The Problem
Most search results return a chain: Blog A cites Blog B who cites Blog C who vaguely references "a study." By the time you read it, the original data has been paraphrased three times and the sample size, methodology, and date are gone. You're reading marketing dressed up as research.
The Fix: 3 Query Types
Every research topic gets at least 3 different queries. Each one surfaces different sources.
1. Primary Source Hunt
This is the one that cuts through the chain. Ask for the original researchers, not articles about the research.
2. Skip the Usual Suspects
Every industry has a handful of sites that dominate search results. They all cite each other. Tell the engine to skip them and look deeper.
Who to skip by industry:
SEO / Marketing:
Medical / Health:
Look for instead:
Legal:
Look for instead:
Political / Policy:
Look for instead:
Financial:
Look for instead:
3. Chain Tracing (The Myth Buster)
When you see the same stat everywhere, ask: who actually ran the study?
What came back: The claim originated from a single company's marketing blog with undisclosed methodology. It was cited by 4 other blogs, each adding more certainty to the language. A separate study from Search/Atlas found no correlation at all.
Optional Queries (When You Need More)
4. Contradictory Evidence
The most valuable findings often go against the mainstream narrative.
5. Freshness Filter
Limit results to a specific time window when you only want current data.
- Last 30 days — very narrow, may return nothing for niche topics
- Last 90 days — good balance for active fields
- Last 6 months — use for slower-moving topics
- No filter — gets the most results but mixes old and new
What to Include in Every Request
- The topic — be specific ("interventional radiology" not "medicine")
- The timeframe — "last 60 days", "last 6 months", "2025-2026"
What You Don't Need to Say
- "Please find citations" — it always does
- "Format it nicely" — the template handles that
- "Use primary sources" — the queries already do that
- "Be thorough" — it runs multiple queries automatically
Quick Examples
Reading the Output
- Primary source = the organization that ran the study. This is what you cite.
- Secondary source = a blog or article describing someone else's study. Use it to find the primary, don't cite the blog.
- "Undisclosed methodology" = they gave a number but didn't explain how they got it. Treat with skepticism.
- "Marketing claim" = a vendor published a stat to sell their product. Not research.
- Contradictory findings = often the most valuable part. If everyone says X and one study says not-X, that's worth investigating.