What does AI-transformation mean for PR?

If you use a PR agency, AI-transforming marketing means you need them to be AI-transformed. But most agencies are far from it. Sure, they all use ChatGPT or Claude. That's table stakes.

According to a recent study, just one in eight — 13% — say their AI is "highly integrated."¹ Worse, only 12% use AI agents.²

Does this matter? Isn't it enough to give everyone in the agency a seat on ChatGPT or Claude? That's a start. It's a very small step.

AI transformation is hard

The goals of AI transformation are simple:

• Better
• Faster
• Cheaper

Really. It is that simple. But achieving AI transformation is not simple. Let's start with what you actually need to do.

Consider a landmark MIT study that examined 106 studies of people working with AI.³ One of their questions was whether combining humans with AI helped or hurt the results of the task being accomplished.

First, the surprising finding: human–AI teams performed significantly worse than the best of human or AI alone. That's tricky to understand, so here is a simple example.

Suppose a human alone scores 55%. AI alone scores 73%. Together they score 69% — better than the human, worse than the AI. That's the paradox. Adding AI helped the person. Adding the person hurt the AI.

So far it seems like the human is the problem. The human drags AI down. We should turn AI on and send the human home.

But the study dove deeper. It turns out it depends on the task. One experiment ran the same interface, the same people, and the same AI against two different tasks — spotting a fake review online, and classifying a bird from a photo:

Task	AI alone	Human alone	Human + AI
Spot fake reviews	73%	55%	69%
Classify a bird	73%	81%	90%

The net: when AI is better than a human, adding a human makes things worse. When a human is better than AI, adding AI helps.

Three strategies, not one

Let's apply this to PR. A task where humans are clearly better than AI is pitching a prominent journalist on a story. A top PR pro with decades of experience and a familiarity with that journalist stands a far better chance of succeeding. But add AI, and guess what you get: better, faster, cheaper.

We call this AI Wingman. AI becomes the R2-D2 to the PR pro's Jedi Knight.

What about tasks AI is better at? Like coverage analysis. Mercedes-Benz garnered 31,802 articles in traditional media outlets in 2025. AI can find, enrich, analyze, grade, and report on those articles in hours. A human reading and tagging each one at five minutes apiece would need roughly 2,650 hours — sixteen months of full-time work, for one brand, for one year. So for coverage analysis, AI is better, faster, cheaper.

We call this Agentic AI. AI automates the task from start to finish. The PR pro simply gets the results.

Some tasks are a combination. Tools like Qwoted send agencies hundreds of journalist requests every morning, asking for help on articles they're writing. An agentic solution delivers only the requests that match a client's focus. But the PR pro crafts the pitch.

This is Human-in-the-Loop.

Why AI transformation is difficult

The first strategy — AI Wingman — is pretty easy. Yes, you have to teach your staff how to prompt, the rules of working with AI, and a host of other things. But AI literacy among PR professionals is generally high, and this phase gets established quickly.

This is where most agencies stop.

Agentic AI is a different species of problem. In 2015, Google engineers published a paper on what it actually takes to run machine learning in production.⁴ Their most-cited illustration has become famous in engineering circles for one reason: the AI itself is a small box, surrounded by nine much larger ones.

The model is the easy part. Configuration, data collection, verification, monitoring — that's the job. And every one of those is software that has to be built, hosted, secured, and maintained forever. It isn't a project with an end date. It's a permanent engineering function inside a PR agency.

Then there's the math nobody mentions. Agentic AI works by chaining steps together, and each step has an error rate. Those errors don't average out. They compound.

The Berkeley-led team behind Measuring Agents in Production — the first large-scale study of AI agents actually running in the wild, accepted to ICML 2026 as an Oral Presentation — found that the median production agent runs just five steps before a human intervenes.⁵ Five. Suppose each of those steps is 95% accurate, which for messy real-world text is generous.

Twenty-three reports in a hundred come out wrong. And you cannot tell which twenty-three. Charts render. Numbers total. Sentences parse. A coverage report built on a broken pipeline looks exactly like one built on a sound pipeline.

That's five steps — the median. A pipeline that genuinely analyzes an enterprise corpus runs many times that.

This is why serious systems don't just run the pipeline. They check it. Every output reconciled against source data, every stage verified end to end, automatically, before anything reaches a human. Not a person auditing the machine — that doesn't scale either. The verification has to be built into the system from the start.

Almost nobody does this.

And Human-in-the-Loop is harder still. It requires everything Agentic requires, plus a judgment call about precisely where the machine stops and the person starts. Put that line in the wrong place and you get what most "AI-powered" PR tools produce: fluent, confident, and obviously written by something that has never met your market.

Then there's the failure nobody talks about. Suppose the machine gets it right 90% of the time. By week three, the person reviewing the output is approving on autopilot. They're still in the loop on the org chart. They stopped reading a month ago. You're paying for oversight you aren't getting — and the errors that slip through are exactly the ones a reviewer was there to catch.

Precision AI-Transformation

AI transformation offers three techniques:

• AI Wingman
• Agentic AI
• Human-in-the-Loop

The decision of which method to use depends on the task. For tasks where humans are better, AI Wingman is best. When AI is better, Agentic AI is best. And when it’s a mix — some parts of a task where AI is better and some where the human is better — Human-in-the-Loop is the answer.

That’s what Precision AI-Transformation is for. Matching the method precisely to the task. Connect Marketing uses 100% Precision AI-Transformation to deliver Quality at the speed of AI.

What this means for you

You're not going to build this. Neither is your team. That's not a criticism — it isn't your job, and the agencies that have done it spent years on it.

What you can do is tell the difference.

Start by being suspicious of anyone claiming they've automated everything. They haven't, and you wouldn't want it if they had. A machine could be taught that one editor wants ninety words and no adjectives while another needs the setup explained. There are hundreds of editors. Each gets pitched a handful of times a year. It would take a year to build and would never pay for itself. So a good agency doesn't. Your pitch gets written by the person who has known that editor for twenty years. That's a decision, not a gap.

Then ask four questions.

1. Which of your work is automated, and which isn't?

   A real answer has things in both columns.

2. Who writes the pitch?

   If it's a machine, ask to see one.

3. How do you verify what your automated systems produce?

   If the answer is vague, there is no verification.

4. What did you decide not to automate, and why?

   No answer means nobody ever decided.

So — what does AI-transformed actually mean for PR? Not that a machine does everything. It means every task in your program has been examined, assigned to the strategy that fits it, and built accordingly. All three strategies. Most agencies stopped at AI Wingman.