Content designers need to understand generative AI writing tools, and be aware of the pitfalls, but shouldn't miss out on their great potential.
Mention generative AI tools like ChatGPT and most content designers will shudder.
That’s not because they’re worried about robots taking over the world, or for their job security.
It’s because they’re thinking of the potentially disastrous effects of AI generated content on users without the necessary quality control.
And the clean-up operation they’ll need to undertake to fix it.
Unfortunately, these negative experiences and perceptions are putting content designers off using AI when it could be a valuable tool.
The trouble with AI generated content is that too many people who don’t understand the importance of good content think that what ChatGPT generates is ‘good enough’ for publishing.
And to be honest, in some rare cases it might be – especially if you have a custom model trained on your own data and have well designed custom prompts.
But when it comes to user-centred design, ‘good enough’ is too low a bar.
It has to meet user needs and content that’s just ‘meh, okay’ won’t do that.
What’s the beef with AI generated content?
The AI content itself isn’t the real issue. It’s the lack of understanding on how to use it to get the best results that’s the challenge.
We’ve tested it. Extensively.
Even with detailed and effective prompts – and lots of back and forth – generative AI content needs plenty of editing to be fully accessible and inclusive, and truly meet user needs.
Why? Here’s a big old list of reasons.
1. It has trouble with house style
In its current state, ChatGPT can’t grasp GDS guidelines. Yes, even if you feed them to it in your prompt. It can get close, but not close enough. It’s like working with someone very keen who hasn’t quite read the manual.
Although the update announced last week might help with this: ChatGPT can now check current sources for information, rather than relying on information from before its training data cut-off date.
2. It gets stuck in a rut
It has a habit of repeating certain cliched words and phrases, over and over again.
In the content design practice at Sparck, we’ve got good at spotting AI generated content when it comes to us for review.
There are obvious ‘tells’.
3. ChatGPT waffles
Generative AI content is also often long-winded and uses overly complex language unless you carefully structure your prompts to avoid it. (Which most users don’t.)
In terms of actual meaning, AI will tend to give you very general or broad points, and just waffle around them.
It doesn’t have access to deeper context and detail because it’s not actually thinking, even if it sometimes gives that impression.
The recently announced ‘chain of density’ prompt model suggests one way to address this problem – but it’s almost as complex as coding!
4. It’s not human
It lacks personality and empathy because it’s not actually capable of feeling.
A key part of being a designer, before we ever put pen to paper, is putting ourselves in the shoes of users.
Our best work comes from a place of understanding.
5. American vs. British
Most generative language AI tools default to American-English, because that’s where they were built.
Which needs correcting if you’re creating content for the UK, which is something non-expert users might overlook.
6. It fibs and steals
AI sometimes lies and plagiarises.
Sometimes even the sources it cites are false.
But it almost always gives the illusion of correctness and rigour, which makes these fibs harder to spot.
7. Poisoned training data
And probably the most troubling problem is that AI has ingested all the biases and stereotypes in its training material.
When you want to create accessible and inclusive content, this can be a huge issue.
Proceed with caution
Because of all of the above I’d say, proceed with caution when working with AI generated content.
Use generative AI alongside your user-centred designers, but don’t think you can simply replace them with a snap of the fingers.
It can’t (yet) lead a workshop or effectively collaborate on a prototype.
How to use AI in content design
Recognising and understanding generative AI’s strengths and weakness is the most important thing.
For me it has just one real strength: speed.
Ask ChatGPT to write you the brief for a workshop, group user needs into categories, or summarise a lengthy document, and it will do it in literally seconds.
Because of this, it’s great for that content design ‘busy work’ that prevents us from spending more time on the creative thinking side of our jobs.
Generative AI can be a great tool for:
- Drafting rationale documents to explain content decisions and changes.
- Creating materials for workshop, pair writing and content crits, including email invites and summaries.
- Turning research findings into content principles.
- Spotting patterns in and organising research findings.
- Creating rough-and-ready code for prototypes to test layouts.
- Ideation (“Give me 10 ideas for…”) or creating a starting point to get ideas flowing.
As someone who is user focused, I’d always say that AI content needs editing before it’s released into the wild, regardless of where it’s going.
Whether it’s for a quick email or publishing on a government website, there’s a user at the other end that you need to consider.
However, using generative AI tools in the ways suggested in this article could save time and brain power for the juicy creative bits.
It could give you back the time and headspace to solve more complex problems and try out a wider range of ideas.
