How to Switch AI Models Without Losing Your Chats, Projects and Settings: A Checklist
A before, during and after checklist for switching AI models in ChatGPT, Claude and Gemini — export your chats first, save project instructions, use the right picker, and test the new model safely.
Switching AI models does not have to mean losing your chats, projects and settings, as long as you back up first and test the new model before you trust it.
Key points
- Back up before you switch: a ChatGPT export includes your chat history, and its download link expires 24 hours after it arrives.
- Projects keep chats, files and instructions together in ChatGPT and Claude, so work there is easier to find after a change.
- Gemini keeps your chosen model for the current conversation, and Claude lets you change model within a chat.
Your work is safer when it lives in a project, not only in one long chat
Projects hold chats, files and instructions together, so a model change does not scatter them.
The fear behind every model switch is the same: weeks of careful chats, gone because you clicked a different name in a menu. A single long chat feels fragile, because everything you value sits in one scroll of text.
Projects are the calmer place to keep real work. OpenAI says ChatGPT projects organise chats, files and instructions together. Anthropic describes Claude projects as self-contained workspaces with their own chat histories and knowledge bases, where you can upload documents and provide context. If your brief already lives in a project, you are not relying on one chat surviving a change. Our pillar on how AI model updates work explains why versions keep moving.
A backup you control is the one step no switch should skip
Export first, check the file, and only then change anything important.
Make your own copy before you switch. In ChatGPT on the web: sign in, open your profile menu, select Settings, then Data controls, and under Export data select Export, then Confirm export. ChatGPT sends an email or SMS when it is ready. OpenAI warns exports can take up to 7 days, and the download link expires 24 hours after you receive it, so download promptly, signed in to the same account.
The downloaded ZIP includes your chat history and other relevant account data. Open it and check it contains the conversations you need. Self-service export is available for Free, Go, Plus, Pro and eligible Edu workspaces, not for Business, Enterprise or Healthcare workspaces, where OpenAI says to contact the workspace owner. You can also use OpenAI’s Privacy Portal. Our guide to AI model names helps you decode what you are switching between.
Follow this checklist before, during and after you switch
Ten small, testable steps turn a risky click into a reversible experiment.
Before you switch:
- Write down the model you use now, exactly as shown in the picker. Our guide on how to tell which AI model version you are using shows where to look.
- Copy your project instructions into a note you control, outside the app.
- Run the ChatGPT export above, download the ZIP, and open it to confirm key conversations are inside.
- In Gemini, pin and rename the chats you return to, so they stay easy to find in the side panel.
- Save a short handover summary of each important chat: goal, decisions made, and next step.
During the switch:
- Change the model using the app’s own picker or model menu, not by deleting the chat.
- Ask one question you already know a good answer to, and compare the reply.
- Check your project files and instructions are still the ones the chat is using.
After the switch:
- Re-run one real task, not just a test question, and read the result properly.
- Keep the old notes for a few days, so you can choose again in the picker if the new model disappoints — see the mistakes people make when a new AI model launches.
Each app handles the switch itself in a different place
The picker moves between apps, and what stays with the conversation is stated clearly only in some cases.
In Gemini, Google is explicit about continuity. With a Google AI plan, you can manually switch models: go to gemini.google.com, at the bottom of the chat box click the currently selected model, click the model you want, and optionally select a thinking level of Low, Medium or High. Google says Gemini will continue to use that model and thinking level during the current conversation, including if you later return to it.
In Claude, Anthropic says the model menu next to the send button controls the model, its effort, and whether it uses thinking, and that you can still change model and effort within a chat. Higher effort means more thorough responses, but they take longer and reach usage limits faster.
In ChatGPT, OpenAI places the picker in the message composer on the web and at the top of the conversation on iOS and Android, with options depending on workspace settings and rollout status. Where an official page does not promise what happens to a chat’s history when you change model mid-chat, this guide does not promise it either. That is why the export comes first.
What this means
A model switch feels risky because your work feels personal, and treating it like moving house makes it manageable.
Let us be honest about why this worries people. A good AI chat is not just text. It is the essay you have revised six times, the business plan you argued into shape late at night, the study notes that finally make sense. Losing that would not feel like a software glitch. It would feel like losing work, because it is work.
The reassuring truth is that most of what people fear losing is not tied to a model. Your chats live in your account history. Your project files and instructions live in the project. The model is the part that answers, and it is the part providers expect you to change. Gemini even documents that your chosen model sticks with the conversation when you return to it. The danger is rarely the switch itself. It is switching blind, with no export and no note of what you were using.
In our view, the healthiest habit is to keep a thin layer of your work outside any one app: an exported backup, copied instructions, and a one-paragraph summary of each important project. That sounds dull, in the way locking your front door is dull. It takes ten minutes, and it turns panic into a shrug. If the new model writes differently, you notice on a test question, not on a deadline.
There is freedom in this routine, too. Once your work is backed up and organised, trying a new model stops being a leap of faith and becomes a choice you can undo. Give the deeper model the difficult draft, keep the faster one for everyday questions, and return to your old favourite when it suits the task. Not one voice you cling to, but a small team you direct, with your judgement in charge.
Bottom line
Back up, note your settings, switch through the picker, and test before you trust.
Export your ChatGPT data and check the ZIP, copy your project instructions somewhere you control, organise the chats that matter, then change model through the picker and test it on work you understand. A switch becomes a reversible choice, not a risk.
Frequently Asked Questions
Quick answers on safety, deleted chats, plans and expired links.
Is my exported ChatGPT file safe to share with anyone?
No, you should not share it freely. OpenAI warns downloaded data can include sensitive account information, and says to keep your download link and exported files secure. Store the ZIP somewhere private, treat it like personal documents, and only open it on devices you trust.
Can an export restore a ChatGPT chat I already deleted?
No, it cannot. OpenAI states an export cannot restore chats that have already been deleted. That is why the order matters: export while the conversations you need are still in your account, open the ZIP, and confirm the important chats are inside before deleting anything.
Do I need a paid plan to switch models manually in Gemini?
Yes, manual switching needs a Google AI plan. Google says users with a plan can manually switch models in Chat Mode and Task Mode, while accounts without a plan use Auto or Flash-Lite in chats. Auto still routes each prompt to an efficient model, but you do not choose it yourself.
Can I still download my ChatGPT export after the link expires?
No, an expired link will not work. OpenAI says the link expires 24 hours after you receive the message, and that if it has expired you should request a new export through ChatGPT settings or the Privacy Portal. Download promptly, while signed in to the requesting account.
{“@context”:”https://schema.org”,”@type”:”FAQPage”,”mainEntity”:[{“@type”:”Question”,”name”:”Is my exported ChatGPT file safe to share with anyone?”,”acceptedAnswer”:{“@type”:”Answer”,”text”:”No, you should not share it freely. OpenAI warns downloaded data can include sensitive account information, and says to keep your download link and exported files secure. Store the ZIP somewhere private, treat it like personal documents, and only open it on devices you trust.”}},{“@type”:”Question”,”name”:”Can an export restore a ChatGPT chat I already deleted?”,”acceptedAnswer”:{“@type”:”Answer”,”text”:”No, it cannot. OpenAI states an export cannot restore chats that have already been deleted. That is why the order matters: export while the conversations you need are still in your account, open the ZIP, and confirm the important chats are inside before deleting anything.”}},{“@type”:”Question”,”name”:”Do I need a paid plan to switch models manually in Gemini?”,”acceptedAnswer”:{“@type”:”Answer”,”text”:”Yes, manual switching needs a Google AI plan. Google says users with a plan can manually switch models in Chat Mode and Task Mode, while accounts without a plan use Auto or Flash-Lite in chats. Auto still routes each prompt to an efficient model, but you do not choose it yourself.”}},{“@type”:”Question”,”name”:”Can I still download my ChatGPT export after the link expires?”,”acceptedAnswer”:{“@type”:”Answer”,”text”:”No, an expired link will not work. OpenAI says the link expires 24 hours after you receive the message, and that if it has expired you should request a new export through ChatGPT settings or the Privacy Portal. Download promptly, while signed in to the requesting account.”}}]}
Related Guides
Keep exploring how model updates, names and versions work.
- AI Model Updates Explained: How ChatGPT, Gemini and Claude Get New Versions
- How to Tell Which AI Model Version You Are Using
- AI Model Names Explained: GPT, Gemini, Claude, Llama and More
- Mistakes People Make When a New AI Model Launches
Sources
The official help pages behind the steps above.
