guhcostan Research Methodology
Core principle
Never answer from training data alone. Always fetch real sources. Training data is stale. Prices change. Products get recalled. Companies fail. The only trustworthy answer is one backed by content fetched today.
Source hierarchy
| Priority | Source type | Why |
|---|---|---|
| 1st | Specialist review sites | Deep expertise, tested methodology |
| 2nd | Community forums (Reddit, Reclame Aqui) | Real user experience, unfiltered |
| 3rd | Official sources (manufacturer, gov) | Authoritative but potentially biased |
| 4th | News articles | Current events, not always deep |
| Avoid | SEO content farms | Thin, keyword-stuffed, unreliable |
Search strategy
Run multiple queries per topic — not just one:
- Broad:
"best SUV 2025" - Specific:
"SUV reliability data 2025 Brazil" - Community:
"SUV Reddit recommendations family" - Critical:
"SUV problems complaints 2025"
One query = one perspective. Multiple queries = real picture.
Reading sources
Use WebFetch — not just search snippets.
A snippet says: "rated best in class" The full article says: "rated best in class but has known transmission issues after 60k km"
Always fetch. Always read.
Output structure
Every research output must include:
- TL;DR — recommendation in 2 sentences
- Why — top 3 reasons with data
- Trade-offs — honest downsides
- Alternatives — ranked options with key differences
- Next steps — what to do now
- Sources — URLs of everything consulted
Gotchas
- Recency: filter for last 12 months on fast-moving topics (tech, finance, crypto, telecom)
- Region: Brazil prices, availability, and regulations differ from US/EU — search in Portuguese first
- Reclame Aqui: always check for services and products sold in Brazil — it surfaces real complaints
- Conflicts: if two sources disagree, surface both — don't silently pick one
- Gaps: if you couldn't find reliable data on something important, say so explicitly
- Professionals: tag decisions with legal, medical, or tax implications — recommend a specialist