Artificial intelligence can now draft a press release, a customer email or a product announcement in seconds, and much of what it produces is genuinely useful. That capability is real, and the time it saves is real. What it does not do is assume responsibility for what it writes. When an AI-generated message reaches a customer, a regulator or the press, accountability still belongs to the company that sent it, and that distinction matters more than the convenience that precedes it.
I work in strategic communications and reputation management, and the questions executives ask me about AI have shifted. The question is no longer whether to use these tools. It is where to let them operate and where human judgment has to stay in the process. That is the right question, and the answer is less about the technology than about what a company is willing to stand behind.
Where the risk actually sits
The danger in automated communications is not that AI writes poorly. It often writes well. The danger is that it writes confidently whether or not it is correct, and external communications are unforgiving of confident errors.
The scale of the accuracy problem is measurable. A 2025 peer-reviewed review of factuality in large language models found that even advanced systems produce inaccurate factual statements in roughly 5 to 10 percent of their responses to general-knowledge questions, and the rate climbs for specialized or less common subject matter. A tool that is wrong one time in ten, in language as fluent as the nine times it is right, is a tool that cannot be the last check on anything a company publishes.
Three failures recur. The first is inaccuracy, when a system states something untrue about a product, a policy or a commitment, phrased persuasively enough that no one questions it. The second is inappropriate personalization, when automated tailoring reaches a customer with information or a tone that is wrong for the moment and the message lands as careless or worse. The third is inconsistent claims, when different automated messages describe the same product or promise in conflicting ways, and the contradictions accumulate where customers and regulators can see them. Any of the three can reach thousands of people before a human notices.
Why external communications are different
Internal uses of AI are comparatively forgiving. A flawed draft summary or first attempt at an internal memo can be corrected before it matters. External communications do not offer that margin. Once a message reaches a customer, an investor or a journalist, it has been sent in the company’s name and carries the company’s accountability, regardless of how it was produced.
In regulated industries the exposure is sharper still, because a claim made to the market can carry legal and compliance weight. “The system generated it” is not a defense a board, a regulator or a customer will accept. The organization made the statement. The tool is simply how.
Where human judgment has to stay
The useful question is not whether to use AI but where a person must remain responsible. A few places are not negotiable.
Accuracy of factual claims requires human verification, because AI cannot reliably know what is true about your specific products, commitments and circumstances, and external statements have to be correct. Approval of anything carrying legal, regulatory or financial weight belongs with an accountable person, not an automated workflow. Judgment in sensitive situations, among them a crisis, a complaint or any high-stakes relationship, calls for human reading of context that a model does not have. Consistency with positioning, the work of ensuring every message reflects what the company actually stands for, remains a human responsibility because the company, not the tool, owns that position.
A practical way to think about it
The sound approach is neither banning these tools nor trusting them unsupervised. AI is well suited to drafting, to handling volume and to producing first versions that a person refines. It is not suited to being the final authority on what a company says to the world.
The companies getting this right treat AI as a capable contributor whose work is reviewed, not as an autonomous publisher. They decide in advance which categories of communication can move quickly with light oversight and which require deliberate human approval, and they make that line explicit rather than leaving it to individual judgment under deadline. The efficiency is captured. The accountability stays where it belongs.
The point that holds
The value of AI in communications is real and worth using. So is the principle that technology does not absorb responsibility. A company remains answerable for every message it sends, whoever or whatever composed the words. The organizations that keep that clear will get the benefit of these tools without mistaking speed for judgment, and they will be the ones still trusted when a competitor learns the difference the hard way.