Clean Lead Data With ChatGPT Feed your AI a tidy list from Google Maps

Raw scraped lists are always a little dirty

Any list of local businesses pulled from Google Maps arrives with the usual mess: duplicate rows for the same shop, phone numbers in five formats, addresses that mix street and city into one field, and empty cells where a website should be. None of that is a scraping fault. It is just how public listings look in the wild.

The good news is that this is exactly the kind of tedious, pattern-based work that large language models are excellent at. Instead of writing spreadsheet formulas or regular expressions by hand, you describe what a clean row should look like and let the model rewrite the file. The trick is getting the data out in a format the AI can read, which is where the capture step matters.

With the Vonsel extension you scrape straight from a Google Maps search and export the results as CSV, Excel, or JSON in one click. That export is your raw material. Feed your AI with data, then let it hand the list back cleaner than it found it.

Export the right file for the job

Before you open any AI tool, decide which format fits. The extension gives you all three, and the choice changes how smooth the cleaning goes. A quick chat pass loves a flat CSV. A model that needs typed columns or nested review data prefers Excel or JSON.

If you plan to keep review text, social links, or opening hours attached to each business, export JSON. It preserves the structure so an AI can reason about nested fields without you flattening them first. For a plain name, phone, and address cleanup, CSV is the shortest path. Our guide on export formats for AI goes deeper on each one.

Clean the list in ChatGPT

Open ChatGPT, attach your exported file, and give it a specific brief. Vague prompts like "clean this" produce vague results. Tell the model exactly what a good row looks like and what to do with the bad ones. A prompt that works reliably reads something like this:

"Here is a CSV of local businesses scraped from Google Maps. Remove duplicate rows where name and phone match. Standardize every phone number to international format. Split the address column into street, city, and postal code. Flag any row missing a website in a new column called needs_website. Return the result as a downloadable CSV and a two-line summary of what you changed."

ChatGPT will process the file, show you a preview, and hand back a cleaned download. Because it explains what it changed, you can spot when it guessed wrong, for example merging two genuinely different branches that happen to share a head-office phone. Review the summary, then ask for a correction in plain language if needed. The OpenAI file uploads guide covers current size and type limits so you know how big a list one chat can swallow.

This same window is a natural place to keep working the leads. Once the list is tidy you can ask it to score each lead or draft outreach, turning a cleanup session into a full list analysis.

Scale it up with Claude Code

A single chat window is fine for a few hundred rows. When you scrape an entire city and land tens of thousands of businesses, you want a tool that works over the whole file at once without pasting it into a browser. That is where Claude Code shines.

Claude Code runs in your terminal and can read a file directly from disk. Point it at your exported leads.csv or leads.json and describe the cleanup the same way you would in chat. It will write and run a small script, process every row, and save a new file, so nothing is truncated by a chat length limit. A working brief looks like:

"Read leads.json in this folder. Deduplicate on place_id, normalize phone numbers to E.164, lowercase all email domains, drop rows with no phone and no website, and write the result to leads_clean.csv. Print how many rows you removed and why."

Because Claude Code executes code rather than eyeballing rows, it is deterministic and repeatable. Run the same command next month on a fresh scrape and you get consistent output. For teams that already lean on it, our walkthrough on analyzing leads with Claude Code shows how the cleaned file becomes the base for scoring and dashboards. Between the two tools, ChatGPT is your fast desk edit and Claude Code is your bulk pipeline.

Where Codex and local models fit

From clean file to a mapped CRM

Once the AI hands you a deduped, standardized list, you have raw material that is ready to sell to. The question is where it lives next. A clean CSV sitting in a downloads folder does not book meetings.

This is where the extension being the means and the Vonsel dashboard being the end comes together. Instead of leaving your polished list stranded in a spreadsheet, load it into the mapped CRM, where every business sits on a map you can carve into territories. From there the dashboard layers on what a chat window cannot: Reviews Intelligence that mines each business's ratings for pain points, an AI email tailored to each business, and value-added databases contextualized with those pains so your outreach lands.

You can also reopen a capture later and add more businesses, attach reviews to leads you already saved, and enrich websites for missing contact details. The AI cleanup gets your data honest. The dashboard turns honest data into a pipeline. If you are weighing this against buying a list, our piece on scraped data versus bought lists is worth a read, and if compliance is on your mind, start with is it legal to scrape Google Maps.