Export leads for AI CSV, Excel, or JSON to feed your AI with data

A good export is fuel for your AI

Every useful AI workflow starts with a file. You cannot ask a model to score your local market, cluster your prospects, or write a hundred first lines if it has nothing to read. The bottleneck is almost never the model. It is getting clean, structured lead data out of Google Maps and into a format the model can parse.

That is exactly what the Vonsel Chrome extension does. You search on Google Maps, capture the businesses you want, and export the list as CSV, Excel, or JSON. The feed your AI with data idea is simple: the extension produces the raw material, and your AI turns it into decisions. The export is the means, not the end.

If you want to skip straight to acting on the results, the paid Vonsel dashboard already bundles a mapped CRM, review intelligence, and an AI email per business. This guide covers the do-it-yourself path: export a file and run your own model over it.

CSV vs Excel vs JSON for AI

All three formats come from the same capture, so you are not locked in. The right choice depends on what you plan to do next. Here is the quick rule of thumb before we get into the tools.

CSV is the safe default because it is a flat, well defined text format that every model handles. If you want the formal rules, the CSV standard is documented in RFC 4180. JSON keeps structure that a spreadsheet flattens, so it shines when each business carries a list of reviews or attributes. The JSON format nests objects and arrays cleanly, which is why scripts and agents prefer it. For a deeper split, see our note on JSON vs CSV for AI analysis and the extension's export options.

Upload the file to ChatGPT and work the list

This is the workflow most people reach for, and it needs zero setup. Export your captured businesses as CSV or Excel, open ChatGPT, attach the file, and describe the job in plain language. The model reads the columns, runs the analysis in its sandbox, and returns a table you can download again.

Attaching a spreadsheet lets the model parse and compute over the rows rather than guess. If you want the official primer on how uploads behave, the agentic coding tools that read local files follow the same principle: give the AI the raw file, then ask precise questions.

Try prompts like these once the file is attached:

For the full step by step, read upload a lead CSV to ChatGPT, and for outreach specifically see writing cold emails from scraped data. The pattern is always the same: clean file in, structured answer out.

Point Claude Code at a JSON export

When your list grows past a few hundred rows, or you want repeatable analysis, a scripting agent beats a chat window. Export the capture as JSON, drop it in a folder, and open Claude Code in that folder. It reads the file, writes the analysis code, runs it, and shows you the result, all from plain language.

Because it works on your machine, nothing leaves your laptop and you can rerun the same job on next week's export in seconds. Ask it to parse the JSON, compute a lead score, filter out low quality entries, and write a clean CSV back. Then ask it to build a small map or chart from the coordinates so you can see coverage at a glance.

Our walkthrough on analyzing leads with Claude Code covers the exact prompts. The advantage over a chat upload is memory and repeatability: your scoring logic lives in a script, so every future export runs through the same rules without you re-explaining anything.

Codex and local LLMs on the same file

The same export feeds any engine you prefer. If you live in a code editor, Codex style assistants can read a CSV or JSON in your project and generate the parsing and scoring script for you, much like the Claude Code flow. The file format does not change, only the tool that consumes it.

Privacy sensitive teams can keep everything offline. Run a local model over your lead data and the export never touches a third party server. For plain lists, CSV is easiest to stream into a local model; for richer records, JSON keeps the reviews and attributes intact so the model can weigh them.

Where the dashboard takes over

Exporting and prompting is powerful, but at some point the analysis needs a home. That is the job of the Vonsel dashboard. The extension captures and exports; the dashboard is where the enriched list becomes a working pipeline.

Instead of scoring a CSV by hand every week, you get a mapped CRM that plots every business, review intelligence that surfaces the pain points buried in Google reviews, and an AI email written for each business from that context. It is the same value the do-it-yourself flow chases, without stitching four tools together. Compare the paths in Vonsel vs Outscraper or browse the rest of the AI Workflows series.