There Are No “Special Words” That Give Away AI-Generated Content
AI writing has the same faults as human writing
“Of course, their content is clearly written by ChatGPT.”
If this sentiment sounds familiar, you’ve probably spent a fair bit of time in marketing meetings over the last year.
At some point during their 15th hour scrolling through hot takes on LinkedIn about how gamechangingly enormous ChatGPT’s impact on marketing was going to be, an awful lot of people decided that the new hallmark of sophistication was to assume every bad piece of writing was generated by AI.
This despite the fact the general tenor of most commercial social posts, SEO filler and spam email had changed not one iota. (It’s also with mentioning that both the film Crash and TV show Derek were released before OpenAI existed.)
And now we’re seeing articles coaching readers to spot AI writing through key phrases. As if LLMs weren’t literally trained to emulate the way humans create content.
The truth is AI writing is because most human writing is bad. It tends toward hyperbole because that’s the internet’s whole thing. (Notice the hyperbole there? As if hyperbole were all anyone ever does on the internet.) And it uses embarrassing terminology, egregious cliches and dull empty phrases because, well, that’s what all of us do too.
The logic, as far as I can see, is this:
- Bad writing is easier to produce than good writing, therefore more bad writing is produced.
- That means AI is trained on more bad writing than good writing.
- And that means AI (rightly) predicts that the next word the average human would choose in any given sequence will be a dumpster fire decision guaranteed to make any half-sentient being either giggle or retch.
So to assume bad writing must be the product of AI is actually quite sweet, a sort of naivety about human achievement masquerading as tech savvy.
We are biased in favour of humans, and that’s a good thing.
But this can shift from healthy anthropocentrism into hilarious delusion at the drop of a hat.
Captcha systems are my favourite example. Only a human would assert its cerebral dominance based on the ability to figure out which bits of a picture are a bus.
(Hint: if you actually want to distinguish between the average human and AI, ask them a basic math problem – and only admit those that fail.)
Our human advantage against AI when it comes to writing doesn’t stem from our superior skills or knowledge – it comes from the fact we are the audience.
So here’s my hot take: the fact people are so eager to claim content is written by AI is a valuable signal. It shows that the risk moving forward is not that we will be constantly tricked by devious AI-generated content, but that we will lose trust that any message is the product of a real, thinking and feeling human.
This presents an opportunity for business content. It will dramatically increase the value of factors that are still the exclusive domain of human content creators. These include:
- Genuine excellence in prose and design
- Humour and irrational ideas
- Utterly unnecessary levels of precision and attention to detail
- And thought leadership born of actual first-hand experience
I also believe this is the most valuable use-case for AI in commercial content production: arriving at the “differentiation” part of the process faster.
ChatGPT can show you all the most obvious, cliched solutions to the problem you’re solving; serve up all the most generic topics, insights and advice; and enable faster market research to figure out how you can ensure you don’t regurgitate the same old pap.
It can aggregate all those mindless “trends” articles you spend the festive season yawning over, and show you what the average person in your industry thinks – so you can actually make a reasonable attempt at “leading” then somewhere more interesting.
And it can remind writers that their most valuable skill isn’t producing lots of words, or meeting that deadline – it’s reading, appreciating and obsessively pursuing real quality content.
