JSON vs CSV for AI analysis Which export actually feeds your AI better?

CSV for quick tables, JSON for structure

Both formats hold the same Google Maps data: business name, address, phone, website, rating, review count, category, and coordinates. The difference is shape. JSON nests and types its values, while CSV is a flat grid of rows and columns.

For most sales questions you throw at ChatGPT, a CSV is perfect. It is compact, the model reads it as a spreadsheet, and it burns fewer tokens. When your leads carry a list of reviews per business, or two phone numbers, or a nested owner reply, JSON keeps that intact so nothing gets flattened into a single messy cell.

Export both from the same capture and choose per task. That is the point of having options in the extension: you decide the format at the moment you know the question, not before.

JSON vs CSV for lead analysis

Rule of thumb: if the answer fits in a spreadsheet, reach for CSV. If the answer depends on structure inside each record, reach for JSON. For a deeper look at every option the extension offers, see CSV, XLSX and JSON export options.

Feed a CSV to ChatGPT

This is the fastest path and it covers roughly eight in ten of the questions a salesperson asks. Capture your leads with the extension, click export, and pick CSV or Excel. Then open ChatGPT, attach the file, and prompt it plainly.

Prompts that work well on a flat lead table:

"Sort these businesses by rating, then flag every one with no website in a new column."

"Group by category and tell me which niche has the most leads with under 20 reviews."

"List the 15 businesses most likely to need a website redesign and explain why in one line each."

ChatGPT reads the columns, runs the logic, and hands back a clean table you can copy straight into your outreach list. Keep the file to a few thousand rows for reliable results. Our step-by-step guide to uploading a lead CSV to ChatGPT covers the edge cases, and analyzing lead lists with ChatGPT has more prompt recipes.

Work JSON with Claude Code

When your task is repeatable, programmatic, or involves nested review data, export JSON and hand it to Claude Code in your terminal. Claude Code can parse the file, run transforms, and even write a small script you can reuse on next week's capture.

Drop the JSON in a folder, open Claude Code there, and ask:

"Read leads.json, dedupe by phone number, and write leads_clean.json with a review_count field added."

"Score each business from 0 to 100 as a web-design prospect based on rating, review count, and whether a website is present. Output a ranked CSV."

"Build a simple HTML page that plots every lead on a map from the coordinates in this JSON."

Because JSON preserves types and nesting, Claude Code never has to guess where a field starts and ends. That matters most when each business carries an array of reviews and you want sentiment or pain points pulled out per record. See analyzing leads with Claude Code for full command examples.

Codex and local models

The same files work anywhere. If you use Codex or another coding assistant, JSON is the friendly input because it maps cleanly to a data structure the model can iterate over. If you run a local model on your own machine for privacy, a small CSV keeps the context window light and the responses fast.

The principle does not change with the tool: flat questions like CSV, structured questions like JSON. Whatever you point at the file, the extension has already written it in a clean, valid format so the model spends its effort on your question, not on repairing broken rows.

The file is the means, not the end

Exporting a file and prompting an AI is a great way to answer a one-off question. But a spreadsheet in a chat window forgets everything the moment you close the tab. That is where the export stops and the Vonsel dashboard begins.

Send the same capture into Vonsel and every lead lands in a mapped CRM you can work for months. Reviews Intelligence reads the review text and surfaces pain points per business. Email Intelligence writes a tailored cold email for each one. Sales Intelligence scores and segments the list. You can reopen a scrape later to add more businesses, attach reviews to captured leads, and keep a full history of what you pulled and when.

In other words, the CSV or JSON is the raw material. The value-added, contextualized database is the finished product. For the philosophy behind that, read value-added database vs raw CSV and the broader idea of feeding your AI with business data.

Which one for which job