All insights

AI can write. But can it read the room?

AI can support research, synthesis, ideation and editing, but the most effective approach is to use it as a thinking partner while keeping first-hand experience, point of view and personality at the heart of the work.

By Julia Traskas

I use AI pretty much every working day. That’s quite a change from a few years ago, when conversations about AI seemed to involve either breathless predictions about the future of work or reassurance that the robots weren’t coming for our jobs just yet. Now, for many of us, it’s simply part of the toolkit.

And having spent a lot of time working with it, I’ve become both more impressed by what it can do and more aware of what it can’t.

For someone who works with words, ideas and information, AI can be extraordinarily useful. It can research a subject, interrogate an argument, analyse documents, identify gaps and turn an unruly collection of thoughts into something resembling a structure.

It can also save me from some spectacularly tedious jobs. But there’s an important distinction between using AI to help you think and asking it to do the thinking for you.

The technology has evolved. So has the way I use it.

My early experiments with generative AI were fairly transactional. Write this, summarise that, give me ten ideas for something else. Sometimes the results were impressive, other times they were fluent, polished and somehow not quite right.

The biggest change hasn’t simply been that the technology has become more capable. I’ve become better at working with it.

Instead of treating AI as a vending machine where you insert a prompt and hope something useful drops out, I use it as part of a process. I give it context, challenge its answers, ask it to research, explain why something doesn’t work and use it to explore thoughts I haven’t quite worked out yet.

Some of the biggest gains are in research and synthesis. Jobs that might once have taken hours can be tackled much more quickly. It’s useful for ideation too, although I’m less interested in asking for 20 fully formed ideas than in using it to provoke one of my own.

Then there are the small efficiencies that add up: checking for repetition, stress-testing a headline, cutting 200 words without losing the argument or spotting where the logic goes a bit wonky.

None of this removes the need for a writer. It gives us more time for the parts of the job that depend on something harder to replicate: perspective, empathy, instinct and an understanding of what sits behind the words.

The part that still needs a human

AI can produce a perfectly competent article remarkably quickly. That’s both its strength and its problem because “perfectly competent” isn’t what I’m aiming for.

Thought leadership needs thought. It needs a point of view, an understanding of the audience and some sense that the person behind it has actually experienced, observed or concluded something. AI can help articulate those things. It can’t manufacture first-hand experience and it can’t read the room.

A human writer picks up on things that are difficult to put into a prompt: the way somebody speaks, the words they would never use, the gap between how a business describes itself and how it actually feels. There’s a difference between a sentence that’s perfectly acceptable and one that sounds unmistakably like the person or organisation behind it.

We read meaning through more than words. We notice tone, imagery, emphasis, what’s said and what’s left unsaid. AI is getting better at interpreting all of this, but it’s still interpreting patterns rather than living the context. It also has an unnerving ability to make something sound plausible when it isn’t quite right. It can flatten nuance, confidently join dots that shouldn’t be joined and produce an answer authoritative enough that you’re tempted not to check it.

The problem is that this confidence can be misleading. AI has a tendency to blow smoke up the prompter’s backside, agreeing enthusiastically with a suggestion before quietly changing position when challenged. “Excellent idea, Julia” can become “yes, that will work much better”, which eventually becomes “you’re right, that isn’t accurate.”

That’s why interrogation matters. One of the less discussed skills of working with AI is knowing enough about the subject, the audience and the person behind the words to recognise when something is technically plausible but fundamentally off.

More content isn’t necessarily better content

Generative AI has made it incredibly easy to produce content at scale. Need six blogs rather than two? No problem. Landing pages for every conceivable search term? AI can churn them out before lunch.

But just because you can produce more content doesn’t mean you should.

Google’s guidance increasingly emphasises originality, first-hand experience and what it calls “non-commodity content”: something that offers more than a summary of information already available elsewhere.

That’s not to say “AI bad, human good”. Google acknowledges that generative AI can be useful for research and structuring original content. The issue isn’t whether AI touched the copy. It’s whether the finished content contributes anything useful, distinctive or original.

If everybody uses the same tools to summarise the same information from the same internet, we risk creating an enormous echo chamber of perfectly adequate content. And perfectly adequate doesn’t feel like a great brand strategy.

The risk goes beyond search

There’s another reason I wouldn’t hand content creation wholesale to AI. Your content is a big part of how people experience your organisation.

They learn what you care about, what you know, what you believe and, ideally, what it might be like to work with you. If you outsource your brand persona to a language model, something will inevitably get lost.

Think about a wedding, christening or significant birthday. You might enlist experts to deal with the big stuff, but the personal touches are yours. They’re what make it feel like your event rather than anybody else’s. Content works in much the same way.

Tone is part of your personality, but this isn’t simply about making AI copy “sound more human”. The bigger issue is authorship. Whose idea is this? What have they experienced that makes them believe it? What can they tell me that I couldn’t get by asking an AI tool the same question myself?

Those questions will only become more important as AI-generated content becomes more widespread.

So what is the intelligent way to use AI?

For me, it comes down to augmentation rather than abdication.

Use AI to get somewhere faster. Use it to research, interrogate, organise, challenge and refine. Let it do the jobs it can do better or faster than you can, but don’t outsource the part that makes the work yours.

Know what you want to say and who you’re saying it to. Question what comes back. Check the facts. Put your experience into the work and be prepared to reject the beautifully written paragraph that doesn’t actually say anything, or the perfectly reasonable one that simply doesn’t sound like you.

The irony is that the more capable AI becomes, the less interesting simply “using AI” becomes. When almost everybody has access to the technology, the advantage lies in knowing what to do with it and, perhaps more importantly, what not to let it do.

Let’s talk.

If you need experienced marketing support, or simply want to discuss where your current approach could be stronger, get in touch.

Discuss your marketing needs (opens in a new tab)