AI Transparency and Disclosure
In one line
Where we use AI, which providers we use, how we label AI content for your audience, and what AI is never allowed to do on its own.
The short version
This short version helps you understand the full text. Read the full text for the complete terms.
- AI drafts text, images, and (later) audio for you. A person in your business signs every post before we publish it. (Full text, sections 2 and 4)
- When you talk to our onboarding assistant, we tell you it is AI. (Section 3)
- Images and video we generate carry Content Credentials (C2PA) and an AI marker in the file. We never remove or fake these marks. (Section 5)
- We switch on each platform's AI label where it applies, for example Meta's "AI info", TikTok's AI-generated label, and YouTube's altered or synthetic content setting. (Section 6)
- AI steps have no power to act alone. They cannot publish, send messages, read your passwords or tokens, or spend money. (Section 7)
- We do not train AI models on your content, and our providers may not either. (Section 8)
- AI can be wrong. Our Checks catch many problems, but you make the final call. (Section 9)
Full text
Read the full text (about 4 minutes)
1. Why this page exists
We want you, your staff, your audience, and platform reviewers to know exactly where AI is involved in qypu. This page also explains how we meet AI transparency rules, including Article 50 of the EU AI Act, platform policies, and consumer protection law on endorsements and reviews.
2. Where we use AI
| Engine and step | What AI does | Who decides | Status |
|---|---|---|---|
| SOE Studio · Brand Kit | Turns your interview answers into a Brand Kit plan you can edit | You | In place |
| SOE Studio · Drafts | Drafts captions and posts in your Voice | You sign each post | In place |
| SOE Studio · Images and video | Generates or edits images and short video | You sign each post | In place |
| SOE Studio · Audio | Generates voice-over or music | You sign each post | Planned |
| SOE Studio · Checks | An automated reviewer checks each draft against your hard facts, never-say list, licence level, and platform rules | Flags go to you | In place |
| SOE Studio · Replies | Suggests replies to comments | You sign each reply | Planned |
| AVM · Valuation | Estimates a value range with the evidence it used | A named specialist reviews it | Planned |
3. Talking to AI
Our onboarding interview is run by an AI assistant. We say so at the start and on screen throughout. You can ask for a person at any time at [CONTACT EMAIL].
4. Human sign-off
No AI output is published, sent, or relied on in qypu without a named person signing it. See the Human Review and Sign-off Policy.
5. Marks inside the file (provenance)
5.1 Media we generate carries a C2PA Content Credentials manifest stating that AI was used, and the IPTC digital source type "trainedAlgorithmicMedia" (or "compositeWithTrainedAlgorithmicMedia" for edits).
5.2 We never strip, replace, or forge provenance data from media you upload or we generate, and we never add fake camera data.
5.3 Some platforms remove file metadata on upload. That is why we also use the platform label in section 6.
6. Labels your audience sees
| Platform | What we do | Status |
|---|---|---|
| Facebook and Instagram | Set the "AI info" label on realistic AI images and video | In place where the API allows; otherwise you are prompted to set it |
| TikTok | Turn on the "AI-generated content" label for realistic AI content | In place where the API allows |
| YouTube | Declare "altered or synthetic content" for realistic generated video | In place where the API allows |
| Add a short disclosure line, for example "Image created with AI" | In place |
You may add more disclosure. You may not remove a label we set, unless the content is not realistic AI content and the platform rules allow it.
7. What AI may never do on its own
- Publish, post, send a direct message, or reply.
- Read access tokens, passwords, or payment data.
- Spend money, change billing, or change who can sign.
- Follow links or instructions found in comments, web pages, or uploaded files. We treat that text as untrusted data.
Drafting steps run with no tools at all. This is "least privilege": each step gets only what its task needs. (In place in code for drafting stages; being extended to every stage — Planned until complete.)
8. Training and data use
8.1 We do not use your content, your channel data, or your audience's data to train or fine-tune AI models.
8.2 Our contracts with AI providers prohibit them from training on data we send. Where a provider offers zero data retention for our account, we use it; the Subprocessors page shows each provider's retention.
9. Accuracy and limits
9.1 AI can produce wrong facts, wrong tone, or content that looks like someone else's work. Our Checks compare each draft with your hard facts and never-say list, scan for risky claims, and enforce platform rules.
9.2 Checks do not guarantee accuracy. You remain responsible for what you sign.
9.3 We plan to publish how often Checks catch problems in a fixed test set, and how often they miss them. (Planned)
10. What we do not offer
We do not offer fake accounts, undisclosed AI personas, fake reviews or testimonials, AI detector evasion, or posting to accounts our customers do not own or control.
11. Machine-readable version
A machine-readable summary is at /trust/ai-disclosure.json.
12. Legal notes
12.1 EU AI Act Article 50. We disclose AI interaction (Art. 50(1)), mark generated media in a machine-readable way (Art. 50(2)), and give customers the tools to disclose deep fakes (Art. 50(4)). Published text is signed by a named person who holds editorial responsibility; where text informs the public on matters of public interest, we still recommend disclosure.
12.2 Endorsements and reviews. AI must never write fake reviews, testimonials, or endorsements (US FTC Rule on Consumer Reviews and Testimonials, 16 CFR Part 465; FTC Endorsement Guides, 16 CFR Part 255; UK DMCC Act 2024; EU Unfair Commercial Practices Directive). Paid partnerships must carry the platform's paid-partnership label.
12.3 Bots. Automated replies that try to sell to people in California must say they are automated (California Business and Professions Code §17941). Replies in qypu are signed by a person before sending.
Change log
- 2026-10-10 · 0.1.0 · First full draft, replacing the earlier short page at /legal/ai-transparency.
Open questions for counsel
We publish these while the page is a draft, so you can see what is not settled yet.
- Confirm qypu's role under the EU AI Act (provider of a general-purpose-model-based system that generates synthetic content, and therefore subject to Art. 50(2)), and the status of any transition period for Art. 50(2) proposed in the EU "Digital Omnibus".
- Watch the Commission's Code of Practice on marking and labelling AI-generated content and align section 5 when it is final.
- Confirm the "editorial responsibility" exception in Art. 50(4) covers signed posts, and whether a disclosure should still be the default.
- Platform label status must be verified per API before publication; mark any platform where labels cannot be set via API as "you set it manually".
Useful for: Business owners, Your staff, IT and security reviewers, Platform reviewers. To save this page as a PDF, use your browser's Print command. Back to the trust centre.