Deep Research

The major services have a separate mode: instead of a quick answer, the service searches for sources, reads them, and assembles a report with links.

The work runs in the background and takes from a few minutes to an hour. You ask the question and come back for the finished text.

How it differs from an ordinary answer

An ordinary answer is built out of what the model absorbed during training. Research mode leans on pages found right now and shows where a statement came from.

The area of use is clear: a market overview, a comparison of vendors, a roundup of requirements by country, recent changes in the rules.

For short questions the mode is overkill. There is no sense waiting 20 minutes for something you would have found with one search.

How to set the task

Name the frame: country, industry, period, the language of the sources. Without it the report gets assembled from the most prominent pages on the internet. Say what has to be in the answer as well: a comparison table, a list with prices, a set of requirements with links.

Ask it to state separately what it could not find. That line shows immediately where the holes in the report are.

Verification stays with you

The links in a report look convincing, but some of them lead somewhere other than promised and some to pages that do not exist. It is the same failure covered in the chapter "Mistakes and Invented Facts", only in prettier packaging.

Open the links behind the key statements and confirm the quotation is there. A report with links nobody opened is no better than an ordinary answer.