Should you say AI was involved? Five questions before you decide

As you may have seen, I'm tinkering with a bit of game development on the side. The team (if you can call them that) is eight AI agents and me. I decide what we're aiming for and what's fun, I give feedback on tweaks, and they do the rest. Often they nail what I want. Sometimes not quite. That's life.

At some point the game might be good enough to sell. Then I have to ask myself: where should it say that AI was involved? And how much should it say? How much of the work was "mine"? What does it mean for the player?

This post is about how I might make that choice. I've found five good questions that could help me.

… and if they help me, they might work for you too?

Why now

On 25 September, Nielsen Norman Group published an article about when to say AI was involved.

Taylor Dykes boils the research down to five questions with an easy name: PACED. Policy, Audience, Context, Expectations and Degree.

One of the findings has been called "the disclosure paradox". People say they want to know whether AI was involved… and then they're let down by the content once they find out. Like when the magician reveals the trick, and some of the magic goes.

Sound familiar? It does to me.

I want to know when others use AI. I was probably a little wary of saying it myself, too. One of the studies noted exactly that: people would rather be told about other people's AI use than disclose their own. That makes sense.

I wonder what it's like here in Norway, with our culture of modesty and the Law of Jante, but that may have to be an article of its own.

The five questions

1. Is there a rule? (Policy)

First of all. Check with the platforms, with your employer, and/or with laws and rules.

The game platform Steam rewrote its rules in January. They now cover AI-generated content that players actually meet, in the game and on its store page. Tools behind the scenes are not included.

For me, that means text an agent has written ends up in the disclosure. How my agents have worked on the code doesn't.

2. Who is reading? (Audience)

People who like technology are often less critical of new technology and AI use. A meta-analysis of twelve experiments found that the drop in trust was smaller among people with positive attitudes to technology. People who take pride in their own writing were, on the other hand, stricter.

Personally, I've read countless threads on Reddit where people rip into games that have used generative AI badly. Jam packaging from two Norwegian brands has also had a lot of attention lately. A Reddit thread is full of detectives trying to solve the mystery of whether one of the labels was made with AI. The maker felt forced to write a page about how the labels were made. Have a look here: https://www.lerum.no/aktuelt/historia-bak-etikettane-til-lerum-topping. Then the question is how much of the process was manual, how much was automatic, and who cares about the nuances.

3. What kind of thing is it? (Context)

The more of the value lies in a person having made it, the more the disclosure matters. A birthday card is not the same as a timesheet reminder. A personal post like the one I'm writing now is closer to the birthday card.

…and yes, to be honest, I write this myself, based on a draft and outline made by AI, just so it's said. It's rubbish in Norwegian, so I spend a lot of time correcting it. It's better in English, so I'm not as fussy there.

4. Does the reader expect to be told? (Expectations)

Next up: would the reader feel fooled if they found out afterwards? If the answer is yes, say so first rather than last.

In games, the norm is becoming quite clear. The platform asks me, and players see the label on the store page. "AI". In a LinkedIn post, the norm is much murkier. How much of your feed is slop? The presentation your colleague suddenly finished before lunch? The email from an acquaintance?

5. How much did AI do? (Degree)

Fixing a paragraph and writing the whole text from scratch are two different things. In one study, ratings of the content dropped in line with how much of it people were told was AI. The bigger the share, the stronger the reason to say so. And the more important it is to be precise. Half AI? All of it? A couple of sentences?

If you've used AI, there are several ways to share it. For example: "The illustrations were made with AI." "The text was proofread with AI." "AI wrote the first draft, I rewrote it."

What I'm taking with me

One finding stands out most. In Schilke and Reimann's experiments, telling people yourself created more distrust than staying quiet.

Being exposed by someone else led to even more distrust than that.

So if it might come out that you've used AI, it's probably best that it comes from you. Control the narrative and be open about it. Then you avoid being caught out when that awkward colleague asks at the next all-hands how much of the report you used AI for.

A caveat

The research here is… mixed. A review of 47 studies on news found no consistent reaction to AI.

Many of the studies were done in a lab or online, with text, images and news.

Few are about games or design work. In one study with money on games, they found no distrust or suspicion at all.

Don't get me wrong; I'm not writing that disclosing AI is dangerous, or harmless. As always, it depends.

What I hope is that the five questions help you work out how to make your own choices. And how I make mine too.

Try it yourself

• Take something you've made this week.
• Ask the five questions. Take two minutes.
• Write one sentence about what AI did. As concrete as you can. "The first draft was written with AI" says more than "made with the help of AI".
• Hesitating? Think about who else might expose you (and how would you handle it? with style, or a nervous breakdown?)

Want to know more about how my agent team works? It's in Building a game with a team of AI agents. Where AI helps in design work, and where it doesn't, Tentativ has written about in a separate article.

Saying so is also a design choice. Might as well make it yourself.