AI & Content Policy

1. Which AI models we use

• Claude (Anthropic) – in every workspace: writing posts and ideas, rewriting, summaries, deriving style rules, the website chat.
• OpenAI (optional): turning voice notes into text (gpt-4o-mini-transcribe), image generation, alternative idea suggestions (ChatGPT).
• Google Gemini (optional): transcribing voice notes when OpenAI is not used, image generation, alternative idea suggestions.

2. What is sent to the AI

Only what a request needs: your brand description and rules, the channel, the topic, excerpts of your earlier posts, your voice note and – if you connected them – short, pre-filtered excerpts from your calendar, e-mails and CRM. Where possible, names, e-mail addresses, phone numbers, amounts and contract details are masked before anything is sent. Passwords, tokens and payment data are never sent.

Providers and locations: Sub-processors.

3. No training on your data

Under their API terms, Anthropic, OpenAI and Google do not train their models on data received through their paid APIs. We do not train any general model with your data either, and we never pass it on to other customers.

4. Learning your voice and style

Sayweek learns from your own posts, the brand rules you fill in and the way you rewrite drafts: this produces short, readable style rules (e.g. “shorter sentences”, “put the number in the first line”) that later drafts follow. They live only in your workspace, and you can view and delete them at any time in the brand settings. No other customer sees them and they are not used to train models.

5. Human approval – you are the publisher

Sayweek publishes nothing without your approval: every post arrives as a draft and only goes out when you (or an authorised member of your team) approve it. Mandatory human approval is on by default and can be locked; two-step approval can also be switched on. Every post has a provenance log (what the AI wrote, how much a person changed, who approved it and when – without storing prompts), and a compliance report can be downloaded from it.

You, the customer, are responsible as the publisher for the published content – its accuracy, lawfulness, the necessary permissions and labels. Sayweek is a tool that helps you write and schedule your posts; see the Terms.

6. AI can be wrong

Language models write convincingly but can be wrong: they may invent facts, figures, quotes, sources or events (“hallucination”) and misread the context. Where Sayweek spots a claim worth checking, it lists it next to the draft; even so, check every fact, name, number and promise before publishing, especially on legal, health, financial or pricing matters.

7. Labelling AI content (EU AI Act, Art. 50)

• Deepfakes: if an AI-generated or manipulated image, audio or video shows a real person, object, place or event in a way that may appear authentic, you must disclose when publishing that it was artificially generated or manipulated.
• Text on matters of public interest: if you publish AI-generated text to inform the public on matters of public interest, you must disclose that it was AI-generated – unless a human has reviewed it and holds editorial responsibility for it. In Sayweek a human approves every post, but the responsibility is yours.
• Machine-readable marking: images generated in Sayweek carry a machine-readable marking (IPTC “trainedAlgorithmicMedia”, XMP metadata). It is not a signed C2PA credential, and some platforms remove it on upload; please don’t strip it. [whether IPTC/XMP marking is sufficient for Art. 50(2) or C2PA / a watermark is needed; whether Sayweek is a provider or a deployer under Art. 50]

8. The social platforms’ rules

The platforms apply their own AI labelling rules, which may be stricter than the law:

• Meta (Facebook, Instagram): realistic AI-generated or altered images, video and audio must be disclosed; Meta may add an “AI info” label to the post.
• TikTok: realistic AI content must be posted with the “AI-generated” label.
• YouTube: when uploading you must disclose if the video contains realistic altered or synthetic content.
• LinkedIn: may show the origin of images and videos that carry C2PA credentials.

In Sayweek you can switch on AI disclosure per brand and channel: a short label (“AI-assisted”), the #AIassisted hashtag or – where the platform’s API supports it (TikTok, YouTube) – the platform’s own AI label. The disclosure is only added to the copy that is sent; if it does not fit, it moves to the first comment, gets shorter or is skipped (and this is logged). You decide which rule applies to your content; where Sayweek cannot set the platform’s label, add it on the platform after publishing.

9. Prohibited uses

You may not use the AI features to deceive, to impersonate real people without their consent, for unlabelled deepfakes, hate, harassment, spam, infringing others’ rights, manipulating elections or anything unlawful. The full list: Acceptable Use Policy.

10. Generated images and rights

Under the providers’ terms you may use the images generated for you, but Sayweek does not warrant that an image does not resemble someone else’s protected work, trade mark or a real person. Do not ask for images of real, identifiable people, other companies’ brands or copyrighted characters, and look at what an image shows before publishing it. When you upload your own photos, you warrant that you have the right to use them. To report infringing content: Copyright notices.

11. Questions or problems

If the AI suggests wrong, offensive or dangerous content, write to us: hello@sayweek.com. How we handle data is explained in the Privacy Policy.