GuidesChoosing a modelAnyone with saved prompts, routines or a product built on a particular AI model
When the model you rely on is switched off
For Anyone with saved prompts, routines or a product built on a particular AI model.
Start here
- What this helps you do
- Providers retire older models on a published schedule. Anything you built on one will eventually stop working or, more often, quietly start answering differently. Knowing it is coming turns a bad surprise into a task you schedule.
- What you leave with
- A repeatable workflow and a starter prompt you can adapt to your own work.
- What to check before trusting it
- You know which of your prompts, routines and documents actually matter enough to test
Best fit today: ChatGPT, Claude, Gemini, Grok or Kimi.Our read of the reviewed tool guides, not tested by us. Reviewed 21 September 2026. Open a tool guide for what each one does with this kind of work.
When to use this
Something you have relied on for months starts giving different answers to the same request, or stops working altogether, and nothing on your end changed.
- Providers publish what is being retired and what to move to, with notice
- A newer model can usually take on the same work once you have checked it
- The model can re-run your saved prompts and compare the results against an answer you already trust
Workflow
List what you depend on: saved prompts, custom assistants, scheduled routines, anything wired into a product
Note which model each one uses and where it runs, because the same model can retire on different dates on different platforms
Read the provider's own deprecation page rather than a summary of it, and turn on their notifications
Keep one example of good output for each task that matters, so you have something to compare against
Well before the cutoff, run those prompts on the replacement and compare against your saved example
Reword anything that drifted, then switch over early rather than on the final day
Prompts to try
Use it, adapt it, and make it your own.
Here is a prompt I rely on, and an example of the output I consider correct. I am moving to a different model. Run the prompt, then compare the new output against my example and tell me where it differs in substance rather than wording. Flag anything that changed in a way that would matter for my work, and say plainly if nothing meaningful changed.
Human lens
- You know which of your prompts, routines and documents actually matter enough to test
- You decide what still working means here, because the change is usually silent rather than an error
- You choose when to move, so it happens on a day you picked rather than the day it stops
What to avoid
- The failure is usually silent. Same request, different answer, no error, so you often hear about it from a person rather than an alert
- A model retiring on one platform does not mean it retires everywhere on the same day
- Preview and experimental versions can be withdrawn at very short notice, so do not build anything you depend on around them
- It is not only models. Surrounding features, assistant tools and individual settings get retired too
Reviewed 21 September 2026