LocalBridge

Research template - 3 to 8 tabs

Extract a table from pages prompt template

Turning pages into a table is the task where invention does the most damage, because a guessed cell looks exactly like a sourced one.

The prompt

Build one table from these pages. They are structured differently, so normalise as you go.

Columns: [LIST THE COLUMNS YOU WANT]
One row per: [THE THING EACH ROW REPRESENTS]

Rules:
- If a page does not state a value, write "not stated". Never estimate, never infer from similar products.
- Keep original units and formats. Do not convert. If units differ across rows, add a units column.
- Quote short values exactly as written on the page.
- After the table, list every cell you were unsure about and say what made it ambiguous.

Then tell me which column has the most gaps, and what that suggests about these pages.

Why this prompt is written this way

'Not stated' as a required value is the single most important line. An assistant asked to fill a table will fill it, and the plausible guess is indistinguishable from the sourced figure once it is in a cell.

Banning unit conversion removes a quiet source of error, and an extra units column costs nothing compared with a silently wrong number.

The uncertainty list at the end tells you which three cells to verify, which is the difference between a usable table and one you have to check entirely.

Getting more out of it

  • Define the columns precisely. 'Price' produces a mess; 'monthly price for the cheapest paid tier, per seat' produces a column.
  • Ask for the table in markdown if you are pasting it into a doc, or as CSV if it is going into a spreadsheet.
  • Spot-check the three cells that matter most to your decision against the original pages.
  • A column full of 'not stated' is a finding about the market, not a failure of the prompt.

Works best with

3 to 8 tabs open at once. Send them together rather than one at a time, so the assistant can compare them instead of forgetting the first by the time it reads the last. Sending tabs to Claude explains the options.

Common questions

Can it handle pages with different structures?
That is what it is for. The normalising happens because you define the columns, rather than letting each page's own layout drive the shape.
How accurate is the extraction?
Good for values stated plainly on the page, weaker for anything in an image, a complex table or a footnote. The uncertainty list is where the weak cells surface.
Can I get the output as a spreadsheet?
Ask for CSV and paste it into your spreadsheet tool. Keep the 'not stated' cells as text rather than converting them to zero.

Other templates

Stop pasting tabs one by one

LocalBridge sends the pages you have open into Claude or ChatGPT in one click, so this template gets the sources it needs.

See how it works