Policy Purpose
Use this opening: We use AI to support research, organisation, transcription, drafting, translation, editing, and media production. We do not use AI to fabricate first-hand experience, customer testimony, expert identity, evidence, or endorsement. A named human remains responsible for the final claim. We disclose material AI transformations when required by law, policy, contract, or audience understanding.
The policy covers public articles, social posts, video, audio, images, newsletters, sales assets, and customer education.
Classify the AI Use
Class A covers assistive, low audience-impact uses: spelling and grammar checks, transcript generation, formatting, internal summarisation, and idea organisation. The requirements are human review, an internal record for important assets, and public disclosure only when required or materially useful.
Class B covers substantive transformation: major rewrites, translation or synthetic dubbing, generated illustration, AI-selected clips, summaries of contested material, and material background replacement. The requirements are source verification, a named approver, a transformation record, and public creation context when a reasonable audience would interpret the asset differently.
Class C covers synthetic identity or synthetic events: AI avatars, voice clones, deepfakes, simulated testimonials, generated documentary-style scenes, and synthetic experts or customers. The requirements are explicit rights and consent, legal and policy review, clear disclosure with the asset, a provenance record, senior approval, and no publication if the audience could reasonably mistake fiction for real experience and the disclosure cannot correct that.
Consent and Rights
Use this policy language: We do not clone or materially synthesise a person's voice, face, likeness, signature style, or first-person account without explicit, documented, use-specific permission.
Consent should cover the asset and model used, channels, languages, duration, editing, future versions, commercial use, revocation or expiry, storage and deletion, and who may approve releases.
A platform's technical capability does not establish legal or ethical permission.
Disclosure Decision
Ask whether the content depicts a real person saying words they did not record, depicts a generated or materially altered real event or place, presents an avatar as an expert, customer, employee, or independent reviewer, changes the apparent evidence, result, or endorsement, needs creation context for the audience to interpret the claim, or triggers notice under law, platform policy, contract, or industry rules.
If any answer is yes, escalate to the appropriate disclosure and review level.
The EU AI Act's Article 50 includes transparency obligations for several AI and synthetic-content contexts. The exact applicability needs legal analysis.
The FTC's Disclosures 101 concerns material relationships in endorsements, not AI generally. It illustrates that disclosure should be clear, conspicuous, and close to the relevant message.
Disclosure Wording
Use specific language for a real interview with AI support: This clip comes from a real recorded interview. AI transcribed the answer and suggested candidate highlights. The speaker approved the final cut.
Use specific language for synthetic translation: The video shows the original speaker. The Spanish voice is an AI-generated translation reviewed and approved by the speaker.
Use specific language for a generated visual: Illustrative scene generated with AI; it does not depict an actual event.
Use specific language for an avatar: This AI-generated presenter reads a script reviewed by the team or named role, and the presenter is not a customer or independent expert.
Avoid vague labels such as AI-powered, enhanced, or a distant policy link when the audience needs immediate context.
Human Review
Every public asset must have an owner who checks the main claim, primary sources, quotations, names, dates, numbers, scope, uncertainty, rights, privacy, disclosure, caption, title, audience fit, and corrections route.
High-stakes medical, legal, financial, scientific, safety, employment, or public-interest content requires qualified domain review.
Google's guidance on AI-generated content emphasizes accuracy, quality, relevance, and useful creation context. That is search guidance, not a substitute for legal review.
Source and Transformation Record
Retain the asset ID, original sources, prompts where material, model and tool, AI output, human edits, consent, reviewer, disclosure text, final file, publication URL, review or expiry date, and correction history.
Define access, retention, and deletion. Do not preserve sensitive prompts or recordings indefinitely without a purpose and lawful basis.
C2PA offers technical standards for content provenance. A small team can also begin with an internal record. Neither proves the truth of the claim.
Prohibited Uses
Suggested prohibitions include fabricated customer or employee testimony, invented first-person experience, unapproved voice or likeness cloning, fake evidence, fake quotations, fake citations, fake results, deceptive synthetic documentary media, removing material caveats for impact, putting confidential data in unapproved AI systems, automated publication of high-risk claims without human review, and evasion of platform or legal disclosure requirements.
Corrections and Incidents
If a material problem is found, pause distribution, preserve the source and versions, notify the owner, assess audience and legal impact, correct, label, or remove the asset, notify affected people where appropriate, document the decision, and update the policy or workflow.
Do not silently swap a file when people relied on the original claim.
Ownership and Review Cadence
Assign a policy owner, legal or external escalation contact, tool-approval owner, content reviewer, security and privacy reviewer, and incident owner.
Review the policy when law, platform rules, tools, or use cases change, and on a fixed schedule suitable for the team.
Honest Limitations
No policy can enumerate every tool or fact pattern. Over-disclosure can confuse audiences; under-disclosure can mislead them. Teams need judgment and current advice.
Disclosure does not make a harmful or false asset acceptable. Fix the content, rights, and process first.
How REC Fits
REC Content Studio uses AI to research supplied context, prepare a guided solo interview, transcribe the real answer, and suggest grounded highlights. The person remains the visible source.
REC does not determine legal obligations, obtain every consent, generate compliance labels, or issue a C2PA credential. Apply the policy: classify the AI role, verify the answer, preserve the source, and let a human approve the final clip.
Adopt the policy before scaling output. The easiest disclosure is the one supported by a workflow that already knows what happened.