Last month, a university communications director told me her team had started using ChatGPT to draft research news stories. "It's so much faster," she said. "We feed it the abstract and the press release, and it writes the article in seconds. We just tweak it a bit."

I asked her to show me one. It was fine. Grammatically correct. Structurally sound. The kind of writing that provokes no objection — and no interest. It read like a competent intern had been given a template and told not to colour outside the lines.

That's the problem. AI can produce competent writing. But competent writing is not what makes people read a research magazine, remember a story, or act on what they've learned. Competent writing is invisible. It's the writing nobody notices — and nobody finishes.

The AI Temptation Is Real

I understand why university communications teams are reaching for AI. They're underfunded, understaffed, and under pressure to produce more content across more channels than ever before. A tool that can generate a 600-word article from a research abstract in 30 seconds feels like survival.

And for some use cases, it genuinely is. If you need a serviceable summary of a grant announcement for the staff intranet, AI will do that job well enough. If you need a two-paragraph briefing note for an internal meeting, AI will save you twenty minutes.

But the content that matters — the content that shapes how external audiences perceive your institution, the content that donors read before deciding whether to give, the content that policymakers cite in evidence — that content needs to be more than competent. It needs to be compelling. And AI doesn't do compelling.

What AI Gets Wrong About Research Writing

1. AI doesn't know what's interesting

AI models are trained to produce the most probable next word in a sequence. That makes them good at producing text that sounds like everything else in their training data. It makes them terrible at identifying what's genuinely surprising, counterintuitive, or newsworthy about a piece of research.

When a journalist interviews a researcher, they're listening for the detail that doesn't fit. The unexpected finding. The experiment that failed before it succeeded. The quote that reveals personality. An AI, given a research abstract, will produce a faithful paraphrase of that abstract. It won't find the buried lede because it doesn't know what a lede is — it only knows statistical patterns in language.

2. AI writes in the voice of the internet

AI-generated text has a distinctive quality. It's smooth, balanced, and anodyne. It hedges. It generalises. It favours safe formulations over specific ones. It never takes a risk because it's been trained on the average of everything, and the average of everything is anodyne by design.

Read enough AI-generated university content and you'll notice the same patterns: the same sentence structures, the same transition phrases, the same ratio of adjectives to evidence. Individual institutions disappear. Individual voices disappear. Everything reads like everything else.

This is a strategic problem. Your university's research communications are supposed to differentiate you. They're supposed to communicate not just what research you do, but what kind of institution you are — your values, your priorities, your intellectual character. AI flattens all of that into a single, generic institutional voice that sounds exactly like every other university using the same tools.

3. AI can't interview anyone

Great research writing depends on great quotes. And great quotes come from real conversations — the kind where a researcher relaxes enough to say something honest, personal, or surprising. AI can't have that conversation. It can only remix and rephrase what's already been written down.

The result is research communications filled with quotes that sound like a press release written them: "We are excited by these findings, which open new avenues for understanding..." Nobody ever said that in a genuine conversation. But AI thinks they did, because AI has read a million press releases and they all said something similar.

4. AI can't verify anything

AI doesn't know what's true. It knows what's probable. For research communication, this is a fundamental problem. Getting a detail wrong — even a small one — undermines trust. When a human journalist writes about research, they check facts, confirm claims with sources, and verify that what they're publishing is accurate. AI hallucinates with confidence.

You can of course fact-check AI-generated text. But at that point, you're doing most of the work anyway — and you're doing it on content that was designed to be generic. You'd be better off writing the piece yourself, or hiring someone who can.

What Actually Makes Research Writing Work

The difference between research communication that works and research communication that doesn't isn't grammatical correctness. It's not efficiency. It's not volume.

It's specificity. It's surprise. It's the human detail that makes a reader think: I didn't know that. Or: I want to know more. Or: I should tell someone about this.

These qualities can't be automated, because they require judgement. They require a writer who can look at a body of research, identify the most interesting thread, and pull it into a narrative that rewards attention. They require someone who knows how to ask a researcher: "What surprised you? What went wrong? What would you tell your neighbour about this work?" — and who recognises a good answer when they hear one.

AI is a tool. It can help with research, summarisation, and first drafts. But it cannot replace the core skill of research communication: the ability to find and tell a story worth reading. That requires a human being. Ideally, one who's trained to do it.

Use AI. Don't Depend on It.

There's a sensible middle ground. Use AI to generate a first draft of a routine announcement. Use it to suggest headlines. Use it to check for inconsistencies. But don't let it write the things that matter — the features, the profiles, the impact stories, the magazine articles that represent your institution to the world.

Those deserve a human writer. They deserve someone who can interview a researcher, identify the story, and tell it in a voice that sounds like your university — not like every university.

At Stokel Publishing, that's exactly what we do. We interview your researchers. We find the human story in the data. We write research magazines, features, and case studies that people actually read — because they were written by a journalist, not generated by a model trained on the average of everything.

If your research communications sound like they were written by AI — even if they weren't — get in touch. There's a better way to tell your stories.