From scrape to AI-powered CRM Feed your AI with real business data

Your AI is only as good as the data you feed it

Large language models are brilliant at reasoning over structured data, and useless without any. Ask ChatGPT to "find me 200 dentists in Chicago with weak websites" and it will invent plausible names, because it has no live database to draw from. The fix is simple: give the model real rows, then let it think.

That is the whole point of the workflow in this guide. You use a scraper to capture the raw material, an AI to reason over it, and a CRM to act on the result. The scraper is not the destination. It is the on-ramp.

The free Vonsel Chrome extension is built for exactly this. You browse Google Maps like normal, capture the businesses you want, and export them as CSV, Excel, or JSON. No credit card, no trial timer, no paywall between you and your data.

Capture and export the leads

Open Google Maps, search for a category and city (say "roofers in Denver"), and let the Vonsel pop-up capture the results. When you are done, export. Choose CSV or Excel if you plan to work in a spreadsheet-style tool, or JSON if you want a machine to parse nested fields like categories, ratings, and reviews.

Each row carries the fields that matter for outreach: business name, address, phone, website, category, rating, and review count. If you are unsure which format fits your task, our note on JSON vs CSV for AI analysis breaks down the trade-offs, and the export options guide covers the buttons.

One habit that pays off: capture more than you think you need. It is far cheaper to let an AI filter a 500-row list down to 40 great fits than to run five separate scrapes. Volume in, precision out.

Work the list in ChatGPT

ChatGPT is the fastest entry point because it will run code on your file for you. Drag your CSV or Excel export straight into the chat box and it spins up a sandbox to read the rows. Under the hood this is the same code interpreter tooling the API exposes, so it can sort, filter, and chart without you writing a line of Python.

Prompts that work well on a fresh export:

The trick is to ask for a downloadable file back. Say "return the scored list as a new CSV" and you get a clean file ready to import into a CRM. For a deeper walkthrough, see how to upload a lead CSV to ChatGPT.

Scale it up with Claude Code

ChatGPT is perfect for a one-off list. When you are running the same analysis every week across many exports, you want something repeatable, and that is where Claude Code earns its place. It works in your terminal, reads your JSON exports directly, and can write a small script that scores and segments any file you drop in the folder.

A realistic session looks like this. You export ten JSON files from ten cities, open Claude Code in that directory, and ask it to "read every JSON here, merge them, tag each lead by city, flag the ones with under 20 reviews, and write a combined ranked CSV." It writes the script, runs it, and shows you the output. Next week you rerun the same script on new files.

Because Claude Code keeps the logic in a file, your scoring rules become an asset instead of a prompt you retype. Our guide on analyzing leads with Claude Code shows the exact commands. This is the difference between doing analysis once and building a repeatable lead pipeline.

Codex, agents, and local models

The same export feeds any model. If you live in an IDE, a coding agent like Codex can turn your JSON into an enrichment script or a small dashboard the same way Claude Code does. If privacy matters and you would rather keep leads on your own machine, run a local model with Ollama and pass it the CSV so nothing leaves your laptop.

None of this changes the shape of the workflow. You capture with the extension, export a clean file, and point whatever model you trust at it. The data format is the universal adapter: every serious AI tool reads CSV and JSON.

From clean file to a mapped CRM that sells

Analysis is not the finish line. A scored CSV sitting in a downloads folder never closed a deal. The point of the workflow is to move those leads into a place where you can work them, and that is what the Vonsel dashboard is for.

Where a bare export gives you rows, the dashboard gives you a mapped CRM: every business plotted so you can plan a territory, plus Reviews Intelligence that reads what customers complain about and Sales Intelligence that turns those pains into talking points. Email Intelligence then drafts a tailored AI email per business, so the personalization you prototyped in ChatGPT becomes a standing feature, not a copy-paste chore.

That is the value-added layer: contextualized databases where each lead carries its own pain points, not a naked list of names. When you are ready to push scraped leads into a system that acts on them, see moving Google Maps leads to a CRM.

Built for the whole pipeline