# A Pre-Production Control Checklist for Healthcare AI UGC
Before producing healthcare AI UGC, complete one control record covering the viewer decision, presenter role, script claims, evidence, prohibited implications, source-media rights, privacy-clean inputs, synthetic-media disclosure, channel variants, accessibility tests, reviewers, version identity, monitoring, and stop conditions. Do not render while any required owner, source, permission, or review route is unknown.
This record is not paperwork after the creative decision. It is the place where the decision becomes producible. A good response tells an editor what may be made, a reviewer what must be checked, and a distributor where the finished version may run.
Start with an accept-or-stop decision
Name a production owner who can accept a complete brief or stop an incomplete one. The owner should not fill gaps by guessing. Missing evidence, uncertain identity permission, unknown data origin, or an undefined clinical-review route is a blocked input.
Mark every control as accepted with an attached source or exact approval, not applicable with an asset-specific reason, or blocked with the responsible reviewer and required artifact. Replace “fine” with evidence such as a licensed source-file receipt, approved script version, disclosure preview, or passing caption check; replace “to be confirmed” with a stop.
Complete the Pre-Production Control Record
The Pre-Production Control Record can be organized into connected fields. Each field should point to a source or decision rather than repeat a promise.
- Audience and the single decision the asset should support.
- Synthetic presenter role, voice, visual setting, and prohibited identities or experiences.
- Exact claim language, evidence, limitation, and clinical-review status.
- Approved source media, usage scope, vendor, retention, and generation settings.
- Privacy-clean input list and prohibited patient-derived material.
- Disclosure wording, on-screen treatment, caption treatment, and platform controls.
- Channel, audience, destination, call to action, accessibility plan, and measurement question.
- Version naming, reviewers, monitoring owner, pause events, and retirement date.
Keep the record linked to the final script and render. If the asset cannot be traced back to the accepted inputs, it has left the governed workflow.
What decision should the asset support?
“Promote the practice” is not a decision. A useful brief might help a viewer understand an administrative step, compare questions to ask, recognize the purpose of a service page, or decide whether to read a more complete resource. The desired action should match that level of understanding.
Also record what the asset will not do. It will not diagnose, recommend care for an individual, promise an outcome, manufacture a testimonial, or imply access and availability that have not been confirmed. These exclusions guide both script and visual choices.
Lock the claim and identity boundaries
The synthetic presenter should have an explicit role. State whether it is a fictional disclosed narrator or a permitted representation of a real person. Record the voice source and visual source. Prohibit patient, clinician, employee, or independent-reviewer implications that are not true and authorized.
The Federal Trade Commission's health-products guidance says marketers should identify both express and implied claims and assess the advertisement's overall net impression. This supports recording not only the script but also the planned visuals, captions, presenter role, and action. The guidance focuses on health-related products and does not supply a universal review answer for every healthcare service.
Qualified reviewers should resolve the exact clinical, advertising, legal, and identity questions. The production checklist surfaces them; it does not grant clearance.
Approve evidence before drafting variants
For every objective AI UGC claim, preserve the exact script sentence and planned on-screen text beside the supporting source passage, source date, applicable audience, limitation, evidence owner, and expiry trigger. A reviewer should be able to reject a changed caption or visual claim without searching through an entire policy or clinical document.
Do not start with a strong hook and search for supporting evidence afterward. If the evidence supports a narrower proposition, write the narrower proposition. A synthetic presenter's confidence must not exceed the source.
Keep source inputs privacy-clean
List the only folders and assets that production may use. Patient records, messages, calls, schedules, portal screens, photographs, testimonials, and distinctive narratives should remain outside the workspace unless a qualified, fact-specific process expressly authorizes the use.
HHS de-identification guidance recognizes formal approaches and notes that re-identification risk is not necessarily zero. A creative team should not label casual redaction or detail blending as formal de-identification. Record uncertainty as a stop, then route it to the organization's privacy and legal reviewers.
Specify the disclosure as a production element
Write the synthetic-media disclosure, timing, size, contrast, duration, spoken treatment where appropriate, caption placement, and platform upload setting. Preview how platform interface elements may cover it. Define who checks the live label after publication.
Disclosure should be understandable before the realistic presentation creates a false premise. It also remains separate from claim review, rights, and clinical accuracy. Marking an asset as generated does not validate what it says.
Place the proposed label on the opening frame, thumbnail, paid placement preview, and landing-page embed, then test it at the smallest intended mobile size. Record any crop, interface overlay, or playback state that hides it and block that variant until placement is corrected.
Test the planned experience before rendering
A storyboard or rough animatic can reveal problems before the expensive final output. Test the opening without sound, the caption density on a small screen, the disclosure placement, the call-to-action transition, and the visual cues around identity and clinical setting.
The test plan should include:
- Transcript-to-source comparison for every objective claim.
- Caption accuracy, timing, readability, and a usable sound-off path.
- Face, hands, anatomy, equipment, uniforms, badges, screens, and background checks.
- Platform crop, overlay, thumbnail, destination, and disclosure preview.
- Final human review of the exact exported file.
Accessibility obligations and appropriate tests depend on the actual experience and organization. Use qualified accessibility guidance where needed rather than treating a caption file as complete assurance.
Assign reviewers by question
Editorial review checks clarity, usefulness, source fidelity, and absence of fabricated experience. A licensed clinical reviewer should handle medical claims when required. Privacy, legal, advertising, rights, accessibility, security, and operational reviewers should address the questions within their qualifications.
Record reviewer role, version, date, decision, conditions, and unresolved issues. Never convert silence or a meeting invitation into approval.
Route each question with the exact artifact attached: medical wording to the clinical reviewer, disclosure and objective claims to advertising or legal review, prompt and source-asset handling to privacy review, and caption plus interaction behavior to accessibility review. A later edit reopens only the decisions its changed text, visual, data, or placement can affect.
End with owners and stop conditions
NIST's AI Risk Management Framework uses govern, map, measure, and manage functions to organize risk decisions across the AI lifecycle and emphasizes documentation and accountability. That supports a pre-production record connecting intended use, owners, testing, and monitoring. The framework is voluntary and does not certify a vendor, clear a campaign, or guarantee that generated media is accurate.
Name who monitors the live version and what causes an immediate pause. Examples include a missing disclosure, inaccurate render, changed service fact, stale source, disputed identity use, exposed patient information, broken destination, unsafe comment pattern, or direction from a qualified reviewer.
Production begins when uncertainty is controlled
The checklist gives healthcare social media marketing teams a faster start because production receives bounded choices instead of unresolved risk. It also makes later changes easier to assess because the original claim, presenter role, inputs, disclosure, and release scope are visible.
AI UGC campaigns can help turn an accepted record into controlled scripts and synthetic-presenter variants. The healthcare organization retains responsibility for applying qualified clinical, privacy, legal, advertising, accessibility, rights, security, and platform review to its facts and final deployment.