What AI does well
- Reading a lot, fast. Dozens of sources in a few minutes, even in languages the team doesn’t speak.
- Summarizing and organizing. From scattered material to a structured picture by theme, competitor and channel.
- Comparing. Putting offerings, prices and communication side by side.
- Keeping pace. Redoing the reading when the market changes, and saying what’s different.
Where it goes wrong
- It invents with confidence. A language model can produce plausible but false numbers and names if it isn’t anchored to sources.
- It uses outdated knowledge. What the model “knows” stops at the date of its training.
- It loses sources. A summary without references can’t be verified.
- It generalizes. It tends to describe “the market” as an average, losing the differences between countries, channels and segments.
The rules for using it well
1. Separate research from reasoning
First you find the sources — articles, data, pages —, then the model reasons on those. A model answering “from memory” isn’t research.
2. A source and a date on every piece of information
Every fact must be able to say where it comes from and when it dates from. That’s the difference between an analysis and a convincing piece of text.
3. Search in the market’s language
A market is read through its own words and its own sources: local press, local forums, local reviews.
4. Flag estimates and gaps
When a figure isn’t public, say so. When a number is an estimate, label it. A “not available” is information; an invented number is a risk.
5. Have the work checked
A second pass that looks for weak points, a score on the quality of the summary, cross-checking multiple sources: checks cost little compared with a wrong decision.
6. Keep people where they’re needed
AI does the gathering and summarizing; the judgment — what matters, what to do — stays with those who know the business. And when you need representative primary data, you still need research with real people.
