PrivacyPick
In This Article
ComparisonsAI Chat

ChatGPT vs Claude: Same Training Default, Different Trackers

Published November 14, 2026By Juan Martinez

ChatGPT and Claude both train on your chats until you switch it off. They differ in what sign-up asks for, ads, website trackers, retention and audits.

chatgpt
VS
claude-ai

ChatGPT is the AI chat app of OpenAI; Claude is the AI chat app of Anthropic. Both companies are in San Francisco, and both make the models their apps run on, so the company that runs the app also runs the model. On the questions people usually ask first, the two give the same answer: you need an account, the company stores your chats in a form it can read, and it trains its models on them until you switch that off.

The differences sit around that shared core: what you hand over to sign up, whether ads and outside trackers are involved, how long copies are kept, and how much of this anyone outside the company has checked. This comparison covers the personal plans, from Free to the top consumer tier.

ChatGPTClaude
Rating5.45.3
To sign upAn email or a linked sign-inAn email and a mobile number that gets texts
Trains on your chatsYes, until you switch it offYes, until you switch it off
After you delete a chatRemoved within 30 days, with legal exceptionsRemoved within 30 days; training copies up to 5 years
Free planUnlimited text chats; other tools cappedLimit resets every 5 hours and weekly
AdsOn Free and GoNone; Anthropic says it will stay ad-free
Website trackersFree users: chat link and title to Google Analytics, whatever the cookie choiceAccount data to 11 services if you accept cookies
Outside audit (SOC 2, ISO)Covers ChatGPT services; scope for personal plans not statedPersonal plans not covered
Security incidentsTwo in 2023No breach of accounts or chats on record
Paid plans$6 Go, $20 Plus, from $100 Pro$20 Pro ($17 yearly), $100 Max, $200 Ultra

Want to compare every field side by side? Open the full comparison table.

The part that is the same

Both apps encrypt chats in transit and on their servers, and both companies hold the keys. That protects the chats from a stolen disk, not from the company, and a valid legal demand reaches the text, because this is storage the operator can decrypt. Both companies are American, with an Irish company acting as controller for users in the EEA, the UK and Switzerland, which brings those users under GDPR.

Training on your chats works the same way on every personal plan of both. On ChatGPT the switch is “Improve the model for everyone” under Settings, Data controls; on Claude it is “Help improve our AI models” under Settings, Privacy. When Anthropic introduced training on personal accounts in August 2025, existing users saw a pop-up, and TechCrunch reported that its training switch was already turned on. Business and enterprise plans of both companies are not trained on by default, so paying for a personal plan does not change any of this.

The thumbs buttons under each answer are a trap on both. Rating an answer can send the whole conversation to the company as feedback, and that can be used for training even after you have switched training off. Anthropic says it keeps feedback for five years, separated from your user ID.

Signing up: Claude adds a phone number

ChatGPT asks for an account before the first message, tied to an email address or a linked sign-in.

Claude asks for the same account and then for a mobile phone number that receives text messages. VoIP numbers, Google Voice and landlines are refused, the step cannot be skipped, and the number cannot be changed after verification. Anthropic says it uses the number only for verification, to stop spam and abuse, and only people physically in a country Anthropic supports can sign up.

Ads and website trackers

OpenAI’s privacy policy says ads are shown to Free and Go users, and the pricing page says the $6 Go plan “may include ads”. The policy also lists advertising and web analytics partners among the companies OpenAI shares data with. In February 2026 Anthropic announced that Claude will stay ad-free, with no sponsored links beside conversations and no answers shaped by advertisers.

The website is a separate question from the app, and here a May 2026 study by researchers at IMDEA Networks and UC3M in Spain gives the detail. The study found that, for free, signed-in ChatGPT users, the website sent the link to each conversation and its automatically generated title to Google Analytics, whether cookies were accepted or rejected. The title is usually a short summary of what you asked. A separate note on the chat-title finding explains it in plain terms.

On Claude’s website, accepting non-essential cookies lets Anthropic’s servers forward your account email, plan, country and the address of the open conversation to eleven advertising and analytics services, among them the conversion tools of Facebook, LinkedIn, TikTok and Google. Rejecting those cookies stopped it. One path stays open whatever you choose: the support-chat tool Intercom receives the conversation’s address together with your email. What the researchers list for Claude are identifiers and addresses, not chat text.

For free users, no cookie choice stopped ChatGPT’s titles from reaching Google; Claude sent more about who you are, but mostly only if you clicked Accept.

How long copies are kept

OpenAI keeps chats, memories and the account until you delete them. After a deletion request it removes the data within 30 days, except where it needs to keep it longer for fraud and abuse prevention, legal obligations, payment records or the record of the deletion itself.

Anthropic publishes more numbers. A deleted chat leaves the history at once and back-end storage within 30 days. With training on, chats may be kept in a de-identified form for up to five years in training pipelines. Chats that Anthropic’s safety systems flag are kept for up to two years and can be used for safety models even with training off. Switching training off keeps old and new chats out of future training, but chats already in a training run or a finished model stay there.

Both apps have a mode for chats you do not want kept. ChatGPT’s Temporary Chat stays out of the history and out of training and is deleted within 30 days, but saving one turns it into an ordinary chat under ordinary settings. Claude’s incognito chats, opened with the ghost icon, stay out of the history, memory and training and are kept for 30 days; they work only outside projects and cannot be turned into a regular chat.

Who reads chats, and who has checked

OpenAI can read conversations for abuse review, for training where it is on, and for legal demands. Anthropic says its staff cannot open conversations by default, but designated Trust & Safety staff can when a chat is flagged under the Usage Policy, and a chat sent as feedback is readable too.

OpenAI holds a SOC 2 Type 2 report and four ISO certificates for its ChatGPT services, though the report is released on request through a portal aimed at business buyers and does not state its period. Anthropic’s Trust Center lists a SOC 2 Type II report and ISO certificates too, but the table of what they cover names the API and business products, not Free, Pro, Max or Ultra. For a personal Claude account, every privacy statement rests on Anthropic’s own documents.

The incident record points the other way. In March 2023 a bug in an open-source library showed some ChatGPT users other users’ chat titles and could have exposed payment details of 1.2% of Plus subscribers active during a nine-hour window; OpenAI took ChatGPT offline, told affected users and published a technical account within days. The same year a hacker reached an internal OpenAI messaging system; no customer data was taken, and the public learned of it about sixteen months later from press reporting. No breach of personal Claude accounts or chats is on record. Both apps and all models are closed source.

What each one can do

Both run the companies’ own models and let you switch between them. Both have web search, file uploads, voice mode, memory that is on by default, and projects (on ChatGPT from Plus). ChatGPT adds image generation, deep research reports, custom GPTs, scheduled tasks on paid plans, Codex for coding and extensions for Excel, Word, PowerPoint and Google Sheets. Claude’s plan table has no image generation; its paid plans add Research, Claude Code, Claude in Chrome, scheduled tasks and add-ins for Microsoft 365, and every plan includes artifacts, code execution and connectors to apps such as Google Drive and Slack.

Apps cover Windows, macOS, iPhone and Android on both. Claude also has a Linux app in beta; ChatGPT has no Linux app.

The bigger daily difference is the free plan. ChatGPT Free allows unlimited text chats on its base model and caps uploads, images, voice and the rest. Claude Free gives Sonnet and Haiku with search, uploads, memory and up to five projects, under a usage limit that resets every five hours and every week, with no fixed message count.

What this comparison leaves out

ChatGPT Business, Enterprise and Edu, Claude Team and Enterprise, and both developer APIs run under different terms, including no training by default. Answer quality is not compared: neither review scores how well either assistant reasons, writes or codes.

Which one to pick

ChatGPT fits if you want unlimited free text chats, image generation and the widest set of tools. Switch training off under Data controls, leave the thumbs buttons alone, and keep in mind that Free and Go carry ads and that, as the May 2026 study found, the titles of free users’ chats reached Google Analytics.

Claude fits if you want no ads, a Linux app and a website whose ad trackers the cookie banner can stop, and if a phone number tied to your account is acceptable. Switch training off under Settings, Privacy, since with it on chats can be kept for five years, use incognito chats for anything you do not want stored, and expect usage limits on the free plan.

Neither suits chats that would hurt you if a company employee or a court read them.

Missing an app?

Tell us what you could not find. Every suggestion goes onto our list of products to review.

This opens a message in your email app, addressed to hello@privacypick.com. Nothing is sent until you press send there.