AI Agents for Lead Research Feed your AI with real Google Maps data

AI agents are hungry for real data

Ask ChatGPT for "50 dentists in Chicago with phone numbers" and you get a confident list that looks right and is mostly wrong. Large language models predict text, they do not browse your local market. Without a real dataset underneath them, AI agents hallucinate businesses, invent phone numbers, and quietly fabricate ratings.

The fix is not a better prompt. The fix is real data. Once you hand an agent a file of verified businesses pulled straight from Google Maps, everything downstream (scoring, segmentation, outreach copy) is grounded in reality. The model stops guessing and starts reasoning over facts you own.

That is the whole idea behind "feed your AI with data." You do the gathering with a scraper, the AI does the thinking. The feed your AI with business data playbook covers the mindset. This post covers the practical flow: how to get the file and how to work it with the two agents most people actually use.

Capture and export the file

Open Google Maps, run a search like "roofers in Denver," and let the Vonsel pop-up capture the listings. When the scrape finishes you export in one click. The extension gives you three formats: CSV for spreadsheets, Excel for a formatted workbook, and JSON for anything programmatic.

Which format you pick depends on the agent. A quick rule: CSV or Excel for ChatGPT, JSON for Claude Code and code-driven agents. The CSV, XLSX and JSON export options guide breaks down when each one wins, and exporting leads for AI shows how to shape the columns before you hand them over.

The fields matter more than the format. Every export includes business name, category, full address, phone, website, rating, review count, and review text. Those are exactly the signals an agent needs to reason about fit and intent, and they come from Google Maps rather than a stale purchased list. For the difference that makes, see Google Maps data vs bought lists.

Work the list with ChatGPT

ChatGPT is the fastest way to start because it reads spreadsheets natively. Drag your CSV or Excel export into a chat and ChatGPT parses it into a table it can analyze, sort, and filter. Its built-in data analysis runs Python behind the scenes, so it handles thousands of rows without you writing any code. See OpenAI's tools documentation for how file inputs and code execution work.

Once the file is uploaded, prompts do the research for you. A few that pull their weight:

Score the list: "Rank these businesses by outreach priority. A high score means a rating under 4.2 with more than 40 reviews and a working website. Add a score column and a one-line reason." That single prompt turns a flat list into a ranked pipeline.

Segment by pain: "Read the review text and group these businesses into three buckets: slow service complaints, pricing complaints, and staff complaints. Return a table with the business name and its bucket." Now you know the angle before you ever pick up the phone.

Draft outreach: "For the ten top-scored businesses, write a two-sentence cold email opener that references a specific pattern in their reviews." Because the reviews are real, the openers land as researched, not generic. For the deeper version of this, read analyze lead lists with ChatGPT.

Scale it with Claude Code

ChatGPT is great for one list in a chat window. When you have twenty exports, or you want the same analysis run every week, you want an agent that lives next to your files. That is Claude Code, a terminal agent that reads the folder your JSON and CSV exports sit in and acts on them directly.

The workflow is simple. Drop your Vonsel exports into a project folder, open Claude Code there, and describe the job in plain English. It writes and runs the script, then hands you the result. No copy and paste, no upload limits, no losing the thread when the list is long.

Real requests look like this: "Merge every JSON export in this folder, drop duplicate businesses by phone number, and write a single ranked CSV." Or: "For each business with a website, flag the ones whose review text mentions booking or scheduling problems, and save them to hot-leads.csv." Claude Code turns those sentences into repeatable code you can run again next month on fresh data.

Because it is scriptable, the work compounds. Once a scoring routine exists, every new scrape flows through the same logic in seconds. The analyze leads with Claude Code guide walks through a full setup, and from scrape to AI-powered CRM workflow ties it into your pipeline.

ChatGPT vs Claude Code for lead research

Most people start in ChatGPT and graduate to Claude Code as their lists grow. Neither replaces the other. You explore a fresh scrape in ChatGPT and you automate the recurring work in Claude Code.

Codex and local models

If you live in a code editor, OpenAI Codex does the same job as Claude Code from the terminal or your IDE. Point it at your exported JSON and describe the transformation. It is a natural fit if your team already builds tooling around lead data.

For sensitive lists or heavy volume, a local model keeps everything on your machine. Tools that run open models on your own hardware read the exact same CSV and JSON exports, with no data leaving your laptop and no per-token cost. The tradeoff is smaller context and more setup. The use local LLMs with lead data guide covers when the privacy is worth it.

Whatever agent you choose, the input is identical: a clean export from a free browser extension. The scraper stays the same, the AI layer is yours to swap. See the Chrome extensions documentation for how browser-based capture works under the hood.

Where the research becomes revenue

An AI agent gives you a ranked, segmented, annotated list. That is a great research output. It is not yet a sales motion. To actually work the leads you need somewhere to see them on a map, track their status, and reach out at scale.

That is the job of the Vonsel dashboard. The extension feeds your captured businesses into a mapped CRM. Vonsel's own review intelligence scores the pain points for you, and it drafts an AI email tailored to each business, so the research your agent did outside becomes contextualized, value-added data you can act on inside. The scraper and the external agent are the means. The dashboard is the end.

You can run both tracks together: export to your favorite AI for exploratory analysis, and keep the working pipeline in Vonsel where the mapped CRM and per-business AI email close the loop. Same data, feeding your AI and your sales motion at once.