AI Disclosure

Effective date: May 11, 2026

Last updated: September 3, 2026

This page lists every place artificial intelligence is used in the TellSophie product, what data each AI service receives, and what we do not do with AI.

We use AI in a few specific places to help shape what we build for your child — and in one place to keep the archive safe. Here is every one of them. We do not use AI for anything we don't list here.

The marketing website uses no AI

There is no AI on tellsophie.com. The signup form is a form. The Frequently Asked Questions are written by a person. There is no chatbot. There is no AI-generated content on the site.

A large language model is used in four places

Provider: A commercial large-language-model provider, working on our instructions under a written data processing agreement. The provider is identified by name in the subprocessor list on the Trust page. Model: A single large-language model selected at launch, from that one provider. Geographic restriction: We contract for United States data residency where the provider offers it as a tier or option. Retention: Per the provider's data processing agreement. We negotiate Zero Data Retention status where available.

Material changes to this list will be reflected here and emailed to active users.

1. Weekly questions

What it does: Reads past recordings to write the next question to ask. What it receives: Your child's life-stage information (age, life phase), the optional family-context free-text you write during onboarding, and the transcripts of earlier recordings. What it returns: The question to ask next.

2. Emotional tone

What it does: Gently summarizes how each recording feels. What it receives: The transcript of the recording. What it returns: A short summary of the feeling of that recording. The summary colors the Living Timeline, and it is read by the highlight-reel curation and the Annual Letter below. Opt-out: A switch in settings turns the analysis off entirely.

3. Highlight-reel curation

What it does: Chooses the most meaningful moments for the year-end reel. What it receives: The transcripts of the year's recordings and their emotional-tone summaries. What it returns: Which moments belong in the reel.

4. Annual Letter generation

What it does: Writes a letter to your child from the year's recordings. What it receives: The transcripts of the year's recordings and their emotional-tone summaries. What it returns: A draft of the Annual Letter.

Three other services process what you record

5. Transcription of your recordings

Provider: A speech-to-text service, working on our instructions under a written data processing agreement. The provider is identified by name in the subprocessor list on the Trust page. Model: A speech-to-text model, with a second model from the same provider as the fallback. What it receives: Your audio. The audio is sent for transcription only. It is not used for training. What it returns: A transcript of your audio. The transcript is stored with the recording, and it is what the emotional-tone summary, the highlight reel, and the Annual Letter read. Geographic restriction: We contract for United States data residency where the provider offers it as a tier or option. Retention: Audio sent for transcription is processed and discarded in line with the provider's commercial terms. We do not authorize retention beyond what is needed to return the transcript.

6. Content safety screening

Provider: An automated content-safety screening service, working on our instructions under a written data processing agreement. The provider is identified by name in the subprocessor list on the Trust page. It screens every recording, and it is the only content-safety provider in the path. What it receives: Video is scanned frame by frame. Audio has no picture to scan, so its text is screened instead. This runs on every recording before it is stored. Why: The scan enforces our content policy and detects content that we are required to report under federal law. Geographic restriction: We contract for United States data residency where the provider offers it as a tier or option. Retention: Per the provider's data processing agreement, with Zero Data Retention status where available.

7. Child-safety matching

Provider: A child-safety hash-matching service, working on our instructions under a written data processing agreement. The provider is identified by name in the subprocessor list on the Trust page. What it receives: Sampled video frames are converted into one-way fingerprints, and only those fingerprints are sent, so the picture itself never leaves our storage. Audio has no frames and is not sent at all. Why: The fingerprints are matched against the known-abuse database. Federal law requires this. If there is a match: We file the report federal law requires with the National Center for Missing and Exploited Children. NCMEC is a statutory recipient of that report, not a vendor we chose. Geographic restriction: We contract for United States data residency where the provider offers it as a tier or option. Retention: Per the provider's data processing agreement, with Zero Data Retention status where available.

What we do not do with AI

• We do not train on your content. Consumer-health data — never sold, never used to train AI. No recording, transcript, or anything about your child is ever used to train an AI model. We do not sell your child's data. We do not show ads. We have explicit contractual commitments that our recordings, our transcripts, and our prompts are not used to train any of the providers' models.
• We do not generate synthetic recordings on your behalf. The product does not produce AI-generated voice or video content addressed to your child. Everything in your child's collection is recorded by a real person. The Annual Letter described above is written text drafted from those recordings; no synthetic voice or video is ever produced.
• We do not clone your voice. No voice fingerprinting, no synthetic voice generation, no voice-cloning model in the application.
• We do not run facial recognition. No face matching. No facial geometry scans for any verification flow.
• We do not run biometric processing. Per our intentional design, the application is not subject to the Illinois Biometric Information Privacy Act, the Texas Capture or Use of Biometric Identifier Act, or the parallel laws in other states.

Vendor list and updates

The full subprocessor list is published on the Trust page. The data processing agreement status and Zero Data Retention status for each vendor will appear on this page when the iOS application launches.

If we add an AI service, change one, or change what data goes to one, we update this page and email all active users.

Contact

privacy@tellsophie.com for questions about specific AI use.

Related

• Trust commitments — storage, export, continuity reserve, and subprocessors.
• Privacy Policy — what the marketing site collects and how to reach us.