AI Ate the Junior Comms Job, Judgement is Still Hiring

Two things are true at once right now. AI has made writing better, on average, and it has made almost nobody more trustworthy for it. A 2023 study of consultants at BCG by Fabrizio Dell'Acqua and colleagues found that access to GPT-4 raised the quality of consulting work across the board, and did the most for the weakest performers, who improved by roughly 40 percent. More recently, a 2026 systematic review found generative AI raised worker productivity by an average of 21 percent, and in the large majority of the studies reviewed, the gains were biggest for the least experienced workers. That's a real, and by now well-replicated, effect. But a separate line of research by Oliver Schilke and Martin Reimann, running thirteen experiments across professions from consulting to teaching to investing, found something less comfortable: the moment people learn AI was involved, they trust the person or organisation behind the work less, regardless of whether the work itself got better. 

Here’s one for the skim readers: The bar for competent writing is rising fast, for everyone, at the same time. Trust isn't rising with it. If anything, it's getting harder to earn, because the craft that used to signal effort and credibility no longer signals much. Everyone has access to it now.

So what does trust actually run on? Not writing quality itself, that was only ever the signal. Underneath trust sit two things: distinctiveness and legitimacy. AI is structurally weak at supplying either.

On distinctiveness: Anil Doshi and Oliver Hauser ran an experiment where writers produced short stories with or without AI-generated prompts. Individually, the AI-assisted stories were rated more creative and more enjoyable. Collectively, they were 10.7 percent more similar to each other than stories written without AI. AI drafts toward the statistical middle of what's already been written on a subject, because that middle is what it was trained on. Take a small, specialised field like digital public goods, where I’m working on some projects: most organisations already circle the same handful of terms—interoperability, inclusion, open standards—that middle is already crowded. A writer who's read enough of the field to know which framing has been used one too many times this year is offering something no model can: originality relative to a specific, narrow professional conversation.

Legitimacy is the harder one. It's built by a communicator's actual understanding of the environment they're writing into: what a partner ministry will read into a particular phrase, which claim needs to stay marked as projected rather than confirmed, what a technical audience already knows that a donor audience doesn't, when the politically careful version of a sentence also happens to be the honest one. None of that lives in a training set. It lives in relationships, institutional knowledge, and judgment built by being embedded in a specific political and cultural context over time. AI can draft a sentence. It cannot know that this sentence, from this organisation, at this moment, is the wrong one to publish.

And that's what you're paying for when you hire proper comms: not for typing speed or beautiful proses, but for fluency in a field where getting the politics wrong costs more than the ChatGPT subscription (yes it is the better writer than Claude).

What this means for junior comms professionals is that they need to learn more quickly. The entry-level skills, clean drafting, competent summarising, reliable tone, are the skills AI now performs adequately, for free, on demand. That used to be how juniors earned a seat at the table while they slowly absorbed everything else: the politics, the history, the unwritten rules of a sector. That slow runway is largely gone, because the work that used to buy juniors that time is the work that's been automated first.

But there's a more hopeful way to read this. If the entry-level tasks are the ones AI absorbs first, the actual apprenticeship shifts to something more interesting: reading the debates, following the politics, understanding why an initiative succeeded or stalled, arguing about it with people who've been in the room longer than you have. Which, it turns out, was the best part of the job all along.