Claude Code promptsfor your Google Maps lead database
Export CSV, Excel, or JSON from the Vonsel extension, then feed your AI with data. Here are the prompts that turn a raw list into insight.
AI Workflows··8 min read
Key takeaways
The Vonsel Chrome extension captures businesses from Google Maps and exports CSV, Excel, or JSON. That file is the fuel for any AI tool.
ChatGPT is fastest for one off questions: upload the file and ask in plain English. Claude Code is best for repeatable jobs over a big lead database.
The extension is the means. The end is a value added, contextualized database your AI can score, segment, and write outreach from.
The idea
Feed your AI with data
An AI model is only as useful as the data you point it at. Ask ChatGPT or Claude for a list of dentists in your city and you get vague, outdated guesses. Hand the same model a clean, structured export of real businesses and it becomes a genuine research analyst, a copywriter, and a data cleaner in one.
That is the whole workflow: capture real Google Maps businesses with the Vonsel extension, export a file, and let the AI work the leads. The extension is the means. A value added, contextualized lead database is the end, and that database is what makes every prompt below actually work.
The best part: you do not need to be a developer. In ChatGPT you upload a file and type a question. In Claude Code you keep the file in a folder and describe the job. Both read your data directly.
Get the data first
Capture Google Maps businesses and export CSV, Excel, or JSON in a couple of clicks. Free download. No trial, no credit card.
Open Google Maps, run a search like "roofers in Denver", and let the Vonsel extension capture the results. When you are done, export. You get three formats and each one has a natural AI home.
Format
Best for
Why
CSV
ChatGPT quick chats
Universal, opens anywhere, easy to paste or upload.
Excel (XLSX)
Spreadsheet review before AI
Sort and skim by hand, then hand the sheet to the model.
JSON
Claude Code and Codex
Keeps nested fields like reviews and social profiles intact for scripting.
Each row carries the business name, address, phone, website, rating, review count, and in many cases emails, WhatsApp numbers, and social profiles from Vonsel website enrichment. That richness is what separates a real captured database from a bought list. For a deeper look at format trade offs, see our note on JSON vs CSV for AI analysis.
3
export formats from the extension: CSV, Excel, JSON
0
lines of code needed to prompt ChatGPT over your list
10k+
rows Claude Code can script across in one pass
Step two, the fast lane
Work the list in ChatGPT
ChatGPT is the fastest way to start. Upload your CSV or Excel export using the attachment button, wait for it to read the file, then ask. Because ChatGPT can run data analysis on an uploaded sheet, it will parse columns, count, filter, and summarize without you writing a formula. OpenAI documents how the model works with your input in its text generation guide.
Here are prompts that earn their keep on a fresh export. Paste them right after the upload finishes.
Goal
Prompt to paste
Clean the data
"Deduplicate this list by phone and website, flag rows with missing emails, and return a cleaned table."
Score the leads
"Score each business 1 to 10 as a web design prospect. Weight low rating, few reviews, and no website. Explain the top 10."
Segment by fit
"Group these businesses into three tiers: hot, warm, cold, based on review count and rating. Give me counts per tier."
Write outreach
"For the five highest scoring leads, write a two sentence cold email opener that references something specific about each business."
ChatGPT is great for a single file and a single question. When your database grows to thousands of rows, or you want the same job to run again next week on a fresh export, Claude Code is the better tool. It lives in your terminal, reads the JSON export straight from your project folder, and can write and run small scripts to transform the whole database at once.
Drop your leads.json export into a folder, open Claude Code there, and try these. They are written the way you would actually type them.
Clean and normalize
"Read leads.json, normalize phone numbers to E.164, lowercase all emails, drop exact duplicates, and write the result to leads_clean.json."
Score at scale
"Add a fit_score field to every record from 0 to 100. Reward no website, rating under 4.0, and fewer than 30 reviews. Save and show me the score distribution."
Mine the reviews
"For each business with reviews, summarize the top three recurring complaints and tag the record with a pain_point field I can use in outreach."
Build a report
"Generate a one page Markdown market report: total businesses, share without a website, average rating, and the ten strongest prospects with reasons."
Split into zones
"Cluster leads by latitude and longitude into five geographic zones and export one CSV per zone for my field reps."
Draft a sequence
"For the top 25 by fit_score, write a three step email sequence that opens with each business pain_point. Output as one Markdown file."
Because Claude Code can write files back to disk, the output is not a chat you copy from. It is a new dataset, a report, or a batch of message drafts sitting in your folder, ready to import. See our fuller walkthrough on analyzing leads with Claude Code and the format checklist in exporting leads for AI.
A prompt is only as smart as the data behind it. Google Maps data comes from real business listings, as described in the Google Places documentation. Capture it cleanly with the extension and the same ten prompts that produce noise on a bought list produce gold on a real one.
The rest of the toolbox
Codex and local LLMs
Claude Code and ChatGPT cover most of what a lead database needs, but two other options are worth knowing. Codex style coding assistants are handy when you want to wire your export into a repeatable pipeline, for example a script that pulls new rows every week and re enriches them. Our note on enriching lead lists with Codex covers that pattern.
Local LLMs matter when the data is sensitive or you simply do not want a lead list leaving your machine. A model running on your own laptop can score and segment a JSON export with no data ever hitting a server. We break down when this is worth it in using local LLMs with lead data.
Whatever model you choose, the export is identical. That is the point of a feed your AI with data workflow: one clean file, any AI on the other end.
The model is the engine. Your captured database is the fuel
Turn Google Maps into an AI ready database
Capture, export, and prompt. Explore the Vonsel dashboard or browse more AI workflows. Free download. No trial, no credit card.
For Claude Code, JSON is usually the best fit because it preserves nested fields like reviews and social profiles, and Claude Code can read and script over it directly in your project folder. For quick ChatGPT chats, CSV or Excel works well. The Vonsel extension exports all three, so you can pick per task.
Do I need to know how to code to prompt AI over a lead list?
No. In ChatGPT you upload the file and ask in plain English. Claude Code adds more power for repeatable jobs across large databases, but you still describe the task in natural language. The prompts in this guide are copy and paste ready.
Is the Vonsel Chrome extension free to export leads?
Yes. The Vonsel Chrome extension is a free download with no trial and no credit card. You capture businesses from Google Maps and export CSV, Excel, or JSON, then feed that file to any AI tool you like.
Add Chrome Extension