Write cold emails with ChatGPT from scraped data

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

ChatGPT writes a great cold email when it knows who it is writing to. The problem is never the prose. The problem is the raw material. A model with no data invents a generic template, and generic templates get ignored. A model with a clean list of real businesses, categories, cities, and websites writes something a prospect actually reads.

That is the whole workflow in one line: capture real businesses from Google Maps, export them to a file, and hand the file to the AI. The Vonsel extension handles the capture and the export. The extension is the means. The email is the end.

Most people skip the data step and paste a company name into the chat one at a time. That does not scale past ten emails. Export the list once and you can draft a hundred drafts in a single prompt.

Export a clean lead file from the extension

Open Google Maps, search a niche and a city (for example "roofers in Austin"), and let the Vonsel pop-up capture the results. Each business comes with the fields an AI needs to personalize: name, category, address, city, phone, website, rating, and review count. When the capture is done, export it. You choose the format that matches your AI tool in the export options.

A quick rule of thumb on formats, covered in more depth in our guide to exporting leads for AI:

Whatever you pick, this file is the asset. It is portable, it is yours, and it works with every AI tool below. That is what "feed your AI with data" means in practice, and it is the same idea behind feeding your AI with business data.

Write the emails in ChatGPT

This is the fastest path for most people. Open ChatGPT, attach your exported CSV, and give it a clear instruction. ChatGPT reads tabular files natively, so you do not need to paste rows by hand. The official OpenAI text generation guide is a good reference for how to shape instructions.

A prompt that works well:

"Attached is a CSV of local businesses I want to reach. For each row, write a 90 word cold email offering website redesign services. Open with a specific detail from the business name or category, keep the tone plain and human, avoid hype words, and end with a soft one line call to action. Return the results as a table with columns: business, subject, body."

The reason this produces usable drafts is that ChatGPT is not guessing at the prospect. It is reading your real captured data and personalizing off it. For a deeper library of prompt patterns, see our ChatGPT prompts for lead lists, and if you want to trim the list first, analyze the lead list with ChatGPT before you generate a single email.

One caution: always keep drafts as drafts. Read every email before it goes out, and stay inside cold outreach law for your region. The FTC CAN-SPAM compliance guide is the baseline for the United States.

Scale it up with Claude Code

When your export grows past a few hundred rows, a browser chat gets clumsy. This is where Claude Code shines. It runs in your terminal, reads the exported file directly from disk, loops through every business, and writes the drafts back to a new file. No copy and paste, no truncation, and the run is repeatable.

Export your list as JSON, drop it in a folder, and ask Claude Code to do the work. A prompt like:

"Read leads.json. For each business, write a personalized 90 word cold email selling local SEO. Use the category and city to make the opener specific. Write the results to drafts.csv with columns business, email, subject, body. Skip any row missing a website."

Because it works against the file itself, Claude Code handles 500 rows as easily as five, and you can rerun the same command whenever you capture a fresh batch. Our walkthrough on analyzing leads with Claude Code covers the setup end to end. Together, ChatGPT and Claude Code cover the vast majority of real cold email jobs: ChatGPT for a quick batch in the browser, Claude Code for repeatable runs at volume.

If you already live in another toolchain, the same file feeds them too. Codex can loop the export in a script, and a local LLM like Llama or Mistral running on your own machine can draft emails offline when privacy matters. The export format does not care which model reads it.

ChatGPT vs Claude Code for cold email

Neither is wrong. Pick ChatGPT to move fast on a small list, and switch to Claude Code the moment the list is big enough that the browser feels slow. The one thing both need is the same: a clean export from the extension.

From draft emails to a working pipeline

Drafting emails is the start, not the finish. Once you have replies, those businesses become a pipeline you need to track, and a folder of CSV drafts is a poor place to run a sales motion. That is where the export stops being enough on its own.

The Vonsel dashboard is the other half of the story. The same businesses you captured live inside a mapped CRM, plotted on a map, tagged by status, and enriched from their websites. Its Email Intelligence writes an AI email tailored to each business, and its Reviews Intelligence surfaces the pain points those emails should speak to. If you want to personalize off real complaints, our note on personalizing cold email with review pain points shows the idea.

So the full arc looks like this: the free extension feeds your AI, your AI drafts at scale, and the dashboard turns the responders into tracked deals with context attached. Raw CSV is a great input for ChatGPT. A value-added database is what closes them.