The risk of AI-assisted content is not that it sounds synthetic. It is that volume outpaces verification. A team that could publish four pieces a week can now publish forty, and the review process that was adequate at four quietly becomes decorative at forty.
What a human must verify
- Every factual claim, statistic and citation — generated references are frequently plausible and wrong.
- Regulated claims in health, finance, employment and similar categories, against both advertising policy and local rules.
- Attribution and likeness: whose words, image or voice appears, and whether permission exists.
- Anything presented as first-hand experience, testimonial or result. These cannot be generated.
Disclosure and governance
Disclosure obligations differ by platform, format and jurisdiction, and they change. Rather than a fixed policy, most teams need a decision rule: who determines whether disclosure applies, what the default is when it is unclear, and where the decision is recorded.
Governance that works is boring: a named reviewer per content type, a written escalation trigger, and a log that can be produced later. Not a committee.
AI is a tool. The judgment about what is true, permitted and appropriate remains human, and remains the differentiator.
Sources & references
- Platform AI content disclosure and labelling guidance
- Advertising standards guidance on substantiation and testimonials