On Claude vs ChatGPT privacy, the two assistants differ most on the data a connector hands them. Claude does not train on raw connector output and its memory leaves health out by default. ChatGPT may train on what it reads from a connected app, and may remember it, unless you switch those settings off. On retention, age limits and business plans the gap is smaller, and we go through each below.
That matters because a watch connector sends more than a typed question does. Ninety days of heart rate, sleep and HRV can reveal an illness or an injury you never mentioned. This page compares the two assistants row by row, says which one suits which athlete, and says plainly who it is not for. Every row was checked against Anthropic’s and OpenAI’s own pages on 15 October 2026.
Key Takeaways
- Claude’s training data “does not include raw content from connectors”, including MCP servers, but it does include the conversation around them, so Claude’s own replies restating your heart rate can be trained on unless you opt out.
- ChatGPT “may use information accessed from apps to train our models” whenever “Improve the model for everyone” is on.
- By default Claude’s memory does not store health topics; ChatGPT’s memory can keep what it reads from an app, and OpenAI says sensitive information can appear in it.
- Both delete a deleted chat within 30 days, but Claude keeps de-identified training data for up to 5 years if you leave training on.
- Claude’s consumer apps are 18 and over; ChatGPT is 13 and over, with a parent’s permission under 18. Business plans are not trained on: Anthropic’s terms rule it out, and OpenAI does not by default.
Claude vs ChatGPT Privacy at a Glance
Claude is the more private default for data that arrives through a connector, and ChatGPT needs two settings changed to come close. The table below covers the consumer plans: Claude Free, Pro and Max, and ChatGPT Free, Go, Plus and Pro.
| Claude | ChatGPT | |
|---|---|---|
| Trains on your chats | Yes, unless you opt out | Yes, unless you switch off “Improve the model for everyone” |
| Trains on raw connector data | No | Yes, if that setting is on |
| Memory and health | Health left out by default | Can remember what it reads from an app |
| Private mode | Incognito: no training, no memory | Temporary Chat: no training, no new memories |
| Deleted chats removed | Within 30 days | Within 30 days |
| Kept for training | Up to 5 years, de-identified, if training is on | De-identified data is not covered by the 30 days |
| Minimum age | 18 | 13, with a parent’s permission under 18 |
| Business plans | Not trained on | Not trained on by default |
| Health product with a no-training promise | US only | US only, 18 and over |
The rows that separate them are the second and third. The rest are close enough that your settings matter more than your choice of assistant.
Model Training: Connector Data Is Where They Split
Both assistants can train on your conversations by default on consumer plans, but they treat connector data differently. Anthropic’s privacy centre says its training data “includes the entire related conversation” and “does not include raw content from connectors (e.g. Google Drive), including remote and local MCP servers”. OpenAI’s connector help page says it “may use information accessed from apps to train our models if your ‘Improve the model for everyone’ setting is on”.
So the raw pull from your watch stays out of Claude’s training, and it does not stay out of ChatGPT’s unless you change the setting. That is the single biggest difference between the two for an athlete.
Claude’s exclusion has a limit, and it is worth being exact about it. When Claude answers “your resting heart rate has been 6 beats higher all week”, that reply is part of the conversation. If you have left training on, the conversation can be used. Anthropic changed its consumer terms in September 2025 so that chats are used for training unless you opt out, which its privacy policy still says today. Some guides written before that change still say it does not, so check the date on anything you read.
To opt out on Claude, switch off the model-training setting in Settings, Privacy. On ChatGPT, switch off “Improve the model for everyone” in Settings, Data controls; our ChatGPT privacy settings checklist walks through it with the other four settings.
How Long Each Keeps Your Data
Both remove a deleted chat within 30 days, and both keep training data longer than that. Anthropic’s retention page says that with training on, it “may retain your data in a de-identified format for up to 5 years in our model training pipelines”. With training off, deleted chats leave its back-end systems within 30 days.
OpenAI’s privacy policy says deleted personal data is removed “within 30 days”, with exceptions for data it has already de-identified and for data it must keep for security or legal reasons. One thing has changed since the 2025 headlines: OpenAI says the court order in the New York Times case that made it keep deleted chats ended on 26 September 2025 .
Neither deletion reaches back into training. A model that has already learned from a chat is not retrained when you delete it, which is why the settings go first. The reasons are set out in full in our guide to whether it is safe to connect your training data to ChatGPT .
Memory: Claude Skips Health by Default, ChatGPT Does Not
Claude’s memory leaves health out unless you ask it to keep it; ChatGPT’s memory has no such default. Anthropic’s memory help page says: “By default, Claude does not store topics related to personal or sensitive subject matter”, and names health first in its list. You can opt in under Settings, Memory.
OpenAI says ChatGPT “may remember relevant information accessed from the app” unless the app restricts it, and its Memory FAQ says “sensitive information can appear in Memory if you share it with ChatGPT”. Memory is a separate control from training, so switching training off does not stop it.
Take Tom, training for a spring marathon with a hamstring that keeps tightening. He asks both assistants why his long runs feel harder and mentions the hamstring. With default settings, Claude answers without saving the injury to memory. ChatGPT can save “has a recurring hamstring problem” and bring it into later chats until Tom deletes it in his Memory settings.
Private Chats: Incognito vs Temporary Chat
Claude’s incognito chats and ChatGPT’s Temporary Chat both stay out of training, and Claude’s goes slightly further on memory. Anthropic says incognito chats are not used for training, are not saved to your history, do not use your existing memory and are not added to it. They are kept for 30 days by default.
OpenAI’s Temporary Chat stays out of your history, creates no new memories and is not used for training while it stays temporary. A copy may be kept for up to 30 days for safety. The difference is that a personalised Temporary Chat can still read the memories you already have.
For a one-off question about a worrying heart rate reading, either mode works. Use it before you ask, because a normal chat cannot be made private afterwards.
Age Limits and Business Plans
Claude’s consumer apps are for adults only, and ChatGPT starts at 13. Anthropic’s consumer terms say “you must be at least 18 years old to use the Services”. OpenAI’s terms set 13, or your country’s minimum age, and require a parent’s or guardian’s permission under 18. A 16-year-old racing their first HYROX can use ChatGPT with permission, but not Claude.
On business plans, neither trains on your data: Anthropic’s terms rule it out, and OpenAI does not by default. Anthropic’s commercial terms say it “may not train models on Customer Content” on Team, Enterprise and the API. OpenAI says of Business, Enterprise, Edu and the API that “we do not train our models on your data by default”, and offers a data processing addendum. A work account is still your employer’s workspace, so it is the wrong place for your own sleep and heart data.
Both companies also run health products with a no-training promise, and both are US only. Anthropic’s covers its health record and Apple Health integrations for Pro and Max subscribers in the US. ChatGPT Health is for users aged 18 and over in the US. An athlete in Leeds, Lyon or Lisbon gets neither, so the consumer rules above are the ones that apply.
Which One to Connect Your Training Data To
If you want the more private default with the least setup, connect Claude; if you already live in ChatGPT, it is fine once two settings are changed. Here is how that plays out for different athletes:
- You want to connect and forget. Claude, with training switched off. Raw connector data is already out of training and health is already out of memory.
- You want the assistant to remember your injury history. Either, but know what you are choosing. On Claude you opt health into memory yourself. On ChatGPT it happens by default, so check what it saved.
- You are under 18. ChatGPT, with a parent’s permission. Claude’s consumer apps are closed to you.
- You only have a work account. Neither for health data. Use a personal account.
- You use both. Apply the same rule to each: training off before the first question, and a private chat for anything sensitive.
The connector you use matters as much as the assistant. When you connect Kracey to Claude or to ChatGPT , the consent screen offers both assistants the same two kinds of data. Training data (sessions, 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 that stays off unless you choose it. Your conversations with the Kracey coach and what it remembers about you are never shared through either.
Who This Comparison Is Not For
This page compares two consumer privacy policies for athletes; it is not legal advice and not a medical-records guide. It is the wrong page if any of these is you:
- You are sharing medical records or anything covered by HIPAA. Neither consumer plan is built for it. Look at the business and healthcare offerings, and their own agreements.
- You need a GDPR answer for your own business. If you are a coach or a club passing athletes’ data to an assistant, the questions about consent and controllers are yours to settle with a lawyer, not with a blog post.
- You want no third party to hold your data at all. Then do not connect either assistant. Every row above describes what happens after the data has left your watch.
- You are reading this in a year. These policies change. Anthropic rewrote its consumer training terms in 2025, and both companies launched their health products in 2026. Check the linked pages before you rely on a row.
FAQ
Yes, by default on Free, Pro and Max, unless you opt out in your privacy settings. Since September 2025 Anthropic uses consumer chats for training unless you switch it off. Raw content from connectors and MCP servers is excluded, but the conversation around it, including Claude's replies, is not.
For data that arrives through a connector, yes, by default. Claude does not train on raw connector output and leaves health out of memory. ChatGPT may do both unless you change the settings. With training off on both and private chats used for sensitive questions, the gap is small.
It can. If Memory is on, ChatGPT may remember information it reads from a connected app, and OpenAI says sensitive information can appear in Memory. Review your saved memories after any chat about an injury or illness, or switch Memory off.
Not the consumer apps. Anthropic's consumer terms require you to be 18 or over. ChatGPT allows 13 and over, or your country's minimum age, with a parent's or guardian's permission under 18.
No. Both remove a deleted chat from their systems within 30 days, but neither undoes training that has already happened. Claude may keep de-identified training data for up to 5 years if you left training on, so change the setting before your first question. ---
Pick One, Then Change the Settings
Claude vs ChatGPT privacy comes down to two rows: Claude keeps raw connector data out of training and health out of memory by default, and ChatGPT does neither until you change it. Retention, age limits and business plans differ less, and both health products with a no-training promise are US only. Whichever you pick, switch training off before your first question and use a private chat for anything sensitive.
If you would rather your training was read by a coach that never trains anyone’s model on it, that is how the Kracey coach works. It is an AI HYROX, running and endurance coach in Telegram that reads your training through intervals.icu and messages you first. Your data stays in Frankfurt and is never used to train AI models, as our data page sets out. You get 14 free days, no card, and they start when your coach can see your training.