paulpan05 Contested-source diversity checklist
Questions that characterize a country, culture, or other group — whether describing it alone or comparing it to another — are easy to answer with whatever sources come up first in a search, and those sources are usually produced by researchers or analysts looking at the group from outside it. That's a real gap: a Pew survey measuring another country's attitudes using an American-designed instrument, an index built by European academics, or a Western bank's economic outlook on a country, is a valid data point, but it isn't the same as domestic scholarship, domestic surveys, domestic official statistics reported by domestic press, or domestic public discourse from within the society being described. Presenting the outside view alone as "the answer" quietly launders one perspective's framework into a universal one, and the person asking has no way to tell that happened unless it's flagged.
The same gap shows up outside country/culture/generation questions whenever there's a live dispute with interested sides: a contested historical narrative, an organization's disputed reputation, or a product/technology controversy. "Is a national airline's safety record really as strong as it claims" answered only from the airline's own PR, or only from its harshest critics, has the identical blind spot as a culture question answered only from outside analysts — it's missing the other side's own data and reasoning. Treat "group" below as covering both an actual demographic/national group and, in these adjacent cases, each side of the dispute (the accused party, the accuser, the industry, its critics) — a company counts here when the dispute is over its record or reputation (e.g. "is OpenAI's safety record as strong as it claims"), just not for an ordinary factual question like its quarterly earnings.
Run through both of these before treating an answer as complete — not after, when asked to check:
1. Source-origin check. For each group or side being described (one in a standalone description, each side in a comparison or dispute), is there at least one source produced from within that group or side — its own academic literature, its own surveys or official statistics agencies, its own press, technical documentation, or public discourse — rather than only outside-looking-in analysis? If a genuine search attempt doesn't turn one up, don't quietly drop the issue — say so in the answer itself, the way you'd flag any other real gap in the evidence. "I could only find outside analyses of X; I didn't locate domestic/first-party sources to check it against" is a more honest answer than silence, and it tells the person what the answer's blind spot actually is. Note that a group's own state or official media, or a company's own PR, is an inside source for confirming facts and figures, but it is not a substitute for independent domestic scholarship, business press, or public discourse when the question calls for how the group or party itself sees or debates the topic — flag that distinction when it matters.
2. Primary-source check on cited numbers. A web search tool's result text is a synthesized summary written by a model, not the source document itself. Before stating a specific number pulled from search results, fetch the page it supposedly came from and confirm the figure is really there. This matters most for numbers that sound clean or persuasive — round percentages, dramatic gaps between groups — since those are exactly the kind of figure that's easy to round off or misstate when summarizing.
Both checks cost one extra tool call each. Skipping them to save a step is the specific failure this skill exists to catch: a plausible-sounding, one-sided, unverified characterization of a group is worse than a slower one that says plainly where its evidence is thin.