Upload a lead CSV to ChatGPT Scrape Google Maps, then let AI do the work

Feed your AI with real business data

Large language models are only as useful as the data you give them. Ask ChatGPT to "find dentists in Madrid" and it will guess or refuse. Hand it a real file of dentists you captured from Google Maps, and suddenly it can rank them, spot patterns, and write a first email for each one. The model does not find the data. You do, then it reasons over it.

That is the whole point of the workflow on this page. You scrape the businesses you care about with the free Vonsel extension, export a structured file, and upload it to the AI of your choice. No API keys, no scripts, no copy-pasting rows one at a time.

Because the extension exports CSV, Excel, and JSON, you are not locked into one tool. CSV and Excel are ideal for a quick chat upload. JSON keeps nested fields (reviews, social profiles, opening hours) intact for heavier work in Claude Code or your own pipeline.

Scrape the leads, then export the file

Open Google Maps and search the way you normally would, for example "coffee shops in Barcelona" or "plumbers near me". With the Vonsel pop-up open, capture the results as you scroll. The extension collects the name, address, phone, website, rating, review count, and more into a tidy list. For a full walkthrough see our guide on exporting Google Maps data to CSV.

When the capture looks right, filter out anything you do not want and click export. Pick the format that fits your next step:

Which is better, CSV or JSON? For a one-off analysis in a chat window, CSV wins on simplicity. For anything with reviews or repeated processing, JSON keeps the structure clean. We break this down further in JSON vs CSV for AI analysis.

Upload the CSV to ChatGPT and prompt it

Open ChatGPT, click the attachment icon in the message box, and select your exported CSV. ChatGPT reads the header row, understands your columns, and can run analysis on the data directly. If you want the mechanics of how the model works with your text and files, the OpenAI text guide is the authoritative reference.

The magic is in the prompt. A vague ask gives you a vague answer. Be specific about the column names and the outcome you want. Here are prompts that work well on a Google Maps lead export:

Claude Code when the file gets large

A chat upload is perfect for a few hundred rows. When you are working thousands of businesses, or you want a repeatable process, Claude Code is the stronger tool. It runs in your terminal, reads the file straight from disk, and can write a small script to filter, enrich, and re-export without you touching a spreadsheet.

Export the leads as JSON so nested review data survives, drop the file in a folder, and open Claude Code there. A prompt like "read leads.json, score each business by opportunity, and write the sorted result to scored.csv" is enough to get a reproducible pipeline. It remembers the columns across turns, so you can refine the logic instead of re-uploading each time.

Prefer to keep everything on your machine? A local model through a runner like Ollama or LM Studio can chew through the same CSV offline, which matters when the data is sensitive. And if you live in a developer workflow, tools like Codex can enrich the file inside your editor. The file format is the same either way, which is exactly why exporting clean data first pays off.

From a raw CSV to a database you can work

Uploading a CSV to ChatGPT is a great first move. But a chat window forgets, and a spreadsheet does not know when you called a lead or what the reviews said about them. That is where the extension being a means, not the end, matters. The file you export is designed to become a value-added database, not a throwaway table.

Inside the Vonsel dashboard, the same leads land in a mapped CRM. Reviews are analyzed for pain points, an AI email is drafted per business, and you can reopen a previous scrape to add more results or attach reviews to businesses you already captured. The AI work you do in ChatGPT becomes context that lives with the lead instead of disappearing when you close the tab.

So treat the CSV upload as the exploration layer and the dashboard as the system of record. You get the speed of a chat and the memory of a CRM. For the bigger picture of turning captures into contextual data, read feed your AI with business data.

One more practical note: the fields you export map to the same attributes Google exposes. If you want to understand what each column means, Google documents them in the Places data fields reference, which helps you prompt the model with precise column names.