How Lota's AI works

Transcripts, summaries, themes, goals, reflection prompts and the answers in Ask are all written by AI, not by a person. In the app, AI-written content and the Ask chat are clearly marked as AI.

Lota's AI works in three steps, each only when you ask for it:

1. Transcription

Turns your recording into text, and works out who said what, for example you and your therapist.

2. Understanding

Reads the transcript and writes a summary organised by topic, picks out recurring themes, and suggests goals and reflection prompts for the week ahead.

3. Ask

Answers questions about your sessions in a chat, such as “What did we say about my sister?”, using the notes on your phone that you're asking about.

The AI only sees what you choose to process or ask about. It doesn't listen in the background and doesn't look through your other sessions unless your question needs them.

• It can't give professional advice. Suggested goals and prompts are ideas to discuss with your therapist, not instructions.
• It can't judge risk or respond to a crisis. No one at Lota reads your sessions or watches your chats, so nobody would notice if you were in danger. If you're in crisis, contact your local emergency services.
• It doesn't know you beyond your notes. It only has what's in the sessions you share with it, so it can miss context your therapist would have.
• It doesn't learn from you. Your sessions are never used to train AI models, so nothing you say changes how the AI behaves for anyone else.

Like any AI, Lota can make mistakes. The most common are:

• mishearing words, especially names, quiet speech or background noise;
• mixing up who said what when people talk over each other;
• leaving out something that mattered to you, or stating it too strongly;
• an answer in Ask that sounds confident but isn't supported by your notes.

Treat what Lota writes as a helpful draft. If something important looks wrong, check it against your recording, and talk it through with your therapist rather than relying on the AI.

• In the EU. All AI processing for Lota happens on servers in the EU.
• Nothing kept. Our AI providers process your content to give you a result and don't keep it afterwards. The results are sent to your phone and deleted from our servers.
• Never trained on. Our contracts don't allow any provider to use your content to train or improve their models.
• Specialist providers. We use AI models from specialist providers rather than building our own: AssemblyAI for transcription, at its EU site in Dublin, Ireland, and Mistral AI, a French company, for summaries and Ask. Mistral runs with zero data retention for Lota, so nothing is kept once it has responded.

Our privacy policy explains the whole journey of a recording, and the privacy overview shows it step by step.

• We chose our transcription provider by testing several on the same recordings, comparing how well each got the words and told speakers apart, and listening to the moments where they disagreed.
• We test changes to the AI before releasing them in a new version of the app.
• We review the problems people report to us and use them to decide what to fix.

If Lota gets something wrong, or you have a question about how its AI works, email support@lotamind.com. Please don't include anything from your sessions unless you want to: a short description of the problem is enough.