Welcome to Attention Matters, the newsletter from Storythings which gives you practical insights and tools on how to tell better stories and grow your audiences’ attention.

Hello!

Last month I wrote about the three kinds of creative friction: good, bad and hidden. Since then, I’ve had some fantastic feedback about how people are using that framework to help their own creative workflows. But the one question I’ve been asked most is this: how can AI help us deal with creative friction?

The pitch from all the big tech companies is that AI will take away ALL the friction in our workflows. Mark Zuckerberg laid out this vision in a May 2025 interview with Ben Thompson’s Stratchery podcast:

This vision conjours up the image of CMOs as gamblers in Las Vegas, putting coins into the one-arm-bandit that is Meta’s AI marketing stack, and crossing their fingers that they get the result they want. But as we all know, the house is the only winner here.

I don’t think any of us want that kind of AI future. And I don’t think it’s going to happen - I think AI really interacts with each kind of friction very differently, and only one of them is a clear win. So here’s my thoughts on how AI affects Good, Bad and Hidden friction:

Good friction is the constraint inside the creative problem that makes us think harder: spending extra time with a contributor to get them relaxed enough to tell their story, changing a format to make it work for a shorter running time, or having to be creative with b-roll and motion graphics when you can’t get live video.

Generative AI seems to be good at making these constraints disappear. Can’t film in person? Generate the footage. Stuck on a story structure? Ask Claude to come up with the structure for you. But that good friction is what makes your work stand out, and if you skip it, you may skip the interesting idea waiting at the end.

In a recent paper from Google DeepMind and University of Maryland, researchers mapped the rarity of narrative structures in text generated by humans and examples of the leading LLMs. As you can see in the graphic, the human texts had much more diversity and rarity in them, whereas the LLMs create less innovation and variety. So using AI to remove creative friction can make your work duller and less distinctive.

What you can do instead:
As I said in my earlier post, your response to creative friction should be to protect it - so make sure you’re not using AI to automate creative tasks that might lead to new or innovative solutions.

Instead, use AI to add more constraints, not remove them. Give AI a role as the most challenging version of your audience or client, rather than as a creative tool. Ask it for ten reasons your format idea won’t work, or to rewrite your creative brief with half the budget. Use it to give you more reasons to do the hard thinking yourself. This is especially useful if it’s a long running format where you might feel comfortable in the workflow.

Bad friction is the organisational stuff around the work, and a lot of this is information friction. When you are in the thick of working on something, the actual status of a project is scattered across spreadsheets, email chains, chat threads and approval tools, and nobody has time to pull it together. This is the kind of thing that AI excels at.

At Storythings, we’re trialling using AI to replace some of our traditional SAAS project management tools, instead bringing together real time data from different sources into simpler, more customised dashboards. This can help you spot where bad friction is affecting a workflow, help you visualise and share this with your clients and stakeholders, and come up with solutions together.

What you can do instead:
Give an AI tool the timeline, feedback rounds and approval trail from your last project, and ask where you waited longest, who was chased most often and which decisions were made more than once. That’s the friction-mapping exercise from my first post, done in an afternoon instead of a workshop. Then use AI to come up with suggestions for how to clear those bottlenecks.

Two important things to remember when you’re doing this. Automating a bad process just gives you a faster bad process, so if you keep getting friction in the same places, you need to dive deeper. And AI can spot where the friction is, but only your teams can tell you why the friction is there in the first place. So use AI as a way to improve the conversations between your teams, not to replace them.

In my original definition, hidden friction is a decision that creates work for someone else without the decision-maker realising. Somebody in the chain still knows about it, even if they’re quietly absorbing the cost.

AI creates a more troubling version: friction that no one can see, because it’s inside the tool.

I saw a great illustration of this in Careful Industries’ report on the risks of AI notetakers. These are tools that we’re probably all using right now, and they seem to be great - nobody has to take on the role of notekeeper, and the AI not only transcribes the conversation, but also provides really useful summaries.

But we might be missing something. In her report, Rachel Coldicutt explains that transcription errors can be hard to detect, and that tools can misread people with strong accents, people speaking English as a second language and people with speech and language disabilities. She gives the example of her own experiences of using AI in meetings, in which “oppressive” became “impressive”.

Now imagine how that could affect contributions in a creative team. Whose idea got summarised away? Whose contribution was garbled, and so ignored? What assumptions was the LLM making as it summarised the conversation to pick out the actions or insights? Did it miss any because of biases or blind spots in its model training?

What you can do instead:
Make sure you’re always looking for hidden friction in your tools, not just your people. Check a sample of AI summaries against the original recordings. Ask contributors how they think their contributions and ideas were represented. Agree clear rules and consent for AI notetakers so anyone who has reservations is heard and can challenge their use. And most of all, take Coldicutt’s simplest advice: have fewer, better meetings.

So here’s the AI version of the rule from last time:

Bad friction: use AI to remove it.
Good friction: protect it from AI.
Hidden friction: expose what AI is hiding.

Next time you add AI to your workflow, ask: which friction is this removing, and who is now paying for it?

I’d love to know your experiences with using AI, so if you have any case studies you’d like me to share in the newsletter, hit reply and let me know!

Thanks for reading!

Matt