Quicktion

AI Property Extraction by Provider

AI extraction lets the model read each email and fill destination properties with values it pulls from the body, subject, sender, or signature. Each provider exposes a different set of field types, but the configuration UX is the same: pick the properties you want filled, and AI fills them on every save.

How extraction works

  1. You select properties on the destination edit page (in the "Fill these properties from the email" section).
  2. When an email arrives, AI reads the body and returns one value per property in a structured JSON response.
  3. The values are coerced to match each property's type (text, number, date, etc.) and written to your destination alongside any statically-mapped fields.
  4. If the model can't find a value for a property, it's left blank — never guessed.
  5. The activity feed on your destination shows which properties were extracted and which were skipped.

Extraction does not override fields that are already set by your static mappings (like "Subject → Card Name"). Static mappings always win for the fields they cover.

Notion

Supported property types:

  • Text (rich_text, title, url, email, phone_number)
  • Number
  • Date
  • Checkbox
  • Select — AI must pick from your defined options (case-insensitive match)
  • Multi-select — AI can pick zero, one, or multiple options
  • Status — same as select

Not supported (filtered from the UI):

  • Formula, rollup, created_time, created_by, last_edited_time, last_edited_by, unique_id, verification, button — read-only or computed
  • Relation, people — AI doesn't know your row/user IDs
  • Files — AI doesn't generate file URLs

Airtable

Supported field types:

  • Single line text, multi-line text, rich text
  • Email, URL, phone, barcode
  • Number, currency, percent, rating, duration
  • Date, dateTime
  • Checkbox
  • Single select, multiple selects

Not supported:

  • Multiple record links, multiple collaborators, single collaborator
  • Multiple attachments
  • Computed fields (formulas, rollups, lookups)

Linear

Linear has a fixed schema, so AI extraction targets are curated to the 5 fields that make sense:

  • Priority — No priority, Urgent, High, Medium, Low (always available)
  • Status — pulled from your team's workflow states
  • Assignee — pulled from your team members
  • Project — pulled from your team's projects
  • Labels — multi-select from your team's labels

Labels merge with any static label IDs you set on the destination — AI labels are added on top, not replaced.

Static defaults (priority, status, assignee, project) take precedence over AI extraction. If you've set a default Status, AI won't change it. To let AI control a field, leave that default blank.

Google Sheets

Sheets extraction is column-driven. In the column mapping section above the AI Processing section, change a column's role to AI extraction and that column becomes an extraction target. The AI section automatically picks up every "AI extraction" column on the sheet.

The column header becomes the property name AI sees in the prompt. Tagging a column called "Priority" tells AI "fill this with the priority you find in the email."

Column positions are tracked via developer metadata, so renaming or reordering columns won't break extraction.

Trello

Trello extraction uses the Custom Fields feature on a board. Custom Fields is included with Trello Standard workspaces and above. If your board has Custom Fields, they'll appear automatically in the AI section when you open the destination.

Supported Custom Field types:

  • Text → free text
  • Number → numeric
  • Date → ISO 8601 date
  • Checkbox → boolean
  • List → AI picks from your defined option names (case-insensitive)

Trello list fields are strictly single-value — multi-select extraction doesn't apply.

If your board has no Custom Fields, the AI section will only show body transformation. Add fields to your board (Board menu → Custom Fields → New field) and they'll show up on the next page load.

Best practices

  • Start with 1-3 fields, then grow. Easier to validate that extraction is working before adding more.
  • Use selects with explicit options when you can. "Status: Triage / In Progress / Done" is more reliable than free-form text.
  • Add an extraction guidance prompt if a field needs context. See the third input on the destination edit page.
  • Watch the activity feed. The AI badge on each row shows you which fields were extracted and which were skipped, so you can iterate.
Was this helpful?