OSINT with an AI agent: a repeatable workflow

Open-source research is ten percent tools and ninety percent discipline. Here is a workflow an agent can carry end to end — with the caveats that keep it honest.

Anyone can run a search. The hard part of OSINT is staying organised, separating what you know from what you assume, and writing it up so someone else can follow it. That is exactly the part an agent can hold for you.

Step 1 — Define the question and the subject

Start with a single, exact identifier: a domain, a username, an email, a company name. Resist the urge to “look around”. Scope is what keeps the work useful and legal.

Step 2 — Collect from open sources

• Public records and search results, each cited with its source.
• Breach corpora for exposure checks on the subject’s own accounts.
• Technical surface data for anything the subject publishes themselves.

Step 3 — Correlate, don’t guess

Two similar usernames are not the same person. The agent should present each match with the evidence that links it and flag how confident it is, so a common first name does not become a false identification.

Step 4 — Write it up

Ask for a structured report: what was checked, what was found, what each source was, and what remains unknown. An exported file you can hand over beats a wall of chat text every time.

Step 5 — Stay inside the lines

• Only investigate subjects and targets you are authorised to look into.
• Use open sources — never attempt to access anything private.
• Keep the raw sources; your conclusions are only as good as what backs them.

The agent does not replace judgement. It removes the tedious parts — the tab-juggling, the note-taking, the formatting — so your judgement has somewhere to land.