Tables are available on PRO plans, and running table analysis costs tokens.
When to use this workflow
Use this workflow after a table has returned enough results that you need to validate them, organize them, and share them. The goal is not just to collect AI output, but to turn the output into something a human reviewer trusts.
This stage is where you:
inspect the returned answers
focus on high-priority rows and columns
mark trustworthy outputs as verified
flag questionable outputs for follow-up
export the finished result set for downstream work
Review results in the grid first
The main table view is the fastest way to scan a large result set. Start by looking across rows and columns for:
blank or error states
low-confidence answers
rows that need closer review
columns whose outputs are inconsistent or unexpectedly broad
If the table is large, use sorts and filters before opening individual cells. This usually makes review much faster than reading cell details one by one in random order.
Use verification and flagged states deliberately
The table UI supports both verified and flagged review states.
Use verification when a cell looks reliable and you want to record that someone has reviewed it.
Use flagging when a cell needs more attention, correction, or escalation.
The interface also supports controls for Highlight Verified and Highlight Flagged, which helps you visually separate work that is complete from work that still needs attention.
Filter the review set
The review controls support both sorts and filters:
Sorts
Use Sorts to reorder the table by source or by structured output columns. This is useful when you want to group rows by a status-like answer, a date, or some other normalized value.
Structured filters
Structured filters let you narrow the table using typed operators that match the column output type. This is helpful for questions like:
show only rows where an enum column is a specific value
show dates after a cutoff
show rows where a numeric score crosses a threshold
show only empty results that still need work
AI filters
AI filters are helpful when you want to screen for rows that match a more judgment-based condition, such as whether a generated answer mentions a specific concept or satisfies a narrative condition.
Change the display while reviewing
The table actions include several display controls that are useful during validation:
Show Hidden Columns
Show Hidden Rows
Wrap Text
verification highlighting
flagged highlighting
Use these before exporting so you can inspect the same result set in more than one way without rebuilding the table.
Open cell details when you need evidence
When a cell needs closer review, open the cell detail view to inspect the fuller response. Cell detail can include:
the full response text
reasoning
confidence score
citations and links
edit controls
verify and flag actions
This is the right place to confirm whether the generated answer is supported by the underlying evidence and whether it should remain as-is.
Correct outputs before sharing
Do not treat every generated answer as final by default. Before export, use the review workflow to identify:
cells that need manual editing
cells that should be rerun after prompt changes
cells that should remain flagged because the source support is weak
The value of the table comes from combining AI scale with human verification, not from skipping the review pass.
Export the finished results
When the table is ready to share, use the export actions available in the table controls.
The current UI supports:
Export as Excel (.xlsx)
Export as CSV (.csv)
Use Excel when you want a richer spreadsheet handoff and CSV when you need a plain structured export for another tool or process.
Save reusable structures for future work
If the table layout itself is worth reusing, use Save As Template after review. This preserves the structure for future projects so your team does not need to rebuild the same column set from scratch.
Saving a template is especially useful when:
the columns represent a repeatable review framework
you expect similar matters to recur
your team wants a consistent starting point across projects
After review and export
Use the verified results to support charting, drafting, or team review, and reuse saved templates for similar future matters. If too many cells remain uncertain, return to the table setup and rerun the relevant parts before treating the output as final.