You’ve been using ChatGPT or Claude for months, maybe years. You’ve entrusted them with your project ideas, professional doubts, code snippets, and personal documents. But do you know exactly what these conversations reveal about you? The Swiss company Proton, known for its VPN and encrypted email, decided to make this question tangible with a free tool: AI Paper Trail. Here’s what it does, why it matters, and how to do the same on your end.
What is AI Paper Trail?
Launched by Proton in summer 2026, AI Paper Trail is a free online service (lumo.proton.me/aitrail) that analyzes the history of your conversations exported from ChatGPT or Claude, then produces a readable report: what your exchanges reveal about your profession, your relationships, your habits, and your interests. The idea isn’t new in spirit — it’s a “mirror” of your own data — but the approach is novel in its simplicity: you drop in a file, and the tool hands you a diagnosis.
Proton stresses a privacy point that doubles as its pitch: the imported data is analyzed without being stored on its Lumo assistant’s servers, and the generated report is visible only to the person who requested it. This is consistent with the company’s positioning, which built its Lumo assistant on open models running on European servers with end-to-end encryption.
What the report really contains
AI Paper Trail doesn’t just list your messages. It structures the information into several parts:
- A privacy-profile label — a tag that reflects your own approach to the topic (cautious, exposed, etc.).
- An AI exposure score — calculated from the number of extracted data points and their nature (address, habits, relationships, etc.).
- The detail of identified information — what leaked from your private life into your prompts.
- A financial estimate — the value, in dollars, that this data represents for an AI provider.
- A shareable summary card, without exposing the personal data it summarizes.
The figure that often shocks is the monetary estimate. Even if only indicative, it recalls a reality: your conversations are a raw material. Every data point has a value on the training and advertising market.
What ChatGPT and Claude know about you (by default)
According to Proton, consumer platforms keep far more than your sentences. In short, they collect:
- the conversations and responses exchanged;
- the imported files (PDF, images, spreadsheets);
- usage data (frequency, features, session length);
- technical information (IP address, browser, operating system, approximate location);
- and your account information.
This data can be retained for a long time, sometimes indefinitely depending on your settings, and used for model training or commercial purposes — not always with your knowledge. This is where the subject becomes serious: it’s not just “the AI remembers,” it’s “the AI builds a profile.”
On the ChatGPT Memory side, the system continuously summarizes the context of your past chats to personalize your responses. You can view and delete these memories in Settings > Personalization > Memory. Claude, for its part, offers equivalent data controls. The habit to adopt: look at what your tools “retain” about you, and erase what bothers you.
Practical guide: export your data yourself
The most powerful thing isn’t Proton’s tool — it’s taking back control of your data. Fortunately, both ChatGPT and Claude include native export.
Exporting your ChatGPT history
- Log in at chatgpt.com.
- Open the profile menu, then Settings.
- Go to Data controls.
- Under Export data, click Export, then confirm.
- You receive a link to a ZIP file containing
conversations.json.
OpenAI details the official procedure on this help page. Tip: you can also go through OpenAI’s Privacy Portal to exercise your rights.
Exporting your Claude history
- In the web app or Claude Desktop, open Settings > Privacy (or Data & Privacy).
- Click Export data.
- A download link is sent to you by email (valid for a limited time).
- The ZIP contains your conversations in JSON format.
The steps are documented by Anthropic on this Claude help page. Once you have the file, just drop it into AI Paper Trail to get the report.
The legal framework: GDPR is on your side
If you’re in Europe, you’re not defenseless. The General Data Protection Regulation (GDPR) gives you a right of access, rectification, erasure, and objection regarding your personal data. The native export offered by the publishers precisely answers this portability right. Concretely:
- you can request a copy of everything concerning you;
- you can refuse the use for training (an option often found in the privacy settings);
- you can delete your account and its associated data.
The problem is that these settings are scattered and disabled by default. AI Paper Trail makes the issue visible, but it’s up to you to act afterward.
Privacy-respecting alternatives
If this realization bothers you, know that there are more sober paths. Proton’s Lumo assistant advertises a “zero collection” policy, with conversations not used for training. Other approaches exist:
- local models (e.g., via Ollama) that run on your machine, without sending anything to the cloud;
- European providers subject to the GDPR rather than the US regime;
- and, of course, good hygiene: never paste a secret, password, or sensitive data into a public chat.
5 habits to limit your exposure
- Audit once with AI Paper Trail — you’ll be surprised by what comes out.
- Disable training in each platform’s settings.
- Clean the memory of ChatGPT and Claude’s controls regularly.
- Anonymize: replace real names and numbers with placeholders before pasting a document.
- Export and delete the accounts you no longer use.
Conclusion: you are the one who controls transparency
AI Paper Trail is a clever initiative: it turns an abstraction — “my data is being collected” — into a concrete, quantified document. But the tool doesn’t do the work for you. The real lesson is that privacy in AI is managed: export, audit, settings, deletion. The industry giants have every interest in keeping you at default; your interest is in leaving the default behind.
As AI sneaks into all our tools, taking back control of your conversations isn’t paranoia — it’s basic digital hygiene. Go build your own “paper trail.” You might never write a prompt the same way again.