AI in PR: Can Agencies Scale Without Losing Judgment?

The problem is the delivery model, not just the writing

PR work is more than turning a draft into something polished. It includes research, understanding a client's business, finding a credible story, identifying relevant journalists and publishers, building relationships, and responding to feedback. Some of it is repeatable preparation. Other parts are context-heavy and judgment-based.

If an agency charges mainly for hours, automation makes it harder to justify pricing. A task that used to take several hours might take less time with software. At that point the agency has to decide whether it will charge less, deliver more, or change what the client feels they're buying. Adding an AI tool doesn't magically solve that tension.

The real question is whether technology improves the whole service outcome: more relevant opportunities, faster preparation, stronger client retention, and/or lower delivery costs without a quality drop. If all it does is generate drafts quicker, the commercial impact may be pretty limited.

Why smaller businesses matter to the opportunity

Hamdi points to a global market of 400 million small and medium-sized businesses. Many of those companies can't afford a traditional PR agency - or don't have the time and confidence to hire one. A more prepared, technology-supported service could make professional PR available to more of them.

But there's a catch. A lower-cost model only works if the agency can keep delivery efficient without turning everything into generic outreach. Smaller clients still need a credible story and relevant targeting; they may simply have fewer internal resources to provide the context an agency would normally gather itself.

So for agency leaders, the useful test isn't "Can AI do PR?" It's: which parts of the service can be made more repeatable without losing the expertise clients actually value? That's the difference between operational improvement and automation for its own sake.

Use preparation to improve the first conversation

This is one of the more straightforward uses of AI: shift more of the research and structuring to before a person ever enters the room. A prospect should be able to tell the agency has actually considered their business, instead of receiving a generic pitch.

Hamdi says prospects have responded as if Pathos is already doing part of the publicist's job. One reaction he mentioned was, "I feel like you're already my PR firm." That's an anecdote, not proof of conversion rates or long-term satisfaction. Still, it captures the core idea: preparation can show relevance before a contract is signed.

If another agency tries something similar, the workflow needs clear boundaries. Automated research can suggest a story angle, but a person has to verify that the angle is accurate, timely, and right for the company. The point is to reduce time-to-a-useful conversation, not to make confident claims from incomplete information.

Match stories to publishers, then check the fit

For delivery, Pathos says it uses algorithmic matching between client stories and publishers to improve the odds that content lands. Matching can help teams consider more combinations of story and outlet than they could review manually - especially as a publisher network grows.

But a match isn't the same thing as editorial interest. A system may spot topical overlap; it still can't guarantee that a publisher actually wants the story, that the timing works, or that the pitch fits the outlet's standards. The human team has to assess relevance and decide whether outreach is worth the effort.

The product roadmap matters here, too. Pathos's results announcement for the six months ended 30 June 2026 said its AI virtual publicist, Pressella, and PathosMind remained on track for general availability in the first half of 2027. That's a planned availability date - not evidence that a broad launch has already happened or that results have been proven at scale.

That distinction matters for buyers and competitors. Pathos's current workflow shows how it's using technology internally. The planned products are another step - one that still needs to be tested as availability expands.

Judgment is not a final proofreading pass

It's tempting to treat "human involvement" like a quick review at the end of an automated process. That can work when the task is low-risk and the output is easy to verify. In PR, an angle can be factually plausible but strategically wrong. Something can be accurate but still reveal information at the wrong time. A pitch might be well written and still irrelevant to the journalist reading it.

Human judgment needs to shape the work, not just clean up wording. That means checking the evidence behind an idea, making sure it fits the client's objectives, and deciding whether outreach is actually appropriate.

It also affects accountability. If an automated suggestion leads to a misleading claim - or a poor outreach decision - the client won't see the model as responsible. The agency does. The team needs to be able to explain how its people checked the work.

Trust depends on relevance, not volume

Automation can generate more material and contact more prospects. But more output doesn't automatically mean more useful coverage. Unfocused pitches waste journalists' time and can damage a client's credibility. More volume can create more noise instead of more results.

A practical rule is to measure relevance before volume. Teams should be able to explain why a story fits a specific outlet and why the timing makes sense. If an AI system can't provide enough context for a person to evaluate its suggestion, that output shouldn't immediately turn into outreach.

This matters even more when you're serving smaller businesses. A client with limited communications experience may miss weak recommendations. In that case, the agency has more responsibility to separate a real opportunity from something that merely sounds persuasive.

Give the system a defined role

A practical workflow assigns responsibility at each stage:

• Research: software gathers and organizes information; a person checks important facts and sources.
• Story development: AI can propose angles; an experienced practitioner decides whether they're accurate, timely, and useful.
• Publisher matching: software can rank possible outlets; a person verifies editorial fit and decides whether to make contact.
• Outreach and response: technology may help with preparation, but relationship management and sensitive communication stay with the team.
• Review: the agency tracks corrections, client feedback, and outcomes to improve the process.

This isn't saying every PR task must be handled manually. It's about making the boundary clear, so efficiency gains don't silently shift risk onto clients or publishers.

Scale the system, not only the headcount

Many services businesses scale by adding people as client work increases. That can be the right move when quality depends on close, individual attention. But it can also squeeze margins if each new account requires a similar jump in staff capacity.

Technology can change that balance by taking on preparation, organizing information, and supporting repeatable decisions. It doesn't eliminate the need for people - it changes what they spend time on. If staff spend less time gathering basic context and more time on strategy and relationships, productivity can rise. If the saved time turns into endless checking of unreliable outputs, the productivity benefit can disappear.

Hamdi wants Pathos to evolve from a services firm into a technology business with a service layer. He's also said Pathos is adding larger and stronger publishers to its platform to support better margins. That only works if the publisher network is actually useful, the matching is reliable, and client demand is strong enough to justify the system in the first place.

Treat repeat business as a quality signal, not a complete answer

Recurring revenue matters because it suggests clients keep seeing value. In Pathos's reported first-half results, recurring customers were a larger share of revenue than in the prior-year period. That's a positive operational sign, but it still needs context: retention rates, the economics of serving retained clients, and the reasons customers renew.

Yahoo Finance reported on 15 October that Hamdi discussed rising repeat and recurring revenue, the goal of building a sustainable, profitable listed business, and a planned first-half 2027 launch of the AI-powered virtual publicist. Those points fit the company's stated direction. But the product's performance in the wider market is still an open question.

An agency shouldn't treat a growing customer count - or faster content production - as proof that the economics are truly scalable. A durable model needs evidence that service quality holds, clients stay, delivery costs stay controlled, and the business can handle exceptions without damaging margins.

Use a small measurement set

A practical scorecard can use four groups of measures:

• Efficiency: time spent on research, preparation, and revisions per client or campaign.
• Quality: human approval rates, factual corrections, client feedback, and the relevance of proposed outlets.
• Business health: client retention, recurring revenue, delivery cost, and profitability by service type.
• Risk: inaccurate claims, unsuitable outreach, complaints, and the amount of rework caused by automated suggestions.

No single metric tells the full story. For example, lower preparation time isn't a win if it leads to more corrections or weaker client confidence. And higher outreach volume isn't automatically good if response rates or editorial fit decline.

Set a baseline before changing the process. Then compare the AI-supported workflow with the old one using similar kinds of work. That won't eliminate differences between clients or campaigns, but it's still better than assuming the tool "caused" improvement just because it was introduced at the same time.

Expand only when the operating evidence holds

A sensible rollout starts with a bounded task, a named human owner, and a clear review rule. Run it long enough to see where time is actually saved - and where staff still need to step in. Track failures as well as wins; otherwise you'll only learn from outputs that happened to go well.

Scale when three conditions hold: the work stays accurate and relevant, the client experience is at least as good as before, and the economics improve after you include review and correction time. If any one condition fails, tighten the task boundary or stop using AI for that part of the workflow.

This approach is less flashy than calling an agency "AI-first," but it creates decisions operators can actually defend. It also helps clients understand what the technology does, what humans are accountable for, and how the service is evaluated.