AI Seedsai.seeds.cn

FIELD NOTES · Workflow

An AI research workflow you can reuse every day

Separate search, source selection, fact-checking and writing so a complete-looking AI answer is not mistaken for reliable research.

AI Seeds Editorial Team2026-07-2410 min read
01

Begin with a defined question

'Research the AI industry' is too broad to guide a decision. Add a region, period, subject and decision, such as 'What are the main uses and barriers to generative AI adoption among Chinese SMEs in 2025–2026?'

State how the result will be used: a market brief, investment assessment, sales proposal or article plan. That purpose determines the depth of evidence required.

02

Use three levels of sources

AI can collect leads, while you assess source quality.

  • Primary sources: legislation, government data, original papers, company announcements and official product documentation.
  • Secondary sources: established media, research institutions and industry reports with a stated method.
  • Leads: forums, social media and reposts without original sources; use these to discover directions to investigate.
03

Build a fact table before writing conclusions

Record the original wording, link, date, definitions and scope for each fact. Explain differences between conflicting sources instead of averaging incompatible figures.

Separate confirmed facts, evidence-based inferences and unanswered questions. When new evidence appears, you can update the relevant finding without rebuilding the whole report.

READY TO USE

Before you start

  • Define a time range and scope
  • Prioritize primary sources
  • Save original links
  • Check publication dates
  • Distinguish facts from inferences
  • Keep track of unknowns