Research a market question without mixing up the evidence

Name the decision, audience and workflow

Open the assistant and state the decision the research should support. “Research restoration software” is a category; “Which field-documentation problem should our product page explain?” is a decision.

Research how small restoration contractors document jobs. Identify recurring difficulties described in public discussions, alternatives mentioned and questions our product page should answer. Keep discussion evidence separate from vendor claims. Include original URLs and dates where available.

Read different sources for different reasons

A buyer discussion can show how someone describes a problem. A vendor page can show how a product presents its solution. An industry article can provide background. These sources should not be counted as interchangeable proof of demand.

Ask for a source-level summary before accepting a broad theme. If several search results point back to the same original post, treat them as one observation rather than several independent buyers.

Change the search when the vocabulary is wrong

If the first pass finds product promotions instead of operational questions, use the language of the work: missing photos, repeat data entry or report handover. Broaden one variable at a time and ask for adjacent findings to be labeled separately.

A source that cannot be opened remains an access gap. An accessible but irrelevant result is a relevance problem. Neither is the same as evidence that the market has no need.

End with a decision and its supporting material

Ask for a report organized by finding, rather than a list of URLs:

For each finding, show: the buyer’s stated problem, original source,
date where available, supporting observations, contrary evidence and
what remains unresolved. End with the three questions our product page
should answer. Keep vendor claims separate from buyer discussions.

Open a source for each major theme and check that the report contains actual observations. Multiple search results pointing to one post count as one observation. Choose a follow-up such as interview questions, an article brief or a company search. Use public-post research when the original discussions are the main output.