We can all agree that video games are an art form. Each game is a mix of visual flair, storytelling, engaging soundtracks, and technical wizardry. It’s this unlimited scope for creativity that makes games so hard to define when we’re researching what is out there and what we should be building as developers. The pre-defined labels we’re required to use when we publish on Steam can be very misleading.

Imagine that you want to create an exciting horror game with online co-op and stunning 3D graphics, and you want to get inspiration for mechanics that players love as well as find the games that already exist - your potential competition.

You might start by thinking of relevant tags and browsing the Steam catalogue or using online tools to check out reviews and sales. You might even have a game in mind that inspired you in the first place, but unfortunately, your search will always be shaped by your own gaming experiences or those of your friends and the research that you compile might feel incomplete. This might leave you wondering if you’ve really found all the potential game mechanics, story opportunities, or audience needs.

We’ve definitely felt this way during our market research and game development. We knew that there must be a better way to explore the Steam catalogue and that’s why we created the Steam Map.

The Steam Map

The Steam Map is a 3D structure of the entire Steam ecosystem built using our statistical models. It pulls in all the data from a game’s Steam page including tags, genres, labels, descriptions, and even images because we’re dealing with artwork here, not just software!

In the Steam Map, every data point is a unique video game and the distance between points (seen in the image below) shows how similar two games are. Where the game appears on the map depends on more than just genres and user defined tags, it’s also influenced by the game’s description and visual style from the gameplay images.

If the points are far apart, it means the games don’t have much in common in terms of visual style and gameplay description. On the other hand, points that are close together indicate games with similar features, art styles, and gameplay. The real strength of the map is its ability to measure the similarity between games. This lets us uncover patterns, identify the boundaries between genres, and most importantly, spot gaps in the market.

The Steam Map is the engine behind all of our tools you will find in the Game Oracle app:

• In Data Explorer, game title and text search is powered by searching for the neighbours of a game or finding all the games in our map that are similar to a particular search query
• Saturation Map is a 2 dimensional version of our map that we provide to users, where areas of the market are coloured according to the total number of games, providing a sense of where the market is saturated and where opportunities might exist
• The Game Gap market segments are created by clustering games together on the map using our in-house clustering algorithm, we then analyse those segments for opportunities

The point is, we love our map and it is behind everything we do here at Game Oracle. We believe that by understanding what makes game similar and correlating that information with player sentiment and commercial performance, we can unlock the key to what makes a game enjoyable for a particular audience.

Enjoying our insights? Subscribe to our free monthly newsletter!

The Problem With Tags

Just how tricky are tags when it comes to finding the games you’re interested in creating? Let’s revisit our earlier game development example.

Inspired by the success of games like Phasmophobia, you want to develop a 3D online co-op horror game. So, you head over to Steam and search for games with the tags: ‘3D’, ‘Horror’, and ‘Online Co-Op’. The search brings back 838 titles which you can visualise on our Steam Map.

The problem is immediately clear. In the main cluster (right-hand side of the image) we can see lots of games highlighted, but there are many games closer, and therefore similar to those highlighted that are not being captured by our selected tags. On the left-hand side of the image we see a separate splatter of games that are also tagged as ‘3D’, ‘Horror’, and ‘Online Co-Op’, and if we focus in on one of these ‘outliers’ we find a remarkable game called ‘King of Swords’.

While King of Swords does look awesome (we definitely want to try it out), would it be the first game you’d think of when brainstorming ideas for a Phasmophobia-inspired game? Probably not. It also wouldn’t be very helpful for gauging performance and revenue estimates for games similar to Phasmophobia. Yet, it has been tagged ‘3D’, ‘Horror’, and ‘Online Co-Op’.

In this example, relying solely on tags and genres to research the game I’m interested in creating has left me with two blind spots. First, if I don’t meticulously examine every game in my search, my estimates for the number of releases, pricing, reviews, and revenue might be skewed by irrelevant titles like ‘King of Swords’. Second, and perhaps more importantly, my understanding of what similar games are available for this audience misses out on horror games that are not using the tags from my search.

Now let us consider an alternative approach.

Games And Their Similar ‘Neighbours’

With the name of a game that inspires us and a map of the entire Steam catalogue at our fingertips, we can zero in on the ‘neighbourhood’ of games that closely resemble our inspirational title. The image below shows the 100 games most similar to Phasmophobia when we use the Steam Map in Data Explorer to find them.

Our Data Explorer tool lets you discover games similar to the title you searched for, ensuring that no stone is left unturned. With a cluster of games that are truly relevant to your project, you can start to get a clear picture of competitor pricing, the expected ratio of positive-to-negative reviews, and estimated revenue for the type of game you want to create.

More importantly, you can zoom in and ask, “What’s missing? What hasn’t anyone else thought of doing in this space?”

The Steam Map can reveal literal gaps in the gaming landscape, sparking inspiration for your next title.

Technical bits (for the nerds)

Although we have spoken about the map rather broadly here, behind the scenes there is actually a few different versions. We have a very dense deep map (like really deep...like Mines of Moria deep) that 'maps' games in high-dimensional space and then a bunch of shallow versions of this map that have been squeezed down into fewer dimensions, the smallest being a 2D map. 

Why all the squashing? Well it is very hard to visualise our very deep map...well actually it's impossible! We want to allow people to explore the marketplace visually in tools like Saturation Map and this means we need a 2D version. It is equally tough to find reliable market segments in our deep map due to a little bugbear called the "curse of dimensionality". So we also use a squashed version of the map to cluster similar games together into market segments (we use about 20 dimensions for this). We do this squashing using deep manifold learning, which is fun because it gives me an excuse to spend money on a beefy GPU.

 

Map Type	Where do we use this?	Advantages	Limitations
Deep	Data Explorer Steam Revenue Calculator Showing similar games on the details page of individual Steam titles	Extremely accurate search results	Impossible to visualise the topology (shape) of the market and the distance between games Search results can deteriorate when looking at very distant games in the high-dimensional space Difficult to create reliable clusters for market segments
Shallow	Saturation Map Game Gap	We can actually visualise and interact with the map Allows us to cluster games into market segments	The absolute similarity between pairs games can be lost in exchange for broader 'clusters' of similar games

 

What does this all really mean? Well, sometimes you will find what games are considered "similar" in Data Explorer differs from what is considered "similar" in Saturation Map and Game Gap, with the results from Data Explorer sometimes suggesting better matches. Unfortunately this is the cost we pay for nice visualisations and practical market segments. We are working hard behind the scenes to improve how exactly we do our squishing of the map and we will continue to ship new and improved maps.

How do you decide if your map is a 'good' representation of the Steam marketplace?

We don't just rely on vibes here at Game Oracle, instead we try and approach the problem with strong scientific rigour. We care about providing the most accurate insights for our customers. We have worked tirelessly to curate an extremely large dataset of similar and dissimilar games. We have created what is known as "triplet" data, where we select an anchor game, a game very similar to that anchor known as the 'positive', and a game that is dissimilar known as the 'negative'. Take "Overcooked" for example, a good positive example would be "PlateUp!" whereas a negative would be "Call of Duty: Modern Warfare". If our map is good, then "Overcooked"  should be very closed to "PlateUp!" but very far away from "Call of Duty: Modern Warfare".

We have used a combination of user-defined tags, common franchises, and manual labelling to create a dataset of over 100k pairs of anchor-positive-negative triplets. We use this data to improve our map, but we can also use this to assess how well the map is performing. The plot below shows how our shallow map performs across three different subsets of our triplet data (as of May 2025):

Wow that is a lot of graphs! Allow me to explain. The top three histograms show the distribution of distances between the anchor and the positive (blue) and the anchor and the negative (orange). If the model is doing a good job then the positive distances should be close to zero (because the positives are similar to the anchor) and the distribution of the negative distances should be shifted to the right.

The bottom three histograms show the distribution of differences between the anchor-positive distance and the anchor-negative distance. If the map was perfect, we would expect the entire distribution to be on the right hand side of the red dotted line. As you can see, it's not doing a terrible job at the moment. For most triplets, the map does an alright job. We want to drive this to perfect however! Over time we're going to work hard to shift those distributions further and further to the right, which will provide the most reliable map of Steam you will ever find.

Subscribe for updates!

So that is our Steam Map, we hope you find it as exciting as we do. We really believe it could open up a lot of opportunities for game development. If you would like to stay up to date with our developments and get free monthly Steam insights then please consider subscribing to our newsletter:

Enjoying our insights? Subscribe to our free monthly newsletter!