Scrape any business in any country and feed it straight to your AI

Any category, any city, one clean export

The phrase "scrape any business in any country" sounds ambitious until you remember where the data lives. Google Maps holds business listings almost everywhere on earth, and every listing exposes the same core fields: name, address, phone, website, category, rating, and reviews. The Google Maps Platform data model is consistent from Tokyo to Berlin to Bogota, which is exactly why a browser-based capture works the same regardless of where you point it.

That consistency is the whole trick. You search a category plus a city (dentists in Madrid, law firms in Toronto, coffee shops in Paris), let the extension read what Maps already renders, and pull out a structured list. No API keys, no country-specific setup, no code. The map is the same shape everywhere, so your export is too.

From there the interesting part begins. A raw list of businesses is not the goal. The goal is what your AI does with it: ranking, clustering, enriching, and writing. That is the "feed your AI with data" idea, and it starts with a clean file you actually own.

Capture and export in three formats

Open Google Maps, run your search, and let the Vonsel extension capture the visible results. When the list looks right, export it. You get to choose the format, and the format matters because different AI tools prefer different shapes. For a deeper look at when to pick each one, see export CSV, XLSX, and JSON options and the plain-English breakdown in JSON vs CSV for AI analysis.

All three come straight out of the extension, so you never convert files by hand. If you want the full rationale on preparing exports for models, our guide to exporting leads as CSV, Excel, and JSON for AI walks through each one.

Feed ChatGPT: upload, then ask

ChatGPT is the fastest way to turn an export into answers because it can read your file directly. Drop a CSV or Excel file into a chat, and its built-in analysis tool loads the rows into a working table it can filter, sort, and chart. The format helps here: CSV keeps every column flat and predictable, which is why the structured export lines up cleanly with what the model expects.

Once the file is in, you drive it with plain language. A few prompts that work well on a Maps export:

For a step-by-step version of this, we cover it in how to upload a lead CSV to ChatGPT and the writing side in writing cold emails with ChatGPT from scraped data.

Feed Claude Code: point it at the file

When your list grows past a few hundred rows, or when you want the same processing every time you export, a terminal agent earns its keep. Claude Code runs in your project folder, reads the JSON export directly, and writes small scripts to transform it. You describe the outcome in words and it produces the code, runs it, and shows you the result.

Because JSON keeps nested fields intact (reviews, categories, coordinates), it is the natural input here. Typical asks look like this:

That last row is the real advantage. Once the pipeline exists, next month's scrape is one command away. We go deeper in analyzing leads with Claude Code, and the broader vision lives in feed your AI with business data.

Codex and local models

If you live in a code editor, OpenAI's Codex plays the same role as Claude Code: point it at a JSON or CSV export and describe the transform. It is a good fit when your lead work already sits inside a repository alongside other automation.

For sensitive lists you would rather not upload anywhere, a local model keeps the data on your machine. The trade is capability for privacy, and it works well for straightforward tasks like tagging, deduping, and summarizing. We cover the setup in using local LLMs with lead data. Whichever engine you choose, the export step never changes: capture in the browser, choose your format, hand it off.

Where the workflow actually ends

Uploading a file to ChatGPT is a great first pass, but a chat window is not a system of record. This is where the extension stops being the point and the Vonsel dashboard takes over. Your captured businesses land in a mapped CRM, so the same list you would have pasted into a prompt is already scored, plotted on a map, and grouped into territories.

On top of that sit the intelligence layers: Reviews Intelligence reads what customers actually complain about, Email Intelligence drafts a tailored message per business, and the enrichment step fills in websites and social profiles the raw listing never showed. That is the difference between a value-added, contextualized database and a raw CSV you still have to babysit. It is the same argument we make in value-added database vs raw CSV.

You can absolutely stop at "export and prompt ChatGPT," and for a quick research task that is plenty. But when the list becomes a real pipeline you work every day, the dashboard is where it belongs. The AI workflow and the CRM are two ends of the same road, and the export is what connects them.