AnonymFilesSummarize an email thread or meeting notes with AI without leaking names
To summarize an email thread or meeting notes with AI without leaking names, do not paste the raw text into the chat: save the thread as an .eml or .msg file (or paste the messages into a .txt file), pseudonymize it on your computer, check signatures, first names and initials, then give the copy to the AI with instructions that keep the tags. The real names come back into the summary on your machine, never at the AI provider's end.
Why is an email thread so risky to paste into an AI?
Because it holds far more than what you read. A ten-message exchange carries the From, To, Cc and Sent lines ten times over, as many signatures with name, title, firm and direct line, confidentiality notices naming the sender, and email addresses built on people's names. Someone mentioned once in the first message reappears further down in three forms: “Mr Martin”, “Jean”, “JM”.
And the gesture is quick: select all, copy, paste into the chat. The whole thread then goes to the AI provider, which receives, processes and keeps it for a while. The CNB's April 2026 guide recalls that any prompt sent to a generative AI transmits data to the company running it (p. 14). The CNIL, the French data protection authority, advises “never sharing confidential information such as personal data” with a consumer service (quotations from French texts are our translations).
When to use it: emails, meeting notes, pasted chats?
| Source | How to prepare it | Watch out for |
|---|---|---|
| An Outlook thread | Save the latest message in the thread as .msg or .eml | The latest message usually holds the quoted history; a message missing from the thread will be missing from the summary |
| The emails of a matter | Save each useful message and drop them in together | The same people keep the same tags from one message to the next |
| Typed meeting notes | Word or .txt, dropped in as is | Initials and first names in quick notes need checking at review |
| Messaging, chat, video calls | Paste the exchange into a .txt file (UTF-8) and drop it in | Usernames and nicknames are not always recognized as names |
| Automatic transcript of a meeting | Export the text, then treat it as notes | A tool transcribing “live” in the cloud receives everything before any pseudonymization (CNB guide, p. 17) |
To save a message: in classic Outlook, File, then Save As; in new Outlook, More actions, then Save as; in Outlook on the web, More actions, then Download, which gives an .eml or .msg file (Microsoft support). The step-by-step guide, Apple Mail included, is in anonymize an email (.msg, .eml).
What should be masked in an email thread or notes, and why?
| Item | Why | In AnonymFiles |
|---|---|---|
| Sender and recipients (From, To, Cc, Bcc) | They name every participant, the client included | Kept in the copy and redacted like the body |
| Email addresses | They often contain the name and the company's domain | Detected by rules, replaced in full |
| Subject line | “Martin v. SCI Les Tilleuls: hearing on the 14th” says it all | Kept and redacted; a matter reference is masked with one click |
| Signatures and confidentiality notices | Name, title, firm, address, direct line | Names, addresses and phone numbers detected; a unique job title needs checking |
| Messages quoted below the reply | Where most forgotten names hide | Part of the body, run through the same detection |
| First names alone, initials, nicknames | They easily slip past any detection | A first name alone can be linked to the full name when unambiguous; “JM” or a nickname is masked with one click |
| Amounts, project names, meeting places | They place the matter without naming anyone | By hand, one click, for the whole folder: amounts are not detected automatically |
| Attachments | Contracts, bank details, ID documents | Never copied into the message copy, which only states how many there are; the useful ones are processed as documents in the folder |
| Technical headers (Received, servers, message ID) | They reveal domains and the message's route | Never kept in the copy |
Keep what gives the thread its meaning: the order of messages, who answers whom (as tags), requests, commitments, deadlines. A test before sending: could a colleague outside the matter tell who the copy is about? If so, a clue remains, often in a signature or a sign-off (“Best, Claire”).
How to summarize an email thread with AI, step by step
- Save the thread or the notes as a file. The most recent message in the thread as .msg or .eml, or the exchange pasted into a .txt file. Keep these files in the matter's folder, on your computer.
- Drop them into AnonymFiles, with other useful documents from the matter if needed. For each email, a text copy keeps the subject, sender, recipients, date and body, quoted thread included; an HTML message is converted to text without loading images or following links.
- Let detection do its work. Names, organizations, addresses, emails, phone numbers, IBANs, private dates and IDs become stable tags: the same person keeps the same tag in the headers, the body, the quoted thread and the other documents.
- Review side by side. Look for initials, nicknames, first names in sign-offs, project names and amounts; one click masks them across the whole folder. Rename key tags if useful, for example [PERSONNE_CLIENTE] (client).
- Validate, export, paste the instructions. “Copier la consigne IA” (copy the AI instruction) copies the instructions to paste at the top of the conversation; the mapping table stays on the computer.
- Ask for the summary with one of the prompts below.
- Reinject the real names. Paste the answer into “Réinjecter un résultat” (re-inject a result): the preview shows the summary with the real names, which you save as text, Word or PDF.
For a few lines pasted from a chat, the same route applies: a .txt file is enough, and the manual method remains possible (see pseudonymize a document for AI).
Which prompts summarize an email thread or a meeting?
Paste them after the instructions. The tags are examples.
1. Thread summary
Summarize this email thread in ten lines at most: what the exchange is about, each participant's position, what was decided and what remains open. Refer to people by their tags, without trying to work out who they are. For each point, give the message it comes from (sender and date or date tag).
2. Who does what
Build a table of the commitments and requests in this thread: who (tag), what, by when, source message. Flag requests left unanswered and conflicting deadlines.
3. Timeline of the exchange
Rebuild the thread's timeline, oldest to newest: date, sender, recipients, what changes in the discussion. Flag quoted messages that seem to be missing.
4. Meeting minutes
From these notes, write structured minutes: participants (tags), topics covered, decisions, actions with owner and due date, open questions. Add nothing that is not in the notes; write “not specified” if a due date is missing.
5. Draft reply
Draft a reply to the latest message from [PERSONNE_2], on behalf of [PERSONNE_1], covering the open points in the commitments table. Professional tone, ten lines at most, tags kept.
What does the AI give you, and what must you check?
A readable summary, an action table, minutes or a draft reply, with tags in place of names. Review it before use: the CNB guide asks lawyers to “reread, check and validate every generated content before using it or sending it to a client” (p. 28).
- Attribution. AI easily confuses who said what in a quoted thread: check each position in the source message.
- Decisions and deadlines. “Agreement reached” or “by Friday” are checked word for word: summaries tend to round off.
- Missing messages. The AI only summarizes what it received; a message absent from the quoted thread does not exist for it.
- Tags. After reinjection, each name is in its place: a tag given to the wrong person changes a sentence's meaning.
- Draft reply. It goes back to a real recipient: review it as an email you sign.
Which rules apply to emails and notes?
- French lawyers. Article 66-5 of the Act of December 31, 1971 covers “correspondence exchanged between the client and their lawyer, between the lawyer and fellow lawyers”, except letters marked “officielle”, as well as “interview notes”. The national bar rules add that all exchanges between lawyers, whatever the medium, electronic included, are “confidential by nature” (RIN, Art. 3.1). A thread with a client or opposing counsel is therefore pseudonymized before any AI.
- Live minutes. The CNB guide is explicit: pseudonymization “cannot apply to producing, live, the minutes of a meeting with a client”, since the information exchanged goes to the AI that writes them (p. 17). Take your notes, then pseudonymize them before asking for minutes.
- In-house counsel and DPOs. Internal emails name employees, customers and suppliers: the GDPR requires data to be “limited to what is necessary” (Art. 5), and commercial exchanges may be trade secrets (French Commercial Code, Art. L. 151-1). See in-house counsel and DPOs and ChatGPT and GDPR.
What AnonymFiles does with an email, and what it does not
AnonymFiles, French document anonymization software, is a desktop app coming soon for Apple silicon Macs and Windows. From an Outlook .msg or an .eml file, it makes a text copy: subject, sender, recipients, date and body, redacted like the rest of the folder; technical headers and images are not kept. It also reads UTF-8 .txt and .md files, which covers text pasted from a messaging or video-call tool. The same person keeps the same tag in every message and document, each copy is reviewed side by side before export, and the real names are reinjected on the computer.
What it does not do: it does not copy attachments into the message copy (only their number is stated), it does not plug into your email client or browser, it calls no AI, and it does not rebuild an .eml on export. Each email is limited to 50 MB, attachments included; an unreadable or nearly empty email is excluded and flagged. Interface in French and English, no account, no telemetry. Pricing: €290 excl. VAT per seat per year, €240 from 3 seats, 14-day trial (pricing). Other situations: use cases and how it works.
Frequently asked questions
Can you ask ChatGPT to summarize a work email thread?
Yes, on a pseudonymized copy. Pasted as is, the thread hands the AI provider the names, email addresses, signatures and quoted messages of every participant. Save it as .msg or .eml, pseudonymize it on your computer, review the copy, then give it to the AI.
How do you save an Outlook thread to pseudonymize it?
In classic Outlook, open the message, then File, Save As. In new Outlook, More actions, then Save as. In Outlook on the web, More actions, then Download. Preferably save the most recent message, which usually contains the quoted history.
What about text copied from a messaging app or a video call?
Paste it into a UTF-8 .txt file and drop it in as a document: it goes through the same detection and review as an email. Pay particular attention to usernames and nicknames, which detection does not always recognize as names.
Can an AI take live minutes of a client meeting?
The French National Bar Council's guide says pseudonymization cannot apply to minutes produced live: the confidential information exchanged goes to the AI writing them. Take the notes yourself, pseudonymize them, then ask the AI for the minutes.
What happens to an email's attachments?
In AnonymFiles, they are never copied into the message copy, which only states how many there are. If the AI needs them, process them as documents in the folder, reviewed and validated like the others; otherwise they do not go.
Are first names and initials detected?
A first name alone can be linked to the same person's full name when the link is unambiguous, and the initial of a name (“J. Dupont”) is masked along with it. Standalone initials such as “JM” or a nickname are masked with one click at review, across the whole folder.