Try this skill in Strawberry

/extract-web-data

Extract Web Data

Strawberry's harness is built for the web, so extraction that takes an hour elsewhere often takes a minute here. Many sites load their data through their own network requests, and reading those directly can pull thousands of rows quickly and cheaply.

Setup

  • What each row is, which fields they need, and roughly how many rows.
  • What the data is for. That decides which fields matter and how deep to go.
  • Where it should end up. If there's an existing sheet or database, look at its columns first.

Worth knowing

  • Before scraping the rendered page, check whether the page's own network requests already return clean JSON.
  • Cover the whole scope: pagination, infinite scroll, load-more buttons, and detail pages.
  • Show a sample of 10 to 20 rows before the full run. At high volume, a small mistake repeats thousands of times.
  • Keep the source link for each row, and keep "not found" separate from "no".
  • Never fill gaps with guesses to make the dataset look complete.
  • For spreadsheets, use the spreadsheets skill.

Copying a directory or a product catalog into a spreadsheet is the kind of job that eats an afternoon. Most AI tools don't help much, because they read pages the way a person does: one screen at a time.

Strawberry's agent is built for the web. Many sites load their data through their own requests in the background, and Strawberry can read those directly. A job that takes an hour in a tool that wasn't made for the web often takes a minute here, and thousands of rows stay cheap.

What it is good at

Anything that repeats on a website and belongs in rows:

  • Every clinic, agency, or supplier listed in a directory for a city or region
  • A product catalog with prices, variants, and stock status
  • Speakers, exhibitors, or sponsors from a conference site
  • Job listings, property listings, or public records
  • Profiles on a marketplace or community site

How a run goes

  1. Say what each row is, which fields you need, and roughly how many rows.
  2. If the data is going into an existing sheet, Strawberry reads its columns first so everything lines up.
  3. You get 10 to 20 rows to check. At high volume, a small mistake repeats thousands of times, so this is where it gets caught.
  4. The full run covers the whole site: every page, infinite scroll, load-more buttons, and detail pages.
  5. The result lands in a spreadsheet, with a link to the source on every row.

Data you can trust

A value that wasn't found stays marked as not found. It never becomes a guess, and it never quietly turns into a "no". If a site blocks access or asks for a login, Strawberry stops and hands it back to you rather than working around it.

Keep it fresh

Once a run works, save it as a skill so the next extraction starts from the same fields and checks. If the data changes over time, like prices, listings, or openings, a Routine can rerun it on a schedule and tell you only what changed.

Research deep enough to bet on

Define the question, evidence standard, and useful final format.

How to prompt AI for web scraping

Examples for turning repeated website content into structured data.