Should brands disclose AI-written content?
Should you come clean that ChatGPT or Claude wrote your content?
This has been a hot debate on marketing webinars and Reddit forums ever since LLMs emerged, with different sectors converging on their own versions of best practice.
So, should you admit AI wrote it? Like so many things, it depends…
Full disclosure of AI content?
In high accountability sectors like academia and journalism, leading journals and media houses are starting to require full disclosure. They want to know which AI tools were used, how they were used, and assurances of a detailed human review of any content generated.
This is the same in many tender frameworks, particularly in the public sector. And there are plenty of high profile AI blunders that demonstrate why professional bodies like the Bar Standards Board and Royal Institution of Chartered Surveyors mandate human oversight of AI outputs.
AI-generated Vs AI-assisted
Many mainstream publishers, including Amazon, distinguish between AI-generated and AI-assisted content, and only require disclosure of the former.
This follows the reasoning that writing tools are nothing new. We’ve had translators and spell checkers for years. Used carefully in the hands of skilled writers, they speed up production and improve output.
We believe writing should be judged on the accuracy, honesty and humanness of the message, not the tools used. Also, the fuzzy grey area between ‘assisted’ and ‘generated’ is a potential quagmire.
So should brands disclose AI content?
It comes down to context and trust.
The BBC promises transparency because its use of AI ‘must never undermine the trust of audiences’. And where trust in human judgement is fundamental to value, investigative journalism for instance, use of AI is prohibited by leading publishers.
NEVER UNDERMINING THE TRUST OF YOUR AUDIENCE is the key point.
We advise against using AI copy fresh out of an LLM. Firstly, because it’s likely to contain errors. But more importantly, because your readers can tell – and for certain businesses, in certain cases, the potential impact on trust is not worth the risk.
Use AI, by all means. But a human review to check for factual errors and weird AI blunders is the bare minimum. Edit with extreme prejudice, because up to 70% of AI-generated copy is usually fluff. And for God’s sake, try and make it sound like a human wrote it.
Once you’ve checked and edited heavily, (and unless obliged to for contractual, compliance or legal reasons) we don’t think you need to disclose every time you use AI to assist with content. A lawyer won’t point out which parts of a contract were drafted by their paralegal – although if you asked, you’d expect transparency.
In our view, it’s more important to have an honest answer ready about when, where and how your business uses generative AI, and definitely about how you protect yourself and your customers against the inherent risks.
How do readers spot AI content?
The stickier issue is what happens when audiences spot generative AI tells in brand comms. We’ve talked extensively about these tells. The list continues to grow, and includes framing through antithesis, loose phrasing, baffling verbosity.
The trust impact is dependent on the stakes; higher stakes equals higher impact. An AI-generated product page, for instance, isn’t likely to raise any eyebrows. But what about a high value press ad or c-suite targeted ABM campaign? What about an update from a CEO?
Where will customers feel misled by AI generated content?
Customers are likely to feel misled where:
We ask our audiences to actively engage: home & about pages, conversion points, thought leadership articles.
The stakes are elevated: crisis comms, rebrands or big-ticket sales & marketing comms.
The message is explicitly attached to a person: a personal LinkedIn post, a by-lined editorial or leadership statement.
Where your audience would reasonably expect human authenticity, AI content will impact trust.
We help businesses craft communications audiences recognise as human.