How Marlo Uses AI
Effective Date: July 27, 2026 Last Updated: July 27, 2026
Marlo is built on AI, and you deserve to know precisely what that means for the most private thing you own: your own voice.
What the AI actually does
Transcription. Your recording is sent to OpenAI's Whisper model, which converts speech to text. Nothing about your entry is interpreted at this stage — it is a transcription.
Cleanup. The raw transcript goes to Anthropic's Claude, which removes filler words and repetitions and adds punctuation, so what you said reads like something you wrote. The meaning is not changed.
Recaps and insights. When you ask for one, your recent entries are sent to Claude to produce a summary, spot recurring themes, or surface a pattern you might not have noticed.
The companion. When you talk to the companion, your message and relevant context from your journal are sent to Claude, which responds.
What the AI does not do
- It does not train on your content. Our agreements with OpenAI and Anthropic prohibit using data submitted through their APIs to train their models. We do not train models on your content either.
- It does not run continuously in the background. Your entries are processed when you record, and when you explicitly ask for a recap or a reply.
- It does not share your content with other users. Ever.
- It does not make decisions about you that have legal or similarly significant effects.
What AI gets wrong
You should hold everything the AI produces at arm's length:
- Transcription errors. Accents, background noise, names, and technical terms get mangled. Always assume a transcript may contain mistakes.
- Confident wrongness. Language models state incorrect things fluently and without hedging. A summary can misrepresent what you meant.
- Invented detail. Models sometimes add specifics that were never in your entry.
- Shallow pattern-matching. An "insight" is a statistical observation about your words, not an understanding of your life.
You can edit any transcript or entry. Where the AI got it wrong, correct it — the correction is what gets stored.
The companion is not a therapist
The companion is a conversational feature for reflection. It is not therapy, counseling, medical care, diagnosis, or crisis support, and it is not a substitute for a professional who is accountable to you.
It has no clinical training, cannot assess risk, does not know your history beyond what is in your journal, and can be confidently wrong about something that matters. Do not use it to make decisions about your health, safety, medication, or relationships.
If you are in crisis: in the US, call or text 988 (Suicide & Crisis Lifeline) or call 911. In the UK, call 116 123 (Samaritans) or 999. Elsewhere, contact your local emergency number. Please talk to a person.
Your controls
- Edit any transcript or entry.
- Delete any recording or entry — it is removed from our systems.
- Export everything from Profile → Export my data.
- Delete your account from Profile → Delete account, which removes your content.
- Skip the AI features. Recording and keeping entries works without ever asking for a recap or opening the companion.
Who processes what
| Provider | What it receives | Purpose |
|---|---|---|
| OpenAI | Your audio recording | Speech-to-text transcription |
| Anthropic | Your transcript and relevant entry context | Cleanup, recaps, insights, companion replies |
Both act as data processors under contract, may not train on your content, and are listed with the rest of our vendors in Subprocessors.
Changes
If we change AI providers, or if a provider changes its terms in a way that affects how your content is handled, we will update this page and notify you before the change takes effect.
Questions: privacy@marlojournal.com