Skip to main content

AI research

Research each record with an AI prompt and save the result.

AI research · Intelligence · Accepts companies or people

Runs an AI prompt against each record and saves the answer to a field you name. Later steps can read that field.

When to use it

Use it to compute something no filter can — qualify fit from a website, summarize what a company does, estimate a segment, or extract a data point from research you fetched earlier.

What it does

For each record, it runs your prompt and writes the result to an output field. It only researches and saves — it doesn't drop or route anything. To act on the result, follow it with a Filter or Route on the field it produced.

Configure it

  • Prompt — what to ask about each record. Type @ to insert fields from earlier steps, like @Page content or @Company domain. See Reference earlier steps with @-mentions.

  • Output name — the field the answer is saved to. This is the name you'll @-mention downstream, so make it clear (for example, fit_score).

  • Output type — the kind of value to save: text, number, yes/no (boolean), or a pick from a set of options. Choosing the right type lets later steps treat the result correctly — a number or yes/no result can be compared directly in a Filter or Route condition.

  • Model — choose the depth of research: fast, balanced, or advanced.

Cost

  • Fast: 0.1 credits per record

  • Balanced: 0.5 credits per record

  • Advanced: 1.5 credits per record

When the prompt includes large referenced content — such as a fetched web page — the cost is 5× that rate. See Credits in workflows.

Gotchas

  • The field name a later step reads must match the output name here exactly.

  • If the prompt references a field that's empty for a record, the answer will reflect that. Fetch or filter first so the prompt has something to work with.

  • Ask for one clear thing per step. Chain two AI research steps rather than asking for several unrelated outputs at once.

Related Articles
Did this answer your question?