Loading…
Loading…
The failure mode of research with a language model is not that it refuses to help. It is that it helps too smoothly: a confident paragraph, a plausible number, a citation that does not exist. Every prompt below is written against that. They separate what a source actually says from what the model inferred, ask for uncertainty to be marked rather than smoothed away, and force a recommendation instead of a tidy balanced list. Use them when you are comparing tools, mapping an unfamiliar field, checking a statistic that sounds too good, or turning a vague question into something you can actually go and answer.
36 prompts in 5 groups
Mapping an unfamiliar field, scoping the question, and finding somewhere real to begin reading.
Getting the substance out of a document, and reconciling sources that do not quite agree.
Statistics, causal stories, methodology, and tracing a repeated claim back to where it started.
Weighing real candidates against your criteria, from a quick side-by-side to a full vendor evaluation.
Counter-evidence, gaps, and the strongest objection, found before someone else finds it for you.
Send a prompt you keep coming back to. A person reads every one.
Suggest a prompt