Build your own lead database and feed it to your AI

A database you own, not a list you rent

Bought lists go stale the moment they land in your inbox. A database you build yourself is different: you decide the niche, the city, and the fields, and you can refresh it whenever you want. The fastest way to start is to pull real businesses straight from the map that everyone already trusts, the same dataset behind the Google Maps Platform.

The free Vonsel Chrome extension captures Google Maps listings as you search: name, address, phone, website, category, rating, and review count. You then export the results as CSV, Excel, or JSON. That single file is the seed of your own lead database, and it is yours to keep, clean, and grow.

Once you have structured data, the interesting part begins. Modern AI tools read a spreadsheet or a JSON file better than most humans, so the export becomes the input for ChatGPT and Claude Code. This is the whole point of the "feed your AI with data" idea: real records in, useful decisions out.

Capture and export the raw records

Open Google Maps, search something specific like "dentists in Austin" or "law firms in Manchester", and let the extension collect the listings while you scroll. You can reopen a previous capture later and add more searches to the same set, so a database can span dozens of neighborhoods or entire regions.

When the capture is done, choose your export. CSV is the universal option and works everywhere. Excel keeps things tidy for manual review. JSON is the format AI agents and scripts prefer because the structure is explicit. If you are unsure which to pick, our guide on JSON vs CSV for AI analysis breaks down the tradeoffs, and the export options overview shows each one in the extension.

A quick note on scope. Google caps a single map search at around 120 results, so building a large database means running several tight searches by category and area rather than one giant query. That is normal, and it actually produces cleaner segments. If you want the legal context first, read is it legal to scrape Google Maps.

Feed the file to ChatGPT

ChatGPT is the fastest place to start because it reads your file directly. Open a new chat, attach the CSV or Excel export, and ask a plain-language question. Its advanced data analysis runs Python behind the scenes, so it can sort, group, and summarize thousands of rows without you writing a single formula. The OpenAI tools documentation explains how file inputs and code execution work under the hood.

Here are prompts that work well on a fresh export:

Because the answers come from your own records, they are grounded and specific, not generic. You can iterate in the same chat: ask it to export the scored list back to CSV, then re-import that into your database. For a deeper walkthrough, see how to upload a lead CSV to ChatGPT and ChatGPT prompts for lead lists.

Automate it with Claude Code

ChatGPT is great for one file at a time. When you have many exports, want repeatable steps, or need real scripts, Claude Code is the better fit. It runs in your terminal, reads and writes files in a project folder, and can execute code, which turns a folder full of JSON exports into a small pipeline you actually own. The Claude Code documentation covers setup and how it works with local files.

Drop your exports into a folder and ask Claude Code to build the tooling around them. A few things it handles cleanly:

Codex, local models, and the wider stack

The same export works with almost anything. Codex-style coding assistants handle the same folder-of-files pattern as Claude Code, so if that is your daily driver, the workflow barely changes. For privacy-sensitive lists you can also run the data through a local model with a tool like Ollama or LM Studio, keeping every record on your own machine. Our note on using local LLMs with lead data covers that path.

Whatever the model, the raw record has limits. Name, phone, and rating tell you who a business is, not why they might buy. That is where a plain export stops and a value-added database begins. The context that actually drives a pitch, like the pain points hiding in someone's reviews, has to be added on top.

From a flat file to a mapped CRM

Feeding a CSV to ChatGPT is a strong start, but you still end up managing files. The Vonsel dashboard is where the same captures become a working system. Every business lands on a mapped CRM, so you see your territory instead of scrolling rows.

On top of that sits the intelligence layer. Reviews Intelligence reads what customers actually complain about, so each record carries real pain points. Email Intelligence drafts a tailored message per business using that context, and Sales Intelligence helps you prioritize. This is the difference between a raw list and a value-added database: the same businesses, enriched with the context that closes deals.

You can also add reviews to businesses you already captured, keep a full history of your scrapes, and enrich websites for extra contact details. The extension feeds your AI. The dashboard is what your AI, and your sales team, work inside every day. Explore the whole AI Workflows series for more ways to combine both.