Analyzing Trends in AI-Powered Mystery Games
The evolving role of generative AI in mystery game development
Recently, a subscriber suggested I write an article on the topic of generative AI in mystery games. I happen to be pretty knowledgeable on AI already, and I thought it was an interesting topic that I hadn’t really covered in depth, so that’s what we’ll talk about today!
Also, through sheer coincidence, a significant amount of AI news has come out over the past 2 weeks anyway. So I’m glad the timing worked out!
General AI trends in game development
The last time I wrote on this topic was my analysis of Steam’s generative AI policy. That was in January 2024 — a whole year and a half ago!
So, how much have things changed?
This month, Ichiro Lambe of Totally Human Media shared some interesting statistics about generative AI usage on Steam: 7,818 titles on Steam disclose its use, which is about 7% of all Steam titles and just under 20% of all games released in 2025.
As the article pointed out, these are only games that actively disclose the truth, and so there are likely even more games using generative AI that aren’t disclosing it. The article goes a bit deeper into what people are using GenAI for, how they are disclosing it, and which games that use it are getting popular. (You will see in a bit that I’m going to do the same analysis for mystery games.)
And speaking of Steam, Gabe Newell himself recently weighed in on generative AI in a new interview:
"I think the more you understand what underlies these current tools the more effective you are at taking advantage of them, but I think we'll be in this funny situation where people who don't know how to program who use AI to scaffold their programming abilities will become more effective developers of value than people who've been programming, y'know, for a decade."
As someone who has been programming for over a decade, I agree.
I discussed it a little bit in this article on prototyping. But with the help of AI, especially in programming, I’m able to get a lot more done in a lot less time. You can apply that accelerated rate of learning to just about anything, though you also need to be careful about learning the wrong things, picking up bad habits, or being totally misled or deceived by the AI.
Recently I went back and looked at some of my old projects from years ago that I gave up on. What I needed most back then was a coding mentor, because I was learning as I was doing and most of the time didn’t really know if I was on the right track or not.
So now, when I feed my code and ideas into ChatGPT and Grok, I can instantly get extremely useful feedback on how to improve those old projects!
But you need to understand that AI is not a magic solution. It takes a certain level of domain expertise, combined with prompting expertise, to leverage AI effectively. That means the people most likely to benefit from AI are also the people who are also the least likely to need it.
So while I think it’s true that people who leverage AI can outpace a traditional expert, I think an expert who learns how to leverage AI is even more powerful.
When generative AI first started to become popular in 2022 and 2023, learning how to use such a cutting-edge tool seemed like it could be an advantage. But these days, I’d argue that not knowing how to use AI actually puts you at a severe disadvantage.
But today’s topic isn’t what I think about AI, so I’ll have to stop there. What I want to discuss right now is how AI is actually being applied to mystery games.
Why use AI for mystery games?
Other than the fact that it’s trendy, why even bother?
Rapid Prototyping: The best example is the game The Roottrees are Dead. The game was quickly built using AI images as placeholders and released for free. Then, after the game became popular, the images were replaced with handmade art for a paid release on Steam.
Efficiency & Immersion: AI can generate boilerplate code, or it can generate swaths of generic images for background decorations, or text for players sift through while searching for the truth. Avoid wasting time on low-ROI assets and spend your real human effort on the high-impact ones.
Dynamic Story: You can give players a new level of agency by allowing them to shape the direction of the story.
Interrogation: The most unique aspect to mystery games, and the most common AI gameplay mechanic among them. Dynamic conversations are a natural fit for generative AI. Some games go beyond pure text and utilize the Whisper API for voice recognition and ElevenLabs for dynamic voices too. So players can speak to NPCs, and they speak right back.
Answer-Checking: Quite a few mystery games used this as a gameplay mechanic too. Players submit their answers to an LLM that verifies or rejects their solution, possibly with a helpful hint to point them in the right direction.
Multiplayer Simulation: While only used in a few obscure games so far, it's an interesting application. In social deduction or other dialogue-heavy multiplayer experiences, AI can be used to act as the other players, allowing you to play alone. Whether it's as fun as playing with real people is up for debate.
A note about my data collection & analysis
In order to find out how AI is being used in mystery games, I had no choice but to use SteamDB. The Steam API itself doesn’t have any way to pull games by tag, nor does it have a way to pull games by AI disclosure. This made things way more difficult than it needed to be, mainly because mystery games are tagged very poorly on Steam.
SteamDB conveniently has a filter to search by AI disclosure, which you can combine with tags, and so it was easy enough to get the numbers for each. But I wanted to go deeper and actually analyze the specifics of each disclosure. That meant I had to go page by page and copy each game URL, then visit each page and copy the disclosure text into a spreadsheet for each tag.
Speaking of AI, I had a lot of help from Grok 4 to code scripts that helped me collect and analyze the data. Even with the help of AI, all of this data collection and analysis still took me several full days of work — but that's a lot better than the weeks it would have taken me otherwise.
The tags I analyzed were “Detective”, “Investigation”, and “Mystery.” There are likely more games that the average person would consider a mystery game that don't explicitly use those tags (and it's not always the case that the developer can choose which tags are even applied to their own game).
More specifically, when I talk about the games from those three tags, here are the exact ways I categorized the data:
"Detective" games are games on Steam with tag "Detective" and may include the other two tags.
"Investigation" games are any games not already listed in the "Detective" pile that may also contain the "Mystery" tag.
"Mystery" games are any games not already listed in the other two piles.
I structured it this way because "Detective" games tend to more accurately describe the kinds of games we typically classify as mystery games. In fact, when looking at games tagged "Mystery", I found that only 30% actually met my definition of mystery game. Many of them were either horror games, jigsaw puzzles, or completely unrelated. But that’s a discussion for a separate article. Just take the “Mystery” tag results with a major grain of salt. Even when we remove two thirds of the games, the percentages don’t actually change.
Statistics for AI in mystery games
Below you can see a table of general stats:
Let’s break down what this means:
In the whole year of 2024, there were 2,192 games released with AI. By July of 2025, there are already 2,491 games released with AI (note: this has now increased to 2600+ since I first made that table). If we extrapolate these numbers, the number of games with AI in 2025 are likely to be double that of last year.
Similarly, for each tag, there is a significantly larger percentage of AI games in 2025 than 2024, growing at rates ranging from 33% to 54% each.
Detective games have the highest total amount of AI games regardless of the year (15.73%) due to LLM-powered chat mechanics for interrogations. Additionally, nearly 20% of Detective games released in 2025 use generative AI in some form.
Now this next table might be the most interesting one:
Here we’re looking at the actual types of content created using generative AI for each tag. I had to manually look at each disclosure text for all games to categorize it properly, so this took a long time! Actually, Grok helped out toward the end, and was surprisingly accurate when I double-checked.
Some key insights from this one:
In-game art is the most prevalent use of AI, averaging around 60%. Very interestingly, this lines up exactly with what Ichiro Lambe said regarding the general trends across all games on Steam (“Visual Asset Generation: This is in about 60% of disclosures.”)!
Chat gameplay shows a statistically significant difference in Detective games (12.86%) vs. the other tags. This suggests an association between AI-driven chat mechanics and the Detective tag, which makes sense. But it highlights the relevance AI has to the mystery game genre, compared to other game genres.
Promo art is significantly higher for the Mystery tag, although keep in mind most of those I wouldn’t really consider mystery games. Perhaps this says something about games that rely on a mysterious premise/atmosphere over gameplay.
The disclosures marked “Unknown” were unclear about how specifically they used AI, and “None” had the disclosure despite not using it whatsoever.
Overall, I’m not too surprised by those results. Maybe you can find some interesting insights, too.
But I wanted to go a step deeper. So I analyzed the justifications people were giving for using AI in the disclosure. Here’s that table:
As you can see, the vast majority didn’t give justifications at all. The most common excuse across all tags was that the generated content was edited manually by a human. Let me define these in detail so you can make sense of them yourself:
Not all AI = Emphasis that other parts of the game did not use AI.
Licensed = Emphasis that the AI usage was properly licensed.
Postprocess = Human-made content was processed by generative AI.
Staff = Emphasis on being a one-person studio or limited staff.
Budget = Emphasis on financial savings from using AI.
Support Artists = Emphasis that despite the AI usage, the developer supports the creativity of human artists.
Assisted = Emphasis that AI “assisted” the developer in some way.
Immersion = Emphasis that AI was used to enhance immersion.
Efficiency = Emphasis that AI created a more efficient process.
Replace = Emphasis that the AI content was just a placeholder.
Transparency = Emphasis that the developer valued being transparent about AI usage.
There were some other unique justifications as well. Some cited its use as “experimental” or as a learning experience. Others said they had discussed it with their community (fans) and they had agreed it was okay. Some intentionally used AI to generate an “uncanny” visual style.
I’m not sure if there’s a real explanation for any of those percentages, but it’s interesting to see that the “Mystery” tagged games had significantly fewer justifications. Again, I attribute that to the “Mystery” tag being applied badly by games that would generally not be considered mystery games.
It’s also a bit hard to tell, but if you add up the percentages you’ll see that at least 30% of all AI-powered mystery games wrote some kind of justification for their usage of AI.
The most unique justification I came across was also quite sad — a voice actor had passed away during development, but AI was going to be used to continue voicing their lines (presumably with the actor’s permission).
Player reception to AI mystery games
This part was quite tricky to figure out, since AI is such a touchy topic and also because there were just so many games to try to analyze. But it seems like these games largely fit into 3 categories:
The game obviously looks AI-generated or was lazily put together using nothing but AI-generated assets, and was either largely ignored or received not-so-good reviews.
The game is largely made in the traditional sense, but includes a very small amount of AI-generated content that gets overlooked, or the AI-generated content is used to enhance the game via optional voice-acting and translations. This type of game essentially succeeds despite its use of AI.
The game uses AI as a core part of its gameplay, such that the game couldn’t be made if generative AI didn’t exist. This type of game succeeds specifically because it uses AI to create an experience unlike anything else.
For the first type, it was very common for me to see many games with AI-generated art that simply never had anyone play them. Others did a better job and managed to get people to at least play them, but the AI parts were always called out in a negative way. But the AI parts were not the only bad parts of each game — it was clear these games always had design flaws or a general lack of polish that just makes them look bad.
When you look at the reviews for the hit game The Roottrees are Dead, some of them mention the AI images as a problem, despite those images not being in the Steam version at all. But out of nearly 3,000 reviews since January, only 3% are negative. This game single-handedly proves two things: that even when you choose to replace AI images, people will still give you trouble over it — and despite those people, the majority won’t care anyway (as long as it’s actually a good game).
Similarly, The Operator currently sits at 6,000+ reviews and 92% positive. The game discloses it used AI to generate exactly one image. Nobody cares.
Then when you look at games that embrace AI for their gameplay mechanics, you can actually see pretty good success.
Doki Doki AI Interrogation (400+ reviews, 82% positive) is a minimalistic experience focusing on a single interrogation of a very simple scenario. Uncover the Smoking Gun (357 reviews, 95% positive) has you chatting with various robots as you go about searching crime scenes and linking evidence to form theories. And recently-released Bureau of Contracts (378 reviews, 78% positive) is a multiplayer ghost-hunting game released this year, with the twist that you can chat with the ghosts using AI!
I know those aren’t majorly huge — all under 1,000 reviews — but those are all a solid increase from Vaudeville’s mixed reviews and Portopia’s very negative reviews in 2023, demonstrating that the mere presence of AI isn’t the problem. As the tools get better and designers get more familiar with them, we can make more enjoyable experiences.
In summary, if your game is well-designed, then its fun will overshadow any negative connotations the AI parts might give it. But if you’re just using AI to cut corners and release a really unpolished experience, then the reviews won’t be too kind to you.
Generative AI in the courtroom
Despite all of that, I think it’s important to at least mention a word of legal caution about using AI generation in your games.
The legal case for generative AI is so complex that it warrants having its own article. Many lawsuits are concurrently ongoing, and certain rulings on its legality just won’t be clear until at least 2026 or even 2027 as the hearings continue. If you’re interested in these developments, you can follow the court cases on various legal blogs which do a much better job of explaining things than I can.
You can also read this document published by the Congressional Research Service in June 2025 (which includes many links to external sources for further reading) to get yourself relatively up to speed. And the White House released its AI Action Plan just a few days ago, signaling that AI will be a major factor over the next few years.
I’m no lawyer, and you shouldn’t take any of this article as legal advice. But the one thing you absolutely need to make sure of — as the Steam GenAI disclosure requirements suggest — is to not use AI to generate material that violates the law.
This can be dangerous if, for example, you innocently type in a prompt and get back a result that you believe is original, but is actually a depiction or excerpt of copyrighted material.
You might even think ChatGPT helped come up with an original idea, until you do some research and see that it just regurgitated the idea from somewhere else.
Instead, you should have AI teach you the skills needed to create original content on your own. Ask it for feedback and converse with it like a creative partner, but always contribute your own ideas, and make all of the changes yourself.
Final thoughts
I know this was a long one today, but I hope this was at least a bit enlightening.
There’s so much more to investigate about AI in mystery games, but I think this is a solid start. Many games with unique AI mechanics are still in development and won’t be released for a while. So it’s almost certainly worth revisiting in a couple years, especially when the legal precedent is more set in stone.
What do you think about the potential for AI in mystery games? Let us know in the comments!
Thanks for reading!
If you want to learn how to make your own mystery game…
Or if you want to discover the best mystery games before anyone else…
Join our 500+ member Mystery Gamedev community today!
And please consider sharing our newsletter with a friend — we appreciate it!