Playbook 7 — Generative AI Content

Why This Gets Its Own Playbook

Generative AI is where most SMBs actually meet AI risk day to day — not in a credit-scoring model, but in a marketing draft, a chatbot reply, or an AI-written customer email. It behaves differently from a predictive tool: outputs aren’t predefined, so you can’t fully anticipate what it will produce, which means the controls have to sit around the output, not just around the model.

Specific Risks to Watch For

• Offensive, biased, or culturally inappropriate content slipping into a public post.
• Incorrect or misleading factual claims (product specs, pricing, medical or legal statements).
• Output that reproduces copyrighted material or a specific person’s likeness/voice.
• A customer not realizing they’re talking to AI, and being misled as a result.

Pre-Publication Checklist — Any AI-Generated Content Going Public

• A human has actually read it, not just skimmed the first line.
• Any factual claim (price, spec, date, statistic) has been checked against a real source.
• It’s been checked for tone-deaf or culturally insensitive framing, especially for images.
• If it’s meant to sound like it came from a specific person, that person has approved it.

Customer-Facing Disclosure Checklist

Under the EU AI Act’s transparency requirement for limited-risk systems, and simply as good practice everywhere else:

• A chatbot or virtual assistant clearly identifies itself as AI, not a human, on first interaction.
• AI-generated images or video used in marketing are labeled where required or where it would otherwise mislead.
• There’s a visible, easy way for a customer to reach a human if the AI interaction isn’t resolving their issue.

Usage Logging — Keep It Simple

• Keep the prompt and the output for anything that goes public — not forever, but long enough to answer “why did it say that” if a customer complains.
• Note who reviewed and approved it before publishing.

IP and Copyright Checklist for Your Tools

• The vendor’s terms confirm commercial use is allowed, not just personal or research use.
• The vendor indemnifies you against IP infringement claims from generated output (Playbook 2).
• You’re not asking the tool to generate something in the identifiable style of a specific living artist, brand, or public figure without rights to do so.

Next Steps

For anything beyond routine marketing copy — customer-facing chat handling account issues, AI-generated content making factual product claims — treat it as Limited-to-High tier under Playbook 1 and apply the fuller review, not just this checklist.