What AI Can't Do for a Communications Team

The case for AI tools in communications is easy to make and mostly correct. They draft fast, research fast, build fast. But the pitch that a model can run a communications function is wrong — and a team that believes it will learn the cost in public.

Here is the honest list of what these tools cannot do. Read the other way, it's the list of what a communications team is actually for.

Quick answer. AI tools can't own a reporter or client relationship, can't reliably tell true from plausible, can't read timing and political context, and can't carry accountability when something goes wrong. They accelerate the production work. They don't replace the judgment.

It can't own the relationship

PR runs on trust built between people over years — a reporter who picks up because of who's calling, a client who shares the hard news early because the team has earned it. A model can draft a flawless pitch. It cannot be the reason a journalist opens the email. The relationship is the asset, and it doesn't transfer to software.

It can't tell true from plausible

This is the one that does damage. The systems generate fluent, confident text — and they will produce a statistic, a source, or a quote that does not exist when they don't have a real one. In a press release, a fabricated figure is a correction. A fabricated executive quote is a credibility event. The tool has no internal sense of the difference between accurate and plausible. Every factual claim it produces is unverified until a person verifies it.

It can't read the room

Communications is timing and context. Should the statement go out today or after the hearing? Does this phrasing invite the regulatory question legal is worried about? Is the client's competitor about to make this the wrong week to announce? Those are reads on people, politics, and the news cycle. A model sees the words. It does not see the room.

It can't carry the accountability

When a release misfires, the client doesn't call the chatbot. Accountability sits with the team and the firm — always. That's not a technicality. It's the reason the judgment can't be delegated: the consequences can't be either. A tool that bears no risk cannot be handed the decision that creates the risk.

It can't protect what's confidential

A consumer AI tool is, by default, a surface the team does not control. Embargoed news, deal details, a crisis not yet public, personal contact data — none of it belongs in a chatbot without checking the tool's data settings and the client's policy first. (More: AI and Client Confidentiality.)

What it can do — and it's a lot

None of this argues against the tools. They remove the slow, low-judgment work — the blank-page draft, the format, the summary of forty pages, the first version of the talking points. That's real time returned to the team. The point is the division of labor: the tools do the production, the team does the judgment. Get that line right and the tools are a multiplier. Get it wrong and they're a liability with good grammar.