A practical, slightly snarky guide to using GenAI in your writing process (hint: don’t let it generate anything)
The Steveification of Brand Voice – a dystopic future in which everyone starts using ChatGPT a little bit too much.
Generative AI tools like ChatGPT are just that – brilliant new tools. Artificial intelligence isn’t artificial or intelligent. It’s trained on text written by real humans, with all their bland clumsiness and inherent bias.
ChatGPT learned to write using text from Wikipedia. This vast data set is actually written by an extremely limited pool of people. One man, Steven Pruitt, has made over six million edits to English Wikipedia, touching at least one-third of all its articles. If you’re letting ChatGPT write for you, you’re letting the most generic version of Steve get his hands on your work. That makes it not particularly good at generating, either.
As more people and brands swamp the online world with AI beigeness, the AI tools will begin learning from AI-generated content. Boring bots, learning from boring bots, learning from boring bots, and so on, until we all die of boredom.
As the internet slowly eats itself, will we begin to see one brand sounding much like the next, all of them a little bit too Steve for comfort? In reality, probably not, because those major brands will take this as an opportunity to further differentiate, to sidestep the mundanity of the GenAI, and lean harder into human diversity, weirdness and humour. And I think I speak for everyone when I say thank F for that.
GenAI being generic in action: my snarky and conversational phrase, “slowly eats itself”, didn’t fly with Grammarly’s AI assistant. It suggested “continues to evolve”. No, AI. Just stop trying.
The same goes for the rest of our businesses and brands. While everyone else spirals into sameness, we (read: you) can be the ones writing in flashing technicolour.
That doesn’t mean you should avoid using ChatGPT and go back to quill and ink. It just means figuring out how and where the tools will most benefit you and your writing.
Surprise, surprise, I have some thoughts on that.
It might be right there in the name, but using generative AI to generate copy for you can end up sending you down a pit of time-wasting and despair (said from experience).
Here’s why:
Gen AI can smash out 3000 words, no problem! WOW! But then you know who has to read, assess and edit that 3000 words? You.
ChatGPT’s goal is to deliver statistically probable answers, not the most accurate ones. I’ve seen it invent an entire business, complete with phone number and Google Map. The business didn’t exist – the street it was on didn’t even exist. Even when it’s not hallucinating reality, AI skims along the surface of a topic, missing critical nuance. It’s why GenAI seems fine until it’s writing in a subject that you’re especially expert in – then you notice the gaps.
You’ll at least need to develop a strategy for each piece before putting it into a GenAI tool. And by the time you’ve done all that thinking and planning, you might as well draft it yourself. It’ll probably be faster in the end, and you’ll also be embedding your understanding of the material and gaining broader context on the topic. Let ChatGPT write it for you, and you miss all that. Will you know how to answer the client’s questions? Will you be able to speak to your presentation?
We’ve already covered this, yes, but please. Sound like you, not Steve.
The hardest part about writing is getting a perspective about what’s on the page. That’s where ChatGPT and its mates can be useful. We all have blind spots, special interests and not-so-hidden agendas that can get in the way of how and what we write.
The robots can play devil’s advocate for you, challenging you on your thinking and what you’ve written. You’ll find maybe 30% of the suggestions are actually useful, and it will take time to wade through them, so don’t think about GenAI as a time-saver but as a quality improver.
When you’re planning each piece of writing, you can use GenAI to find holes in your strategic thinking. For example:
Remind ChatGPT of the goal, audience persona and brand persona of your piece. Tell the bot that it’s an expert editor and then give it your final draft. You want it to highlight:
Each time, specifically ask for a bullet list, not a rewritten draft – GenAI loves to rewrite stuff without being asked.
Another great example of GenAI’s general uselessness at genning: it suggested this headline for my conclusion,
“Write like a human. Use AI like a tool.”
With my millennial grasp of slang, that sounds like you should use AI while being an absolute dickhead.
Why let the bots have all the fun? They don’t even know what fun is. So think about the parts of writing that are least interesting and least rely on your human brain. Often, AI models are ideally suited to that.
Once you’ve decided on the key messages and overarching angle, GenAI can throw it together in a logical flow. However, it could very well emphasise the wrong things and link ideas in misleading ways, so this is really just a good place to start.
One of the few places that GenAI will save you time is when you’re writing from long or complex source material. A classic example of this is when you’re writing from interview notes or a transcript. Instead of scrolling back to find that really good quote, just ask the bot to surface it. I recently had to draw info from a bunch of reports about house prices in different areas of the country. Collating and combining the info would have been very boring, and I would have been very bad at it (accuracy with numbers: not my strong suit). I asked ChatGPT to do it, and it performed…OK. My proofreader still found a bunch of little errors.
If you’re making claims, it’s really important to have reliable, primary sources to back them up. That means original research, published on peer-reviewed sites or by reputable businesses. Finding these can take some time, so it can be easier to search using natural language models. Caveat: it’s not very good at this yet, so always, always check that the link supplied is usable.
It’s not bulletproof, but GenAI can be a good first pass to find typos and other errors. It’s particularly helpful in surfacing inconsistencies that other digital writing assistants seem to miss – do you have full stops on some bullets but not others? Are you spelling it ‘advisor’ or ‘adviser’? How are you capitalising GenAI?
Again, and I can’t stress this enough, specifically ask it to supply changes, not a proofed version. Lots of its changes will be dumb.
Think about it like this: GenAI is your error-prone research assistant, proofing intern or slightly clueless colleague who sometimes surprises you. All useful, if managed carefully.
GenAI is also Steve. There’s nothing wrong with being Steve if you are, in fact, Steve.
But I’m guessing neither you nor your brand is a white American man who has dedicated his 40+ years to updating Wikipedia.
Keep GenAI where it belongs – in the back seat. Let your ridiculous, creative, unpredictable, surprising, nuanced, interesting and educated human brain take the wheel.
The lead copywriter and creative director at Words for Breakfast. She spends much of her time working with the copy writing team across a variety of projects.
