When Not to Use AI in PR

We use AI every day at Firebrand. I’m using it to help write this post, even. Agents track award deadlines, monitor coverage and find speaking opportunities. We use it for research, drafts and call prep. We spend a fair amount of time working out where else it might help.

We also spend time undoing things people have done with it.

Usually, someone has taken a shortcut that looks reasonable from their end. They generate a brief, rewrite a release or polish a quote, then send it over. What they don’t see is the work that follows: checking unfamiliar claims, restoring language that was there for a reason and figuring out which changes they wanted.

That’s frustrating for the agency. It also takes time away from the work clients hired us to do.

Here are a few places where we’d rather you put AI down.

Briefs and RFPs

A good agency takes your brief seriously. We read every word and assume you meant to include it.

If you give AI three bullet points and ask for a full brief, it will fill the gaps with assumptions you may not notice. Your agency won’t know which details came from you and which the model added. It might also pull in old messaging you stopped using two years ago.

We may spend time developing an angle you never wanted, or asking about a market you have no intention of entering.

Three paragraphs in your own words are enough to get us started. Tell us what you’re trying to achieve, who you need to reach and what’s getting in the way. We can ask questions from there.

The same goes for RFPs. We’re seeing more long lists of questions that turn out not to matter to the company asking them. When we ask for clarification, the answer is often, “Oh, you can ignore that.”

Before sending an AI-generated RFP, read it as though you’ll have to assess every answer. Do you need all that information to choose an agenc

Quotes for the media

Please give us your actual opinion.

It can be rough. Three bullets or a voice note will do. Your PR team can help make it clear and concise, and you can check that it still says what you mean.

An AI-generated answer gives us much less to work with. On a rapid-response opportunity, other sources may have asked the same model the same question. The result is several executives offering the same observation, just as the reporter is looking for someone with something different to say.

There’s also the publication’s policy to consider. Where an outlet prohibits AI-generated contributions, sending one can put the opportunity and the reporter relationship at risk. That affects your agency and the other clients it represents too.

Getting a response out quickly helps only if it’s worth using.

Edits to work that’s nearly finished

This one will be familiar to most agencies. A release has been through two rounds of review and is nearly approved. Then it comes back rewritten because someone asked AI to tighten it.

Some sentences are shorter. But a qualified claim has become an absolute one, a product detail has changed and a point removed in the previous round has returned.

Now we have to work through the whole document again. We also have to establish whether those changes were deliberate. Did you want a stronger claim, or did the model just make one?

If you want us to tighten a release, send that instruction. If the opening is dull, tell us. If the argument takes too long to get going, say where you lost interest. That feedback is more useful than a replacement draft whose decisions nobody can explain.

Contributed articles carry an extra risk. We’ve seen a hand-written piece receive an AI polish during review, then get flagged by the publication’s AI check. Whatever the merits of that check, the author and PR team still have a problem to resolve.

Pitches

We hold ourselves to the same standard.

AI is useful for researching a reporter’s recent coverage and helping us prepare. We don’t let it write the pitch.

The pitch needs to explain why this particular story belongs with this particular journalist. Research helps us make that judgment, but we still have to make it, and write an email we’re prepared to put our name to.

Thought leadership

This is where I’m most reluctant to hand over the writing.

If you ask a model for an opinion on an industry trend, you’ll usually get a plausible answer based on what’s already been said. It may be fine. It’s unlikely to contain the thing you’ve noticed through years of doing the job.

That might be a customer conversation, a failed launch or a decision you got wrong. Those experiences give you something to contribute beyond a summary of the discussion.

Writing also helps you work out what you believe. I often start with an idea and discover halfway through that I can’t defend it, or that the interesting part is something I’d treated as an aside. I’d be reluctant to lose that process just to publish more often.

Where we draw the line

When we ranked our own PR workflow by risk, research, list-building and monitoring were at the safer end. These are useful jobs for AI, provided someone checks the results. We use it for them constantly.

We take more care when the work represents someone’s judgment or goes directly to a reporter, customer or other person whose trust matters.

That matters more with the backlash against AI in communications. People are becoming less patient with material they suspect nobody bothered to think through.

We’ll keep changing how we use these tools. For now, my request is modest: send us the short brief you wrote yourself. Tell us what you think about the reporter’s question. Give us your feedback on the draft.

We can do a lot with that.