Skip to main content

Run AI analysis across rows and columns

Explain how to run analysis on a table and understand the token-consuming parts of that workflow.

Written by Caleb Harris

This workflow is available on PRO plans and costs tokens to run.

When to use this workflow

Run a table when you already have the basic structure in place and want &AI to fill the cells across the selected rows and columns. This is the right step when:

  • the table already contains sources and at least one useful column

  • you want answers generated at scale instead of one row at a time

  • you want to resume a partially completed table

  • you need to apply filters or rerun work after refining the table

Make sure the table is ready to run

Before selecting Run All, confirm that:

  • the rows represent the sources you actually want analyzed

  • the columns ask the right questions

  • output types and constraints match the kind of answer you want back

  • any web-search or deep-research settings are intentional

Start the run

From the table detail page, use the main table action button:

  • Run All appears when the table has incomplete work that has not been processed yet

  • Resume All appears when some work is already complete and some is still incomplete

In both cases, &AI starts processing the unfinished cells for the current table. The table UI is built to show partial-progress workflows, so resuming is a normal part of working with a large table.

How runs process rows and columns

The run operates across the intersection of:

  • rows, which are the selected documents or products in the table

  • columns, which are the prompt-backed or search-backed analysis instructions

Each cell moves through a status lifecycle such as pending, processing, success, or error. The workflow is designed so you can inspect the table while work is happening and return later if the run is still in progress.

Use the main analysis controls

The header and actions area expose the controls you use most during and after a run:

  • Run All or Resume All to process unfinished work

  • Cancel to stop an active table-level run

  • Sorts to reorder the table by source or by structured result columns

  • Filters to narrow what you are looking at

  • row and column visibility controls for hidden items

  • wrap-text controls so long answers are easier to scan

These controls are especially helpful when a run returns a large number of outputs and you want to focus on a subset before reviewing everything.

Use structured filters and AI filters

The Filters control supports two different filter patterns:

Structured filters

Structured filters are tied to a specific column and adapt to the column's output type. Depending on the output type, the available operators can include options such as:

  • Includes

  • Equals

  • Does Not Equal

  • Is Greater Than

  • Is On Or After

  • Is Empty

  • Is Any Of

This is the better choice when the column output is already normalized enough to filter deterministically.

AI filters

AI filters let you choose a column and write a natural-language filter query that describes which rows should match. After you create the filter, you run it from the same filter panel, and the filter can become stale if the prompt changes or the table changes.

Use this when a simple exact-value filter is not enough and you need semantic matching over the generated output.

Add or refine columns while the workflow evolves

If your workspace allows prompt creation, you can add new prompt-backed columns or update your table structure before the next run.

This is useful when the first pass shows that:

  • a question is too broad

  • you need a new output type

  • you want to split one large question into several narrower columns

  • you want to add a search-backed column for additional evidence gathering

Cancel or resume when needed

Longer runs may not finish in one uninterrupted session. The interface supports cancellation and later resumption so you can stop the active work and return when you are ready.

If the table already contains some completed cells and some unfinished cells, use Resume All instead of rebuilding the table from scratch.

After the run finishes

After cells populate, review confidence and reasoning, verify reliable cells, flag outputs that need follow-up, and apply sorts or filters to focus on the most important answers. Once results are trustworthy, export the table or save the structure as a reusable template.

Did this answer your question?