Structured Data Extraction

Overview

Some questions aren't really chat questions — they're spreadsheet questions. "List every product on this page with its price" or "give me every row of this table" is something a written answer handles badly: the AI reading and re-typing hundreds of rows is slow, and it's easy for a few to get skipped or paraphrased along the way.

Structured Data Extraction skips that. It scans the current page directly for a large repeated list, grid, or table, and pulls it out as real rows and columns — every item accounted for, ready to download.

Why this instead of just asking

Asking "list all the products on this page" as a normal question works for a handful of items, but breaks down as the list grows: the AI has to read and re-type every item, which is slow, costly, and prone to accidentally skipping or shortening entries in a very long list. Structured Data Extraction reads the page's actual repeated structure directly and returns every row it finds as real data — nothing is summarized or paraphrased along the way.

When to use it

Reach for this whenever what you want is really tabular: product listings, search results, directories, pricing tables, comparison charts, leaderboards. If you only need a couple of specific numbers, a normal question is faster — this is for "give me all of it."

How to use it

1. Open the Evelant side panel on the page with the list or table you want.
2. Click the Extract structured data from this page tool button in the panel's toolbar.
3. Evelant scans the page and reports back how many rows it found and what the columns are.
4. A data card appears with a preview and two download buttons: Download CSV and Download JSON.
5. You can also just ask questions about the extracted data directly in the same chat — "which of these is the cheapest," "sort these by rating" — since it's now part of the conversation.

A worked example

Say you're comparing laptops on a retailer's search results page, with dozens of listings each showing a name, price, and rating.

1. Load the search results page and scroll through so every listing you care about has rendered.
2. Open the side panel and click Extract structured data from this page.
3. Evelant reports something like "Extracted 84 rows from this page (columns: Name, Price, Rating)."
4. A data card appears with a preview and download buttons.
5. Instead of downloading right away, you ask in the same chat: "which of these is the cheapest with a rating above 4 stars?" — and get an answer computed directly from the extracted rows, not a re-scan of the page.
6. When you're done exploring, click Download CSV to keep the full list for later.

What it works well on

• Long product or listing grids (marketplaces, directories, search results)
• Tables with a clear header row
• Repeated card layouts (the same fields appearing over and over — name, price, rating, and so on)

What it doesn't work on

• A page with nothing repeated. If there's no large repeated pattern to find, Evelant tells you it couldn't find a list or grid worth extracting — it won't invent structure that isn't there.
• Content that isn't visible on the page, such as items behind pagination you haven't loaded yet, or hidden behind a login wall Evelant can't see.
• Extraction is capped at a very large but finite number of rows per page — normal browsing pages won't come close to it, but it exists to keep extremely large pages responsive.

Tips

• Load everything you want first. If a page paginates or loads more items on scroll, get all the items you want visible before running extraction — it only reads what's currently on the page.
• Check the column names in the summary message before downloading — if they don't look right, the page may not have the layout this feature expects.
• Ask follow-up questions instead of re-extracting. Once data is in the chat, questions like "how many of these are over $50" are answered directly, no second extraction needed.
• CSV for spreadsheets, JSON for anything else. Use CSV if you're opening the result in Excel or Sheets; use JSON if you're feeding it into another tool or script.
• Ask analytical questions before downloading. Filtering, sorting, and comparing within the chat is often faster than opening the file in another program for a quick check.

Example use cases

• Pulling every listing from a real-estate search into a spreadsheet to compare price per square foot yourself.
• Extracting a comparison table of specs across several products into CSV for a side-by-side review.
• Grabbing a full leaderboard or ranking page as structured data to analyze trends instead of eyeballing it.
• Turning a long FAQ page into rows of question/answer pairs for your own reference.

Troubleshooting

"I couldn't find a large repeated list or grid to extract on this page." The page doesn't have a pattern this feature recognizes as tabular. This is common on pages built from lots of unrelated content rather than one repeated structure.

The row count looks too low. Some content may not have loaded yet — try scrolling through the full list first (so lazy-loaded items render), then run extraction again.

A column is missing or looks wrong. The page's markup may present that field differently across rows (for example, a price shown as plain text in some rows and inside an icon+text combo in others). Extraction does its best to match the overall repeated pattern, but a genuinely inconsistent layout can affect one column more than others.

Frequently asked questions

Does this use a chat credit? The extraction itself is a page scan, not a model call — the credit cost is the same as any other message in that conversation once you start asking questions about the results.

Can I extract from a PDF? This feature looks at the page's own layout, so results depend on how the PDF is being viewed rather than being a dedicated PDF-table extractor — for spreadsheet-style data trapped in a PDF, converting it to CSV manually may work better.

Where does the downloaded file go? Wherever your browser normally saves downloads — Evelant doesn't store the extracted file anywhere else.

Can other people in a Shared Room see the extracted data? The summary message and data card appear in the chat like any other message, so yes — anyone in that shared conversation can see and download it too.

Can I extract from more than one page and combine the results? Each extraction covers the page you're currently on. For data spread across multiple pages, extract each one separately and combine the downloaded files yourself.

What if the page has more than one table or list? Extraction targets the largest repeated structure on the page. If a page has several distinct tables and you only want one of them, it works best on pages where the section you care about is the dominant repeated pattern — scrolling so only the relevant content has rendered can help.

Related features

• Find on Page — for pointing at one specific fact rather than pulling out an entire dataset.
• Full-Page Deep Scan — for reading an entire long page as context rather than extracting its tabular structure.
• Code Sandbox & File Generation — once you have extracted data, ask Evelant to turn it into a chart or a formatted spreadsheet.