mysticlab
Dream Psychology

Does AI Store My Dreams?

When an AI reads your dream, the text can end up in three places. What OpenAI, Google and Anthropic keep, 12 questions to ask, and 5 apps compared.

Updated Sep 29, 202611 min read
A dreamer reflecting on an interpretation

When you send a dream to an AI dream app, copies of it can exist in more than one place, for different lengths of time. This guide follows a dream to each of them and gives you twelve questions to check any app.

The short answer

Usually yes, for a while, and in up to three places. A journal app keeps your dream in its own database for as long as its policy says. The AI provider behind it may keep a copy for a limited time: up to 30 days at OpenAI and Anthropic, and 55 days on Google's Gemini API, with exceptions such as legal requirements. A dream ends up inside an AI model only if someone trains on it; OpenAI's and Anthropic's APIs and Google's paid Gemini API don't by default, but an app's own policy may allow it.

Where can your dream end up?

In three places, each with its own rules. (For what the model does with a dream while it reads it, see how AI analyzes dreams.)

  • 1. The app's database: your journal. This copy is what lets you search old entries and see patterns. The app's privacy policy decides where it is stored, who can read it and how long it stays. Is a dream journal private? covers this layer.
  • 2. The AI provider's logs. An app that uses a cloud AI model sends your dream to the provider through an API, the business connection that lets one company's software send requests to another's. The provider may keep the request and the reply for a limited time, mainly to detect abuse. The next section has the details.
  • 3. Inside a model. This happens only if the provider or the app trains a model on your text. Calling the data "anonymized" doesn't settle the matter: EU data protection authorities say a model trained on personal data counts as anonymous only if it is very unlikely that people could be identified from it, or their data extracted from it through queries, and that this has to be judged case by case (EDPB, 2024).

Do AI providers keep what apps send them?

Yes, for a limited time in most cases, and the rules differ by provider and by type of access. Here is what three major providers say about their APIs, checked on 29 September 2026:

  • OpenAI API. Since 1 March 2023, data sent through the API has not been used to train OpenAI's models unless the customer opts in. Abuse-monitoring logs, which can include prompts and responses, are kept for up to 30 days, unless longer retention is required by law or reasonably necessary to protect OpenAI's services or others. Approved customers can get Zero Data Retention, which keeps their content out of those logs (OpenAI). For the Chat Completions endpoint, OpenAI lists no stored application state by default, with narrow exceptions such as encrypted prompt-cache data kept for up to 24 hours (OpenAI). The same endpoint has an optional store setting that saves outputs for OpenAI's model-distillation and evaluation products (OpenAI API reference).
  • Google Gemini API, unpaid tier (for example the free quota on the API, or Google AI Studio without a billing account). Google uses what is submitted, and the responses, to provide, improve and develop its products and machine-learning technology. Human reviewers may read and annotate it, and Google says not to submit sensitive, confidential or personal information (Gemini API terms).
  • Google Gemini API, paid tier. Prompts and responses are not used to improve Google's products. They are logged for a limited time to detect abuse and for legally required disclosures (Gemini API terms), a period Google's abuse-monitoring page puts at 55 days (Google). Apps offered to users in the EEA, Switzerland or the UK may use only the paid services (Gemini API terms).
  • Anthropic API. By default, Anthropic does not use inputs or outputs from its commercial products, including the API, to train its models (Anthropic). It deletes them within 30 days, except for features with longer retention under the customer's control (such as its Files API), separate agreements such as zero data retention, enforcement of its usage policy, and legal requirements (Anthropic).

In the EU, a provider that works as an app's processor, meaning it handles data on the app's behalf, may process that data only on the app's documented instructions (GDPR, Art. 28(3)(a)).

Are chatbot apps different?

Yes. When you paste a dream into a consumer chatbot yourself, its own settings apply, not the API rules above. Anthropic, for example, says chats from its consumer Claude apps can be used for model training if you allow it in your settings, if a conversation is flagged for a safety review (to improve how it enforces its usage policy), or if you explicitly opt in, for example through a testing program; incognito chats are not used, even with the setting on (Anthropic). Check the data settings of whichever chatbot you use before you paste a dream in.

How can you check an AI dream app?

Ask where the dream goes at each step, and write down only what the app and its provider actually state. For each of the twelve questions below, record Stated, Not stated or Contradicted (one of its sources says the opposite of another), with the link and the date you checked. "Not stated" means you did not find it in the app's own pages; it does not mean the app doesn't do it.

  • 1. Which AI providers are named? Where to look: the privacy policy and its list of processors (also called subprocessors).
  • 2. Is it API access, a consumer product or a model on your device? Where to look: the policy, or a technical or method page.
  • 3. Does that provider train on this kind of access by default? Where to look: the provider's own documentation (see above).
  • 4. Has the app opted in to sharing data with the provider, or to training? Where to look: the app's policy; only the app can tell you.
  • 5. How long does the provider keep data, and with what exceptions (abuse logs, legal holds, zero data retention)? Where to look: the provider's documentation.
  • 6. Where does the app store your journal, in which country, and for how long? Where to look: the app's policy.
  • 7. Does the app train its own models on dream content? Where to look: the app's policy.
  • 8. Does it sell, license or share datasets, including "anonymized" or "aggregated" ones? Where to look: the app's policy.
  • 9. Does it ask for separate, explicit consent before AI processing, and can you withdraw it? Where to look: the app itself, and its policy.
  • 10. Which identifiers go with the dream (email, account ID, profile details)? Where to look: the policy or security page.
  • 11. Does your data leave your region, and on what legal safeguard? Where to look: the policy.
  • 12. What can you delete: one entry, the whole account, and the provider's copy? Where to look: the policy and the app's settings. Both app stores require in-app account deletion from apps that let you create an account (Apple; Google Play).

How do five dream apps compare?

As their own pages described them on 29 September 2026, all five name an AI provider, but they differ on where data is kept and on what their policies allow for training, sale or licensing. MysticLab publishes this guide and is one of the five, so every row, ours included, repeats only what the app's own pages say (their claims); nothing here was tested independently.

App (checked 29 Sep 2026) AI provider · data location Training, sale or licensing
MysticLab OpenAI, only after in-app consent · EU (Hetzner, Helsinki, Finland) (privacy and security) Not used to train AI models; no sale of personal data (privacy policy)
Dreamly OpenAI API (method page) · Supabase servers in the United States (policy) Policy allows sharing or selling anonymized, aggregated dream data to third parties (policy)
Oniri OpenAI, plus Replicate for images · not stated; processors may be outside your country (policy) Does not sell personal information; anonymous data may be used for statistics or research (policy)
DreamApp OpenAI · UK company; transfers outside the UK possible (privacy notice) May use submitted information, including dream content, to develop and train models; may license or sell de-identified dream datasets, only with explicit opt-in (privacy notice)
DreamStream "Google providers" · may be processed in the United States or other regions (policy) Does not sell personal data or "broker" dream content (policy)

Two notes on reading it:

  • "Not stated" is not "no". Oniri's policy names its processors but not where data is stored. DreamStream's policy names "Google providers" but not which service or tier; on Google's Gemini API, the paid or unpaid tier decides whether Google may use prompts to improve its products (Gemini API terms), so question 3 is not stated for DreamStream.
  • Our own limits on the same twelve questions are in the next section. For features, prices and ratings too, see the dated comparison of dream apps.

How does MysticLab handle your dreams?

Here are the same twelve questions answered for MysticLab, where Luna interprets your dreams, limits included; the privacy and security page has the full detail.

  • AI provider (1–5): Luna's replies come from OpenAI's API, and only after you agree on a consent screen. OpenAI doesn't train on API data unless a customer opts in, and we haven't. It may keep API data for up to 30 days to detect abuse, unless the law requires longer.
  • Storage and our own use (6–8): your journal is on our server in Helsinki, Finland (EU), encrypted in transit but not end-to-end, so our server can read your dreams, as it must to pass them to Luna. We don't use your dreams to train AI models and don't sell personal data. We use no embeddings or retrieval (RAG) database, the techniques explained in AI dream analysis technology; search is plain full-text search.
  • Consent, identifiers and transfers (9–11): withdraw consent in Settings and Luna stops reading new dreams. Your email address and account ID are never sent to OpenAI; your display name, age, gender and zodiac sign are, if you added them to your profile. OpenAI is in the United States, so dreams Luna reads leave the EU.
  • Deletion and export (12): deleting your account in the app erases everything in it immediately. A single dream you delete leaves your journal, and its stored copy is erased when you delete your account. There is no self-service export yet; email us for a copy.

The free web analyzer keeps less: text you submit there goes to our server and to OpenAI to write one short reading, and is not linked to an account or stored by us. To enforce the one-reading-a-day limit, we keep a one-way code derived from your network address, and a counter, for up to 48 hours; OpenAI's retention above applies to that text too.

Frequently Asked Questions

Do AI dream apps train on my dreams?

It depends on the app, not just its AI provider. OpenAI's and Anthropic's APIs and Google's paid Gemini API don't train on API content by default, but an app can still train its own models if its policy allows it: DreamApp's privacy notice, for example, says it may use dream content to develop and train models. Look for a training or AI section in the app's policy.

How long does OpenAI keep dreams sent through its API?

Up to 30 days in abuse-monitoring logs, unless longer retention is required by law or reasonably necessary to protect OpenAI's services or others. API data isn't used to train OpenAI's models unless the customer opts in, and approved customers can get Zero Data Retention, which keeps their content out of those logs.

Is it harmless if an app trains on "anonymized" dreams?

Not automatically. EU data protection authorities say a model trained on personal data counts as anonymous only if identifying people, or extracting their data through queries, is very unlikely, judged case by case. Treat "anonymized" or "aggregated" in a privacy policy as a claim to check, not a guarantee.

Can a person read the dreams I send to an AI app?

Sometimes. Google says human reviewers may read and annotate content sent to its unpaid Gemini API tier. Other providers' pages may not say either way, so don't assume nobody ever reads it, and remember that an app without end-to-end encryption can read your dreams on its own servers (ICO).

Does deleting a dream in the app delete it at the AI provider?

Not necessarily. The app's copy follows the app's deletion rules, while the provider's copy follows the provider's retention rules, such as OpenAI's abuse logs of up to 30 days, or longer where the law requires. Google Play tells developers that rely on service providers to ask them to delete a deleted account's data too (Google Play).

References

Web pages were checked on 29 September 2026. Provider and app policies change, so check the current version before relying on one.

Analyze your own dream

Describe it and get a short reading from Luna — free, no account needed.

Try the Free Dream Analyzer
Download on theApp StoreGet it onGoogle Play

Related Guides

Is a Dream Journal Private?

Who can read your dream journal depends on where it lives. Compare paper, on-device, cloud and AI journals, then check any app with 10 questions.

How AI Analyzes Dreams: Complete Guide

How AI dream analysis works: the four lenses (Freud, Jung, traditional, modern), what a language model actually does, and an honest view of its limits.

AI Dream Analysis Technology: NLP & LLMs Explained

An industry overview of LLMs, embeddings and retrieval (RAG) in dream apps, and which of these MysticLab uses: only a general-purpose LLM.