Ink and watercolour illustration of one robot handing a bundle of notes to another robot over a desk with an open tabbed ledger

Quick answer

Keep a simple dated log for each important prompt or AI skill. Every time you change it, write down what you changed, why, what you expected and what actually happened, then save the full new version. When you move to a new AI model, re-run a few saved test tasks and use the log to see which changes still help, rather than starting from scratch.

Key takeaways

  • Record the reason and the result, not just the new wording. That is what makes a lesson reusable.
  • Keep three to five test tasks for each prompt so you can compare versions fairly.
  • Store lessons outside any one AI tool, in a document or spreadsheet you own.
  • Re-check after a model switch. Some fixes become unnecessary; a few may need undoing.

📘 Main article: This guide follows our news round-up Has Google Built Self-Improving AI? The SIM Paper, the Gestalt Stealth Model and This Week’s AI News, where Google’s research showed that AI systems improve more steadily when they remember past changes.

Most of us tweak our prompts constantly: add a line here, remove a rule there. A month later we can’t remember why the prompt looks the way it does, and when a new model arrives we either keep everything blindly or start again. Google’s Self-Improvement Memory research found that an AI optimiser did better when it kept a record of each change, its reason and its result, and those lessons even carried over to a different model. You can borrow that idea with nothing more than a spreadsheet.

Why keep a prompt change log?

A change log turns trial and error into knowledge you keep. It helps you:

  • avoid repeating fixes that didn’t work;
  • undo a change that made things worse;
  • explain to a colleague or assistant why a rule exists;
  • move to a cheaper or newer model without losing months of tuning.

What should each entry record?

FieldExample
Date and version22 Nov 2026, v7
Prompt or skillWeekly newsletter draft
What changedAdded: “Keep each section under 120 words.”
WhyDrafts were too long to read on a phone
What you expectedShorter sections, same key points
What happenedSections shorter; one tip got cut, so added “keep all tips”
Model usedThe model and version you tested with
Keep, revise or undo?Keep

💡 Tip: The “why” and “what happened” columns matter most. Without them, the log is just a list of old versions.

How do you set up a change log?

  1. Pick your important prompts. Start with the two or three you use every week, not all of them.
  2. Create one spreadsheet or document with the columns above. A Google Sheet, Notion table or plain document is fine.
  3. Save the current full version of each prompt as version 1.
  4. Write three to five test tasks for each prompt: real examples, including at least one awkward one.
  5. Change one thing at a time. Run your test tasks, then fill in the log before moving on.
  6. Save each new full version with its version number so you can roll back easily.

If you use AI skill files, the same approach works; our guide on improving an AI skill without filling it with one-off rules pairs well with this.

How do you use the log when you switch AI model?

Treat a new model as a fresh test, using your log as the checklist.

  1. Run your test tasks on the new model with your latest prompt.
  2. Read down the log. For each past fix, ask: is the problem it solved still there?
  3. Try removing fixes that look unnecessary, one at a time, and re-test. Newer models often need fewer instructions.
  4. Note the new model’s name in the log and record anything you change.
  5. Keep the old model’s final version saved in case you need to switch back.

This is also a good moment to decide which jobs need a premium model at all; see How to Split a Project Between a Fast, Cheap AI Model and a Premium One.

⚠️ Warning: Don’t assume a fix that worked on one model will work on another. The Google researchers found many lessons transferred, not all of them. Always re-test.

How do you keep the log from becoming a chore?

  • Only log meaningful changes, not typo fixes.
  • Batch your notes: five minutes at the end of a session is enough.
  • Ask your AI to help. Paste the old and new versions and ask it to fill in “what changed” for you; you add the why and the result.
  • Review monthly and delete rules that no longer earn their place.

Frequently asked questions

Do I need special software for a prompt change log?

No. A spreadsheet or simple document works well. What matters is recording the reason and result for each change.

How many test tasks do I need?

Three to five real examples per prompt is usually enough to spot whether a change helps or hurts.

Should I keep old prompt versions?

Yes. Saving the full text of each version lets you roll back quickly if a change makes results worse.

Will my prompts work the same on a new AI model?

Often mostly, but not always. Re-run your test tasks and use your log to decide which fixes to keep.

The bottom line

The difference between people who get steadily better results from AI and those who keep starting over is often just memory. A simple change log, with reasons and results, means every improvement you make is yours to keep, whichever model you use next.

author avatar
Garry Knight
I'm Garry Knight, the person behind Prodify Digital. I write about email list building, email marketing, SEO, AI search and the tools that connect them. My aim is to make online marketing easier to understand, so creators and small business owners can make informed decisions about building an audience and keeping people engaged. Here you'll find straightforward guides and product reviews that explain what something does, where it fits and which limitations matter. The focus is on clear explanations and useful next steps—not hype, shortcuts or promises of easy earnings.

Leave a Reply