Visitor card digitizer — GPT Project
Sign in to track installsSnap a photo of your stack of paper visitor cards. The GPT extracts the names, emails, phone numbers, and notes into a clean CSV ready to import straight into Brevo or any contacts tool — no typing, no guessing handwriting. Copy each block into ChatGPT's Custom GPT builder — Name, Instructions, Capabilities — in order.
Quick install.
Grab the install file below — drop it where the setup guide tells you.
Filename: visitor-card-digitizer-gpt-paste.txt
Setup guide.
Step-by-step from “fresh install” to “first run.” Skim it once, then keep this page open while you go.
Before you start
This GPT reads photos of paper visitor cards and outputs a CSV ready for Brevo import. You'll need a ChatGPT Plus or Team account (custom GPTs require Plus+) and any phone with a camera. Setup is 5 paste-in fields, takes 5 minutes.
Step 1 — Create the GPT
In ChatGPT, profile menu → My GPTs → Create a GPT. Click the 'Configure' tab at the top (skip the conversational setup). You'll see fields for Name, Description, Instructions, Conversation starters, and Capabilities.
Step 2 — Paste the 5 fields
Open the .txt file from the Quick Install button. It's structured as 5 labelled blocks. Copy each block into the matching ChatGPT field. Conversation Starters takes one starter per line. For Capabilities, leave Web Browsing / Code Interpreter / DALL·E all OFF — the file explains why.
Step 3 — Test with one card before you use a stack
Save the GPT. Open it. Upload one well-lit photo of a single card and ask it to extract. Verify the CSV looks right and the confidence flag matches your impression of the handwriting. Once that works, scale up to a stack — multiple cards laid out on a table in one photo work, but the more cards per photo, the more chance of low-confidence reads.
Personal data — handle accordingly
Troubleshooting.
Hit one of these? Here's the fix.
Lots of fields come back blank with low-confidence flags
Photo quality is the lever. Re-shoot with better lighting (natural light, no shadow on the card), the camera straight overhead (not at an angle), and one card per photo for hard-to-read handwriting. The Instructions field explicitly tells the GPT to flag rather than guess — that's the right behaviour, but it surfaces photo-quality issues fast.
GPT extracts emails that don't match what's actually on the card
This is a real failure — the Instructions field says 'never invent a field'. If you see invented emails, the Instructions probably weren't fully pasted (the 'Confidence handling' section should be present). Re-save the GPT with the full Instructions. Until then, manually verify every email row against the photo before importing to Brevo.
Discussion.
Per-integration ratings and comments open with the v3 community release. For now, send feedback to [email protected] — David reads every one.