Why AI Cannot Do PR

AI cannot do public relations for you. Skeptical? That is fair: I’m certainly incentivized to say that you can’t replace me with a Claude subscription. Plus, the boundaries around AI capabilities shift all the time. I could claim that AI can’t compete with the quality of human writing and strategy—but even if that’s true right now, it may not always be.

But this is not just cope: there are structural reasons why you can’t replace an agency with an agent. If you’re unaware of these issues, you could torch media relationships and damage your brand reputation to the point that you’ll need extensive professional help to recover. 

Caster can provide that help, but it is a lot easier and more efficient to conduct media outreach for clients the press doesn’t already hate. For your sake and ours, you should understand why AI cannot substantively fulfill the function of a PR professional.

AI Doesn’t Know How to Talk to People

A client told me recently, “Your most valuable asset is your rolodex.” No lies detected: Caster has spent over 25 years building relationships with press, influencers, event organizers, and decision makers across the tech media landscape. We draw upon that mutual understanding and respect to put your stories in front of the right people, with context for why they should care. Good PR involves knowing your targets and engaging in back and forth on angles, audience, connections to other coverage and trends, exclusives. You need a real rapport.

AI struggles to get off the starting blocks with this kind of task. It isn’t even a question of whether or not a model is “smart enough” to build a media list or write a pitch. Let’s imagine that, someday, LLMs close the empathy gap and develop the ability to converse convincingly with a person (We are…a long way from that point at the moment, but I concede that it’s possible).  They would still struggle to identify the right targets, because they don’t have visibility into leading media platforms.

AI Doesn’t Know Who to Talk To

The vast majority of leading news organizations block AI bots from crawling their sites. They do so both to protect their content’s value and to reduce costs. The majority of Internet traffic is now bot-generated rather than human-generated. Traffic drives up infrastructure costs, and no publisher wants to pay for AI agents, who don’t click on ads, to access their site.

This might explain why AI is so bad at identifying pitch targets. I’ve tested this (“know your enemy,” right?), and the results are uniformly abysmal: Across models, I get short lists that include:

• A handful of the most famous journalists on the planet covering a given space.

• People who left the beat three years ago.

• Pay-to-play guest contributors, often directly employed by a client competitor. 

This makes sense when you understand that AI doesn’t have access to the best and most recent reporting in your space. It is perpetually behind on understanding the media landscape, so it can never really get to know the players.

Journalists Can Tell It’s Not Human

AI is having a bit of a PR crisis of its own. Over half of Americans are “more concerned than excited” about AI, and the backlash against “AI slop” has led many sites and apps (including LinkedIn) to add tools to flag, label, or block AI content. You might think AI created a great LinkedIn article or digital event flyer, but your audience, highly attuned to the style and cadence of AI content, may find it repulsive on principle.

No one is more attuned to or aggrieved by AI content than journalists. Some editors run every submission they receive through an AI detector like GPTZero and flatly reject anything that scores “possibly AI.”  Yes, such tools can be fooled by eliminating AI “tells” like em dashes; short, emphatic sentences; and contrastive negation (“it’s not this, it’s that!”). But that won’t work for long.

The European Union’s AI Act requires AI providers to mark any content generated with its point of origin. These machine-readable marks won’t be easily removeable: as Anthropic explained in a blog about how they are planning to comply, their watermarks will be baked into text generation at an algorithmic level. Text won’t say “written by Claude” anywhere visible to human eyes, but anyone with Anthropic’s detection API will be able to tell instantly with nearly 100% accuracy, even if you re-type the text by hand.

The only way to get around that kind of watermarking is to write the content yourself. When submitting anything to a platform that doesn’t accept AI content, or allows it to be tagged, it must be written by a person. Otherwise, your reputation is on the line. You can use AI to brainstorm and assist your research, but beyond that, you’re on thin ice.

AI Will Get Better—But Not At Everything

There are many aspects of my job where I don’t think AI excels. However, reasonable people can disagree about the quality of its output. In some areas, notably coding, it is undeniably narrowing the gap with human performance. But there are some things an LLM cannot do because of their very nature.

LLMs cannot be original or authentic. At the end of the day, that’s what the media and the public want from you. Using an LLM for outreach or content production can feel like a betrayal: when you attach your name or your company’s name to something, it needs to come from you. You can’t promise people a homecooked meal and then give them a can of Spam.

So, to return to my thesis: AI cannot do PR. As long as there are still real journalists and media outlets looking for stories, and real people on the Internet willing to engage, you need humans to talk to humans.