Generative AI is here to stay, of that there is no doubt. What has been subject to fierce debate among copywriting circles is the extent to which we should use AI and how good it is, and whether we should be concerned about the likes of Gemini, ChatGPT and Claude taking work away from us.
All valid questions and subjects I will always enjoy talking about among peers. But the reality is that we, as a copywriting trade, have to deal with that first statement – AI isn’t going away, so how can we make the most of it?
Our clients are a mix of agencies, businesses and publishers who are all facing up to generative AI and answering similar questions.
One clear area of value that copywriters can exploit is in creating humanised content that stands apart from copy produced by generative AI. Similarly, editing AI generated copy is another useful skill that we are seeing more demand for, as is the ability to brief an AI engine properly (a topic for another discussion).
In all of these scenarios, being able to tell whether something has been AI generated, and doing so with conviction, is crucial.
I mention conviction because what follows comes with an enormous disclaimer attached to it: technically, unless you are in the room at the same time, there is no way for anyone to categorically prove that something they have been sent has been produced by generative AI.
But there are a lot of signals which point to content or passages of writing being AI-generated. Below is a list we have compiled based on our own experience of handling copy of all shapes and sizes, both in terms of material sent to us and using engines ourselves to help us with our work.
Again, I cannot present these as hard truths, and some of them are things I naturally want to do in my own writing.
8 signs copy has been AI-generated
So, how can you gauge that something has been written by generative AI?
Some of the telltale signs are purely visual and very easy to identify, while others require a closer examination of specific language and writing patterns.
Let’s take a look, starting with the obvious visual clues.
1: Use of dashes
Personally, I have nothing against dashes and used to use them quite a bit in my own writing. One of the biggest adaptations I’ve had to make is using them sparingly, as AI engines churn them out obsessively.
There are three signals relating to dashes. The first and most obvious sign something is written by AI is if it contains enormous em dashes — these ugly things.
The second tell relates to volume. Different copywriters will have different takes on what is a natural volume of dashes to use. For me, if dashes appear multiple times in most paragraphs of a piece of content, it is a sign of AI generation.
Another sign is the way in which dashes are used. Often I see generative AI produce content full of constructs that use dashes to break up sentences – like this to explain or elaborate on something – when commas can do exactly the same job.
2: Massive paragraphs
A lot of content we work on is thought leadership, be it ghost written bylines for trade publications or blogs for company websites. When I see this type of writing laden with very large paragraphs, often more than seven or eight lines deep, it immediately causes me to question whether AI might have written it.
3: Repetitive paragraph structures
In a similar vein, when AI gets on a roll and is asked to produce whole articles, it often defaults to very predictable paragraph structures.
Specifically, paragraphs which start with a short and snappy opening sentence, followed by much longer sentences. There is nothing wrong with this, and the use of short sentences to grab attention is a well-trodden copywriting technique. However, when almost every paragraph reads the same way, it is a flag.
4: Loads of short sentences one after the other
When a list of things is presented in lots of short, individual sentences, that is another sign that something might be written by generative AI. This is especially the case if such a construct is used more than once in a piece of thought leader content like a blog or bylined article.
5: Lists of three
Sticking with lists, another telltale sign of AI involvement is when a large number of sentences contain a list of three. Again, there is nothing wrong with this construct – once again, it is a matter of volume.
In addition, I quite often see sentences with two lists of three within it, which feels unnatural for a human writer to produce.
6: ‘Not just X, but Y’
There are some other sentence constructs which, when overused, are signs that AI has written something.
The one that irks me the most goes along the lines of: It’s not just about this one thing, but something much bigger. Similarly: We’re not just talking about this. Instead, it is about this.
Another flag when looking at thought leader or educational pieces is when multiple sentences in an article give advice in the same way. In particular, I am referring to sentences that start with ‘For’. Example: For organisations seeking to do this, the best way to approach it is this way.
7: Overuse of specific words and terms
Going into even more granular detail, there are particular words and phrasings that tend to pop up multiple times in a piece of copy when produced by generative AI.
Some of my favourites include:
- ‘actually’
- ‘fundamental shift’
- ‘approach’
- ‘compounds’
- ‘comprehensive’
There are many other buzzwords, cliches and flowery metaphors to watch out for.
8: Broken hyperlinks
AI can also make mistakes with formatting and presentation, one of the most obvious being the insertion of hyperlinks that don’t work. If you are reading a piece which contains broken links, it is unlikely a human copywriter will have written those passages where the links are contained.
Should copywriters use AI detectors?
While often used interchangeably, copy editing and proofreading are distinct processes with different focuses. Everything I have mentioned stems from experience and instinct, and this will build up over time as you become more familiar with AI and make greater use of it.
There are also several online detection tools. These are free and simple to use, and will highlight areas of content that it deems to be written by AI. ZeroGPT, Grammarly and Copyleaks are among them.
I am not an advocate of these tools for a few reasons. First, they are extremely inconsistent – put a piece of writing through five different tools, and they will all tell you something different (this blog scores from 0% up to 56% AI generated). Second, they are temperamental and can react drastically to very subtle word changes. Third, it is essentially AI telling you whether it thinks AI wrote something. And finally, I find myself writing more like AI to get content through a detector – in particular, they seem to like lots of dashes.
However, in spite of these reservations, copywriters need to accept that some clients will use them to verify what they are publishing or passing over to their clients is human content. Rightly or wrongly, they are changing how we write.
How can copywriters use AI effectively?
We have been experimenting with generative AI engines for quite some time now, and there is still a long way to go before they master the art of professional humanised copywriting.
Where AI can be particularly useful is behind the scenes, be it helping with initial ideation and planning, digesting large amounts of source material, or taking notes during calls and summarising transcripts. No doubt it is helping copywriters become more productive.
We’ve written about the pros and cons of using generative AI, and our position now is largely unchanged from when that was published in March 2025.
Do you want to find out more about our copywriting and editing services? Contact us to discuss how we can support your next project.
