AI Disclosure

{brand.name} holds no AI provider account of its own and never resells generation. You connect your own key, currently Anthropic or OpenAI, and every step runs on it, billed to you by that provider at their price. The service assembles the prompt from your project, sends it to that provider, validates what comes back, and stores the result.

That also means the terms that apply to the generation are the ones between you and your AI provider: their training policy, retention and regional processing, not ours.

A compiled prompt includes only what the step needs:

• Format profile: the working title, brand profile and any research notes.
• Research planning and claim extraction: the format profile, your source list and notes.
• Brief: the format profile, claims, sources and notes.
• Hooks, titles and packaging: the brief, claims and current script.
• Script drafting, analysis and humanizing: the brief, claims and the script itself.
• Shot list and image prompts: the brief, format profile, transcript and visual style.
• Repurposing: the brief, claims, script and chosen title.

Every response is validated against the expected shape before it is stored, because it becomes part of your project. We record which step it came from and which prompt version produced it, so a script version can be traced to what generated it.

We do not store the prompt text itself. It is rebuilt from your project whenever you ask for it.

Anything the model states as fact. Generated claims arrive marked UNVERIFIED and only a person can mark them supported. Idea scores are labelled as inference and come with a list of things to verify.

Dates, figures, quotations, studies and statements attributed to real people are exactly the categories where a language model is most confidently wrong. Attach a source before you publish.

Whether you tell your audience that AI assisted your work is your decision and may be governed by the rules of the platform you publish on. Throughline does not add watermarks or disclosures to your output.