It can be, but only once you have changed two settings, and the safety promise you have probably read about does not cover you if you live in the UK or the EU. The “not used for training” rule for ChatGPT health data belongs to ChatGPT Health, a separate space open only to adults in the US. Connect your watch data anywhere else in ChatGPT and it is an ordinary chat: eligible for model training by default, open to Memory, and not un-trained by deleting the chat afterwards.
That is worth knowing before you hand over 90 days of heart rate, sleep and HRV. Asking ChatGPT about your training is genuinely useful. This article covers what happens to that data once it is there, where ChatGPT health data rules apply and where they do not, and the controls that are yours to set.
Key Takeaways
- ChatGPT Health keeps health chats, memories and connected data out of model training, but OpenAI offers it only to users aged 18 and over in the US. It is not available in the UK, the EEA or Switzerland.
- In ordinary ChatGPT, the “Improve the model for everyone” setting is on by default for personal plans, so your training data can be used to train future models.
- Data from a connected app or connector becomes part of the conversation once ChatGPT uses it in a reply, and from then on it follows your training setting.
- Deleting a chat does not delete the Memory entries ChatGPT saved from it, and switching training off does not undo training that has already happened.
- Temporary Chat keeps a conversation out of training, history and Memory, and is deleted after 30 days.
What ChatGPT Health Promises, and Who Actually Gets It
ChatGPT Health is a separate space inside ChatGPT where health chats, files, connected apps and memories are kept apart and are not used to train OpenAI’s foundation models. OpenAI launched it in January 2026, and its help centre lists who can use it: logged-in Free, Go, Plus and Pro users, aged 18 or over, in the US, on the web and iOS. Users in the European Economic Area, Switzerland and the United Kingdom are not eligible.
When we searched “ChatGPT health data” in October 2026, most of the results were coverage of that launch, so it is easy to read the no-training promise as a rule for the whole product. It is not. Health has its own memories, and OpenAI says Health information does not flow back into your main chats. The reverse does not hold the same way: your main chats sit under the normal rules.
So the honest answer depends on where you are. A runner in Denver who connects their data inside Health gets the stronger protection. A HYROX athlete in Manchester or Munich cannot open Health at all, so every conversation about their training happens in ordinary ChatGPT.
What Happens to Your Training Data in an Ordinary ChatGPT Chat
In ordinary ChatGPT, your conversations can be used to train OpenAI’s models unless you switch off “Improve the model for everyone” in Settings, Data controls. OpenAI turns it on by default for personal plans (Free, Plus and Pro). Business, Enterprise and Edu workspaces are not trained on by default.
How your data arrives makes less difference than people expect. You can paste a Garmin export, upload a CSV, or connect a service through an app or a custom MCP server. OpenAI’s own apps documentation draws the line here: it does not train on data synced from a connected app unless that data becomes part of your ChatGPT conversation, for example in a response, and your training setting is on. When you ask about last week’s long run, the whole point is that the data ends up in the reply.
This table sets the two spaces side by side.
| ChatGPT Health | Ordinary ChatGPT | |
|---|---|---|
| Who can use it | 18+, US only | Everyone with an account |
| Used to train models | No | Yes, unless you switch training off |
| Memory | Separate Health memories | Shared across your chats |
| Connected apps | Kept inside Health | Data used in replies joins the conversation |
| Deleting a chat | Removes it from Health | Removes the chat, not saved memories or past training |
The row that matters for most readers of this blog is the second one. Outside the US, the training switch is the only thing between your sleep and HRV and the next model.
ChatGPT Memory Can Keep What Your Watch Data Revealed
ChatGPT Memory can save facts about you from a conversation and carry them into future chats, and it does this whether or not your data is used for training. OpenAI treats Memory and model training as two separate controls. Switching off training does not stop ChatGPT from saving memories, and switching off Memory does not opt you out of training.
For an athlete this matters because the useful questions are revealing ones. “Why has my resting heart rate been up for a week?” can leave a saved note about an illness. “Is my HRV low because of the new medication?” can leave one about the medication.
Memory often writes these down without being asked. A September 2026 audit of 179,057 ChatGPT conversations found that 21.31% contained personal health data. It also found that over 95% of memory profile entries had been extracted implicitly, without the user prompting ChatGPT to remember anything. The sample came from India, Nigeria, Brazil and Pakistan, so the percentages will differ elsewhere, but the mechanism is the same product.
You can see what ChatGPT has saved in its Memory settings and delete entries one by one. It is worth looking after any conversation about an injury.
Deleting a Chat Does Not Undo Training
Deleting a ChatGPT chat removes it from your account straight away and from OpenAI’s systems within 30 days, but it does not remove anything already used for training, and it does not delete the memories saved from it. OpenAI’s deletion article adds two exceptions to the 30 days: chats that have already been de-identified and disassociated from your account, and chats it must keep for security or legal reasons.
Switching training off works the same way. It applies to conversations from that point on, and it is not retroactive. A model that has already learned from a conversation is not retrained when you change the setting.
Take Priya, who is training for her first HYROX in Glasgow. She connects her training to ChatGPT at the start of March, asks it about her runs for eight weeks with every default left on, then reads about the training setting and turns it off at the start of May. Her chats from May onwards stay out of training. The eight weeks before that were eligible the whole time, and deleting them now removes them from her history without reaching back into anything they were already used for.
The lesson is about order. The settings protect what comes after them, so they go first, before the first question.
Is Training Data Even Health Data?
Some of it clearly is. Pace, distance and lap splits say little about your health on their own. Heart rate, HRV, resting heart rate, sleep and anything you say about an injury or illness can reveal a lot. Under UK and EU GDPR, data concerning health is a special category that needs a stronger legal basis, usually your explicit consent.
That is why it pays to split your data in two before it goes anywhere. Training data answers most of the questions athletes actually ask: am I running enough, was this week harder than last, what would I run at HYROX today. Recovery data answers fewer questions and carries more of the risk.
What You Control
ChatGPT privacy comes down to four things you control: what leaves your training app, whether ChatGPT may train on it, what ChatGPT remembers, and whether the connection stays on. The first is set before the data reaches ChatGPT, which makes it the strongest.
Before It Reaches ChatGPT
Choose a connection that lets you decide what is shared, because not every ChatGPT connector asks. When you connect Kracey to ChatGPT as a custom MCP server , the consent screen shows two kinds of data. Training data (sessions, laps, pace, volume, best efforts and race predictions) is always included. Heart and recovery data (heart rate, HRV, sleep, resting heart rate, VO2max and load) is a separate tick, and it stays off unless you choose it.
Some things are never shared through any scope: your conversations with the Kracey coach, what the coach remembers about you, and anything you told it about an injury or illness. The connector is read-only, and every tool and the scope it needs are listed in the connector docs . If you use a Garmin, the Garmin page explains how your sessions reach it through intervals.icu without your Garmin password.
Inside ChatGPT
- Model training. Turn off “Improve the model for everyone” under Settings, Data controls, before your first question.
- Memory. Review saved memories after any chat about illness, injury or medication, or switch Memory off.
- Temporary Chat. Use it for one-off questions. It does not appear in your history, does not create memories, is not used for training, and is deleted after 30 days.
Afterwards
- Disconnect. Remove the connection from Kracey’s Settings and it stops working at once.
- Delete. Delete chats you no longer need, knowing what deletion does and does not reach.
A step-by-step lock-down checklist for these settings is on its way to the blog.
FAQ
No. OpenAI's help centre lists Health as available to users aged 18 and over in the US, and says users in the UK, the European Economic Area and Switzerland are not eligible. UK and EU athletes use ordinary ChatGPT, where the standard training and Memory settings apply.
It can. Data from a connected app or connector is not trained on while it only sits in the connection, but once ChatGPT uses it in a reply it is part of your conversation. With "Improve the model for everyone" on, which is the default on personal plans, that conversation can be used for training.
Mostly, within 30 days, but not completely. Deleted chats are removed from OpenAI's systems within 30 days unless they were already de-identified or must be kept for legal or security reasons. Deleting a chat does not delete the memories saved from it, and it does not undo any training already done.
It is the most private option in ordinary ChatGPT. Temporary chats are not saved to your history, do not create memories, are not used for training and are deleted after 30 days. OpenAI may still review them for abuse during those 30 days.
Only if your question needs it. Most training questions are answered from sessions, pace and volume alone. Share recovery data when you want ChatGPT to read your HRV or sleep, and set the training and Memory controls first. ---
Your Data, Your Call
Connecting your training data to ChatGPT is reasonable once you know the rules: ChatGPT health data protections live inside ChatGPT Health, Health is US-only, and everywhere else the defaults favour training and memory. Turn training off before the first question, check Memory after a sensitive one, use Temporary Chat for one-offs, and share heart and recovery data only when the question needs it.
If you would rather your training data never trained anyone’s model, that is how the Kracey coach handles it. It is an AI HYROX, running and endurance coach in Telegram that reads your training through intervals.icu and messages you first. Your data is stored in Frankfurt and is never used to train AI models, as our data page sets out in full, with the receipts on the security page . You get 14 free days, no card, and they start when your coach can see your training.