Generate outreach sequences with AI Feed your AI with scraped Google Maps leads

Data first, then let the model write

An outreach sequence is only as good as the context behind it. A generic three-email cadence lands in spam. A cadence that references the prospect's category, city, and website reads like it was written for one business, because it was. The trick is giving your AI real business data to work from instead of asking it to invent details.

That is the whole point of the feed your AI with data approach. You scrape a clean list of local businesses, then hand the file to a model and let it draft the sequence. The Vonsel extension handles the first half: you open Google Maps, run a search like "dentists in Austin", and it captures every listing into a structured table you can export.

Each row carries the fields a language model needs to personalize: business name, category, address, city, phone, website, and social handles. The scrape is the raw material, not the finished campaign. What you build on top of it is where the value lives.

Export the leads from the extension

Once a capture finishes, open the panel and choose your format. CSV and Excel are ideal for uploading into a chat window. JSON is the cleanest input when you want a script to loop over records field by field. If you are unsure which to pick, our guide on CSV, XLSX, and JSON export options walks through the trade-offs.

Before you export, filter. Trim the list to the segment you actually want to contact so the AI is not writing to restaurants and law firms in the same batch. A tighter file produces a tighter sequence, and it keeps your token usage down when you paste it into a model.

Draft a sequence in ChatGPT

ChatGPT is the quickest way to see results. Open a new chat, attach your CSV or Excel export, and give it a clear brief. File uploads let the model read the whole spreadsheet, and its structured outputs are strong enough to return one ready-to-send cadence, as the OpenAI text generation guide describes.

A prompt that works well looks like this:

"Attached is a CSV of local businesses I want to reach. Write a 4-touch cold outreach sequence for each row: email 1 introduction, email 2 value, email 3 case study angle, email 4 breakup. Personalize the opening line using the business name, category, and city. Keep each email under 90 words and return the result as a table with one row per business."

Because the model can see every column, it grounds each opener in a real business instead of a placeholder. You can then refine the tone in a follow-up message: shorter, more casual, or industry-specific. For a deeper treatment of writing the copy itself, see how to write cold emails with ChatGPT from scraped data.

The limit of the chat window is scale. Paste 40 businesses and it works. Paste 2,000 and the model truncates, loses columns, or gets slow. That is where you move to a scripted approach.

Generate at volume with Claude Code

When you have a folder full of exports, Claude Code is the better tool. It runs in your terminal, reads files directly from disk, and can write new files back out. That means it does not just answer once: it can iterate over every record in your JSON and produce a personalized sequence per business, saved as its own file.

A working instruction to Claude Code reads plainly:

"Read leads.json in this folder. For each business, generate a 4-email outreach sequence personalized with its name, category, and city. Write each sequence to a file named after the business slug inside an /outreach folder. Use a warm, concise B2B tone and keep every email under 90 words."

Claude Code loops through the records, keeps the same proven structure on every one, and hands you a tidy folder of drafts. Because it works from JSON, nothing is truncated and no column is dropped. You can re-run it with a tweaked tone in seconds. To go further, our walkthrough on analyzing leads with Claude Code shows how to score and segment the same file before you generate.

The two tools pair naturally. Prototype the sequence in ChatGPT, lock the format, then hand that format to Claude Code to run across the whole list. That split covers the bulk of this workflow.

Codex and local models for the last mile

Two smaller options round out the workflow. Codex-style coding assistants can enrich the export before generation: dedupe rows, normalize phone formats, or split one big file into per-city batches, so the sequence generation starts from cleaner input.

A local model running on your own machine is worth it when the list contains data you would rather not send to a hosted API. You lose some polish, but you keep the file entirely offline and still get a serviceable first draft you can hand-edit. For most people these are edge cases: ChatGPT and Claude Code do the heavy lifting.

ChatGPT vs Claude Code for sequences

Whatever the model drafts, remember that outreach to businesses still has to follow the rules. The US CAN-SPAM compliance guide covers what every commercial email needs, and it is worth confirming your scrape and send are on solid ground with our note on whether it is legal to scrape Google Maps.

Where the Vonsel dashboard takes over

A folder of AI drafts is a great start, but it is still just text. Inside the Vonsel dashboard, the same captured leads live in a mapped CRM. Email Intelligence can generate an AI email tailored to each business, and Review Intelligence surfaces the pain points hiding in a company's public reviews, so your sequence opens with something the prospect actually cares about.

That is the difference between a one-off batch of drafts and a repeatable engine. You reopen a scrape to add more businesses, layer in review context, track who replied, and route follow-ups. The extension feeds the machine. The dashboard is the machine. For the bigger picture on turning a raw file into value, read feed your AI with business data and how to write an AI email for each business.