Turning an empty Gmail inbox into a searchable log of every AI prompt worth keeping
The problem this solves
If you use more than one AI tool — say, refining a prompt in ChatGPT, then reusing it in Claude or Gemini — the good prompts and their results tend to just vanish. A month later you remember “I had a great prompt for this” but not the exact wording, which tool produced it, or where the result went. There’s no dedicated place any of it lives.
The concept, in plain terms
This system uses Gmail — not a new app, just the Gmail you already have — as that missing storage place. The whole idea in three sentences:
- When a prompt gives you a result worth keeping, you email it to yourself in a simple format.
- Depending on which address you send it to, Gmail automatically sorts and labels it — new vs. proven — with zero manual filing.
- Later, you search your inbox like a database, and the prompt, the tool that made it, and a link to the result are all sitting right there.
That’s it. Everything below is just the mechanics of setting that up.
Why Gmail specifically
Three Gmail features do all the actual work here, for free, with nothing to install:
- Filters — rules that auto-label or archive incoming mail based on things like who sent it or which address it was sent to.
- Plus-addressing — Gmail lets you add
+anythingbefore the@in your own address (e.g.you+capture@gmail.com), and mail sent to it still lands in your normal inbox. Filters can then target that exact address. - Threading & search — replying to your own email keeps a running history, and Gmail’s search is strong enough that you rarely need rigid folders.
How it works, step by step
- Get a great result. You’ve been refining a prompt in some AI tool and it finally nails what you needed.
- Email it to yourself, using the template below, to one of two addresses depending on status (see below).
- Gmail auto-labels it based on which address you used — no manual tagging required.
- Big files (PDFs, spreadsheets, context docs) go in a Google Drive folder, not as email attachments — the email just links to the folder.
- Search later using the label, a project tag, or a keyword, same as searching anything else in Gmail.
Setting it up
1. Two addresses (using Gmail plus-addressing)
Skip separate accounts or domains pretending to be different systems — that pattern looks a lot like phishing to mail providers, and it doesn’t buy you anything plus-addressing on one account doesn’t already do:
you+capture@gmail.com→ default landing spot for anything newyou+proven@gmail.com→ send here once something’s tested and actually works
Set up a Gmail filter for each (Settings → Filters and Blocked Addresses → Create filter): “sent to you+proven@gmail.com” → apply label proven. That’s the entire setup — a couple of filters, no new software.
2. The capture template
Paste this into the email body every time:
PROMPT:
[paste the final version]
TOOL: [which AI tool produced the result you’re keeping] OUTPUT: [link to the result] CONTEXT: [Google Drive folder link, if files were involved] NOTES: [optional]
Only the prompt is mandatory — everything else is optional, to keep this under 10 seconds.
Subject line: #project — short description. Leave the tool name out of the subject; it’s already in the body, and Gmail’s full-text search finds it regardless.
3. Files and outputs live in Drive, not Gmail
Attachments eat storage and don’t version well. Instead: one Google Drive folder per project, created once, linked from every capture email. Gmail holds the searchable text log; Drive holds the actual files.
4. Version history via threads
Keep replying to the same email thread while refining one line of thought — the thread becomes a running revision history for free. The moment you branch (trying a different fix in a different tool), start a new thread instead and paste a link to the original in the NOTES field. Gmail threading can’t represent branches, so don’t force it to.
5. If you want to automate this later
Eventually you might want a script to auto-file things to Drive or send a weekly digest. Two rules if you go there:
- Never authenticate by “From:” address — it’s trivially fake. Trigger automation off your own labels or your own authorized script, not the claimed sender.
- Never let automation execute open-ended instructions straight from an email’s content. Fine: “turn this into a Doc.” Not fine: “do whatever this email says” — that’s how a random phishing email hijacks the system.
6. Weekly backup ritual
You’ll forget to capture things in the moment sometimes — everyone does. Once a week:
- Skim your AI chat histories (most tools keep their own).
- Backfill anything worth keeping.
- Pick your best prompt of the week and copy the final version into a running “Playbook” doc per project — a clean, curated reference, separate from the raw inbox log.
Ideas worth trying later
- Weekly digest — a script that emails you a summary of the week’s captures, grouped by label.
- Browser bookmarklet — select text, click once, pre-fill a
mailto:to your capture address. - Search cheat sheet:
label:proven #project-x,has:attachment label:project-x,is:starredas a lighter-weight status flag if you’d rather skip the second address. - Playbook docs — one living doc per project with only the proven prompts, rewritten clean — the actual reusable asset over time.
What changed from the first draft, and why
| Problem | Fix |
|---|---|
| Extra copy-paste step on every prompt | Only capture after something works, not while iterating |
| Multiple lookalike email domains looked like phishing | One account, Gmail plus-addressing |
| “From:” address used as security | Automation triggers on your own labels/account, never the sender field |
| Tagging discipline assumed to hold | One mandatory tag; status comes from which address you send to |
| Linear threads forced onto branching work | New thread + cross-link convention |
| Large files clogging email | Files live in Drive, linked not attached |
| Productizing before validating | Left out — prove it works for you first |