannot establish, and what requires another test.

A clear next step

See the method applied to your question.

Tell us which business question your research needs to answer.