Segment leads with AI Export from the extension, then let ChatGPT and Claude Code sort every row

Feed your AI with real business data

A raw list of 800 businesses is not a strategy. It is a wall of rows. The value shows up the moment you split that wall into segments you can actually work: dentists with under 20 reviews in one bucket, four-star restaurants without a website in another, high-volume gyms in a third.

That split is exactly what a language model is good at. The catch is that AI needs clean, structured input. This is where the Vonsel Chrome extension comes in: it captures businesses straight from Google Maps and exports them as a tidy CSV, Excel, or JSON file. The extension is the means. Segmenting the data is where the work begins.

If you want the wider picture of what it means to hand a model your own data instead of asking it to guess, read feed your AI with business data. This post is the practical version: file in, segments out.

Export a file the AI can read

Open Google Maps, search your niche and area, and let the extension capture the results. When you export, you choose the format. Each one suits a different downstream tool, and the extension writes all three from the same capture. See CSV, XLSX, and JSON export options for the details.

CSV is the plain-text workhorse defined by RFC 4180, which is why every AI tool ingests it without complaint. Not sure which to pick for a given job? Our note on JSON vs CSV for AI analysis breaks it down. As a rule: CSV for ChatGPT, JSON for Claude Code.

Segment with ChatGPT: upload and prompt

ChatGPT is the fastest path for a single file. Open a chat, click the attachment icon, and upload the CSV the extension gave you. Once it loads, you describe the segments you want in plain language and the model does the sorting. This mirrors the flow in our guide to uploading a lead CSV to ChatGPT.

A prompt that works well on a Google Maps export looks like this:

"Here is a CSV of businesses I captured from Google Maps. Segment the rows into groups by (1) rating band (under 4.0, 4.0 to 4.5, above 4.5), (2) review count (under 20, 20 to 100, over 100), and (3) whether a website is present. Give each segment a short label, a count, and one sentence on how I should approach it. Return a table."

ChatGPT reads every row, applies your rules, and hands back tidy buckets. From there you can go deeper: ask it to flag businesses with weak ratings and high review counts (a reputation-management angle) or to pull every listing with no website into its own list. If you want to rank inside a segment rather than just group, pair this with scoring leads with ChatGPT.

One habit worth keeping: ask ChatGPT to output each segment as a separate downloadable CSV. That way every bucket becomes its own working list you can act on, one campaign at a time.

Segment with Claude Code: scriptable and repeatable

When the list is large, spread across several exports, or something you will refresh every week, a chat window gets tedious. Claude Code is the better fit here. It runs in your terminal, reads files off disk directly, and writes real code, so it can segment thousands of leads and save the results without you copying anything by hand.

Drop your exported JSON or CSV into a folder, open Claude Code there, and describe the job:

"Read leads.json in this folder. Group the businesses into segments by category and by review-count tier. For each segment, write a separate CSV to an /out folder with a filename like category_tier.csv, and print a summary table with the count per segment."

Claude Code writes the script, runs it, checks the output, and fixes its own errors if a column is missing or a value is blank. The next time you export a fresh batch from the extension, you rerun the same script and get the same segmentation instantly. Our walkthrough on analyzing leads with Claude Code goes further into this loop.

This is the real gap between the two tools: ChatGPT segments this file, once. Claude Code builds a segmentation you can run on every future export without rethinking the prompt.

Codex and local LLMs for private lists

ChatGPT and Claude Code cover most needs, but two other options fit specific situations.

Codex-style coding assistants work much like Claude Code: point them at your exported file and ask for a segmentation script. If your team already lives inside a particular coding assistant, there is no reason to switch. The extension output is plain CSV and JSON, so it reads cleanly in any of them.

Local LLMs matter when the list cannot leave your machine. If you are handling a client database and want nothing uploaded to a cloud service, run a model locally and feed it the same CSV. You trade some reasoning quality for full privacy, which is often the right call for sensitive B2B data. Because the file is small and structured, even a modest local model can group rows by category and rating without trouble.

Whatever the engine, the pattern is identical: export from the extension, hand over the file, prompt for segments.

From segments to a working pipeline

Segmenting a file with AI is a strong first move, but a folder of CSVs is still not a sales operation. The segments live in your downloads, disconnected from where the outreach actually happens.

That is the role of the Vonsel dashboard. It takes the same captured businesses and turns them into a mapped CRM, so each segment becomes a territory you can see on a map, an AI email written per business, and a database enriched with review context and the pain points behind each listing. The extension feeds the data. The dashboard is where segments become deals.

You can keep working purely in ChatGPT and Claude Code if a lightweight loop is all you need. When you want the segments to live somewhere they drive real follow-up, the dashboard is the end the whole workflow points toward. Browse the rest of the AI Workflows series for more ways to put your exported data to work.