AI Columns

Run a prompt across every paper in a table to score, classify, or summarize them - great for literature reviews.

AI columns let you run a prompt across every paper in a table - in your library or in search results - and put the answer in its own column. They are ideal for screening and extraction in literature reviews.

Adding an AI column

  1. Click Add Column (or AI Column on mobile).
  2. Type what you want (for example "Relevance to CRISPR delivery, 1-10").
  3. Choose the column type:
    • AI - a free-form text answer (summaries, extractions).
    • Score - a number from 1 to 10.
    • Yes/No - a yes/no decision, with an optional justification.
    • Plain - a normal column you fill in yourself (no AI).
  4. Pick the Model and the Sources the AI may read.
  5. Press Enter to create the column.

You can also start from a saved template via Manage Prompts (see Prompts settings).

Sources the AI can read

  • In search: Title and Abstract.
  • In your library: Abstract, Notes, Highlights, and PDF Content.

Toggle each source on or off in the Sources menu.

Generating values

  • Auto mode generates values automatically, including for new results as they arrive.
  • Manual mode lets you generate on demand - a single cell, or the whole column with Generate All.

Values stream into the cells as they finish, and a progress indicator shows how many remain. Rows are processed in parallel, and a failed cell can be retried without redoing the rest.

Managing columns

  • Edit a column to change its prompt, type, model, or sources, then regenerate.
  • Delete, reorder (drag the header), resize, sort, and filter columns like any table column.
  • Yes/No values show as colored pills, scores show as numbers, and AI text can include citations.

Each generated cell is a paid AI request unless you use a free-tier or local model. Running a prompt across hundreds of papers can add up, so start with a small batch to check the output first.