Of everything mentioned in the Fall 2025 “This Season on Zwift” Press Release, Zwift’s planned release of personalized recommendations carries the most potential as a game-changing feature. Launching this November, the promise of these AI-powered recommendations is that your fitness data and preferences can be used to serve up a recommended activity, any time you ask.
Let’s dive deeper into this, Zwift’s first public use of AI, and discuss planned features and future potential.
The Challenge: Maximizing Training and Fun for Everyone
Veteran Zwifters like myself are always asking for more. More roads, new features, fresh events. But do we consider what Zwift is like for someone new to the game and/or new to cycling?
It must be overwhelming.
There are thousands of events each week. Hundreds of routes you can ride and workouts to choose from. Thousands of riders milling about, around the clock.
In addition to all that content, numerous metrics are floating around that new cyclists don’t understand. What is FTP? Why does Companion say I’m “overreaching,” and how is that tied to my Training Score and Stress Score? What is XP, what are Drops, and do I care? What’s a good weekly goal? What’s an ideal cadence? Is my heart rate too high? How does Zwift Racing Score work? And let’s not even talk about zMAP and zFTP…
On top of all this, consider that, while every Zwifter wants to maximize their training results, everyone is targeting different results. Some are training for a particular outdoor event. Some want to lose weight. Some are just looking to maintain fitness until they can head outside again, while others are dedicated indoor riders training to win Zwift races. And fitness levels vary: while one rider might consider a full hour on the trainer to be a max effort, another may need two hours before it qualifies as a challenge.
A New Solution?
Zwift has built an ecosystem that can be leveraged to build cycling fitness efficiently and effectively, but many customers need help figuring out how to make that happen.
In the past, Zwifters who wanted to train with intentionality may have paid a cycling coach for a training plan, or dug in and done their own research, or perhaps asked a knowledgeable friend for advice. Zwift’s personalized recommendation engine will soon be the new option on this list. And it will be front and center in the game and Companion, a flagship feature continuously improving thanks to the constant influx of fresh data.
How It Works
No, I’m not going to dive into the depths of LLMs and other AI tech. But let’s talk about how Zwift’s personalized recommendations will work from the perspective of a common Zwifter.
Here’s the first screenshot Zwift shared, showing the beta personalized recommendation block at the top-left of the homescreen:
A few things to note here:
- The interface is very simple. If you want to take Zwift’s recommendation, just click “Start Ride” and go.
- The card explains why this activity was recommended: “This ride fits within your typical weekly activity duration and will help maintain your current fitness level.”
- It also notes the estimated activity duration, distance, and elevation.
If you want to do something different from what Zwift recommends, there’s a shiny “Tune” button, which takes you here:
The Tune screen gives you access to a vast array of recommendations, but through a very simple interface. There are just two things you can tune:
- Activity Type: choose between riding a route, a workout, an event, or hopping into a RoboPacer group
- Activity Length: click an arrow to generate a recommendation that is longer or shorter than what you’re currently seeing
The secret sauce, of course, is the recommendation engine at work behind the scenes. Since it is constantly learning from your activities and feedback, the recommendations it puts forth – regardless of the tuning options you choose – will be tailored to you. That means:
- Routes you like to ride, and perhaps routes whose badges you haven’t yet earned, especially if they’re the type of route you typically ride
- Workouts that help you improve a particular weakness, or work toward a fitness goal
- Events you’ve enjoyed in the past, or similar events
- Robopacer groups riding at the pace you would typically ride
Already working through a training plan, or working with a coach? No problem. Zwift tells me that, if you’re using a third-party workout provider like TrainerRoad, their workout of the day will take preference as the recommendation.
Companion and Game
Zwift says, “Personalized Recommendations will display both in the home screen in the main Zwift App and also in Zwift Companion.” That’s a very good thing, and I’m hoping the tuning tool is also available in Companion, so we can dial in our next activity before hopping on the bike.
Related New Features: Outdoor Connections and Goal Auto-Adjust
Zwift announced two features/changes rolling out in September that will work in tandem with personalized recommendations.
First, an improved onboarding experience will help more Zwifters hook up connections to Wahoo, Garmin, and Hammerhead so their outdoor rides are automatically brought into Zwift and included in key fitness metrics. Outdoor connections launched back in April 2025, and Zwift knows that if riders do significant training outdoors, Zwift needs to see those activities before providing helpful day-to-day recommendations.
Second, Zwift is rolling out the option to have your weekly goal auto-adjust. Instead of having a static goal to hit a particular number for kilojoules, stress points, calories, distance, or time, Zwift will use your recent activities to “adjust your upcoming weekly target goal, helping to support your continual development.”
When?
Zwift says personalized recommendations will launch in November 2025.
Of course, the launch of this feature doesn’t mean the end of its development. That’s not how AI works! I’ve already heard many positive things from Zwift staffers who use the tool internally, and they’re promising that the recommendation engine will continue to improve as it receives feedback from an increasing number of users.
(You can improve an engine like this by changing the parameters used to make decisions, but you can also improve it by feeding in post-activity feedback, such as a star rating scale asking something like, “How well did this activity help you work toward your fitness goals?” Zwift will be doing both.)
What About Other Activities?
Ever since Zwift launched their Fitness Metrics in April, some Zwifters have been asking when running and other workout types will be included. This has been the only consistent complaint about Zwift’s fitness metrics, in fact.
It’s a fair question… because lots of Zwift riders don’t just ride. Many run, lift, ski, hike, swim… you get the picture.
For these athletes, Zwift’s Fitness Metrics, and therefore their personalized recommendations, may not prove useful until Zwift begins to support a wider variety of activities. And there are lots of activity types! As one reader commented recently, “After all, cycling is represented by about 8 of the ~180 activity profiles on a regular Garmin watch. In Strava, I believe cycling is 5-6 out of the 48 available activity types (but lots of activities there are logged under a false flag since the actual activity type is missing).”
Zwift has said they will support running in their Fitness Metrics, but they haven’t given an ETA. Hopefully they’ll figure out a way to do it sooner rather than later, and hopefully they’ll also figure out how to easily include other activity types.
Wrapping It Up
There are things about AI that concern me, such as search engines presenting AI-generated answers to questions and stealing traffic from actual content creators. (This impacts my bank account directly.) The unknowns of AI are even more concerning. Where will it all take us? What are the unintended consequences?
But using AI to deliver personalized recommendations? That’s a use case I can get behind. It’s been discussed for years, and some training platforms have actually done it. But if it can be done well by Zwift, in a way that leverages their massive community and content library, it will enhance the training of tens (hundreds?) of thousands of riders worldwide.
And that’s a big deal.
Your Thoughts
What do you think of Zwift’s plans to offer personalized recommendations? Share your thoughts below!
I think it’d be cool if we could also tell it custom goals we have (not just: “Weekly miles – 100) but other things like “I’m training for Ironman 70.3 ———-, or Revel Big Bear (revel is a road series known for downhill, PR races), Boston, also my goal time is 3:30 for the marathon” Also things like long term goals “I have a yearly goal total of 5000 miles, or cycling 12 Everest’s in elevation and I ride Monday through Saturday, sometimes rest on Friday before really big days” basically like Zwift’s version of Chat GPT, where you can talk to it and it’ll be like “here you go!”
Ooh baby, just what we all need: the Venn diagram of Zwift, Strava, and TrainingPeaks being a circle.😖
(Perhaps rhetorical) Who has those goals and is not already tracking Fitness Life with a thing that is NOT A CYCLING GAME?
Interesting……I guess I would work on your graphics and make them more realistic and true to reality. The giant squirrels, dogs and cats are ludicrous but make me laugh when I see them The people are robotic and stiff, as well. AI could be used for realistic effects?
Pedal harder and you won’t be staring at critters. Voila.
🤣🤣🤣🤣 Absolutely…..makes no difference if the graphics are lousy.
Frankly, you understate the case dramatically (probably in an effort to not offend HQ)….the personalized recommendations are USELESS if they don’t take into account ALL your training.
Frankly a massive chunck of people on Zwift probably have cycling as their majority (or only even) sport, meaning they have no other activity types or they don’t really impact training stress over the week.
Sure, it’s dramatically for your specific usecase but I’m gonna take a wild guess and assume this has 0 impact for many other users.
Also: most serious multi-sport athletes probably have something like a Garmin smart watch anno-2025 meaning they already have these stats in other apps like Garmin connect.
Clearly you don’t have a good grasp on the user base of zwift. More importantly, at some point HQ will try to charge us saying they have added this therefore added value. And since it is useless to those of us who do other things I don’t want to pay for it
Hopefully recommendations can be silenced or uninstalled. Totally not interested.
There needs to be an “opt out” button. Simple as that.
As others said, fitness trends are useless since it misses non-cycling activities, and I wish I could turn the whole thing off.
Well I’m a fan of longer routes (+50k) and I’ve done them all except Knights of the roundabout (France). Here I don’t think AI can do much so I’m hoping for something where I could do many shorter ones without stopping/saving. It would be nice to do them shorter ones in Makuri in one sweep for example.
AI rocks 🤘 we need more AI bullshit. Hope we can turn it of, like the not so awesome strava AI
Zero interest in this. So superficial that it is useless.
I can think of 20 things I’d rather they spent their time developing.
Feels like at the last board meeting some investor said, why aren’t we using AI? And then we get this.
How about using AI coding to speed up road development? That I would be interested in seeing.
How do you know it’s superficial and useless? All we have is two screenshots!
I think, if Zwift can give someone a ride recommendation tailored to their fitness and habits… that’s a really big deal, and a features many thousands riders will happily use.
True, but it seems fair to assume the examples Zwift provided would be a strong representation of what they plan to offer. Telling me that I normally ride for 30 minutes so I should do the Suki Playground route because it will take about 30 minutes is hardly insightful. Even referring to that as “AI” seems like a stretch.
I think even the most novice rider, with a minute or two of contemplation could come up with a workout plan for themselves that eclipses these types of suggestions, and it doesn’t appear to come anywhere close to what a typical rider would consider to be an actual training strategy.
I fit into the group of people who does more than just cycling so until I can include my outdoor runs the feature is no good to me.
That said, assuming that does come then it could be great for me. I’ve always struggled to get the balance between running and cycling right such that one of the two activities ends up getting short shrift.
I’d like to be able to give it goals. For example, I have a half marathon in November to get ready for but I want to at least keep my cycling ticking along. If it could do this sort of thing then the Zwift subscription would really increase in value for me.
How long until the most-requested aspect of this feature – turning it all off – is implemented?
What would help me the most is being able to see the stress point cost of at very least a past completed activity, without praying that there is a screenshot of the final in-game summary screen (which does show it in the top right)
Bizarre this is missing, obviously you can look at daily totals, but for such a simple stat it seems like an oversight to not have it readily available when looking at an activity in review.
Even better still would be having the stress point cost of a to-be-selected workout displayed rather than a vague X/5 flames rating.
Maybe I’m missing something, it seems such a simple and fundamental training stat to ommit.
There had better be a way to turn it off. I don’t need a room full of computers wasting electricity to tell me what routes and workouts I like.
There’s already so much extra going on in Zwift that I just don’t hardly pay any attention to it anymore. It’s supposed to be fun and it feels like work to keep up. Recommendations sounds nice but it needs to push you and not always recommend a coffee ride
Agreed Eric… AI is going to be Scary! It (Zwift Employees Included) is/are going
to replaced by AI… It is just in it’s Infancy, and it (AI) will only get Better, and
Better, and Faster and Faster… Particularly in the Industries that have Platforms
like Zwift… Only time will tell 🤔
Maybe they should focus on improving the navigation rather than wasting money on AI. This feels like a relatively unneeded service, especially given that it competes directly with coaches and other AI training programs being people to Zwift. What would be more helpful is an easy way to look up routes/workouts. I mean how is there not an easy search mode for routes/workouts. Given the number of routes and workouts, sometimes I know I want to ride a certain one and I am fumbling around trying to find it, when I should be able to just start typing its name. It’s like website 101.
Unfortunately, it’s worthless to me if swimming, rowing, SkiErg, training in the gym, etc. aren’t included. In Strava, this works great for fitness and the Freshness Chart by entering your effort factor for sports such as swimming.
Zwift workouts are notoriously poor overall, will they be using AI to improve the quality of the workouts as well? Or is it just using an algorithm to suggest a route or workout from the existing library?
Given that ZHQ is trying to emphasize the quality-workout side of things, will they be implementing the fix for the intentional bug that rounds workout watt targets to 5w for everyone, or is that going to still be limited to integration partners?
Whilst most on here don’t think it’ll be useful, id argue there are others for whom it would be. Newer users for example where they need a nudge to try new things. |Lets be honest this board isn’t flush with zwift novices as a rule.
As always this is phase 1, based on use it could change.
For all those who say it doesn’t include my other activities, fair enough, but you must know what you want from your cycling part? Could phase 2 be where you say over the next 3 weeks I want to do 4hrs a week and want to improve on my climbing, and it suggests a progression of courses to do that based on what you’ve done before.
There’s room for more as well, surely the end goal is you specify your wish and it spits out a custom course with targets for sector, maybe that’s 10 repeats up the hilly KOM with a progressive target time for each ascent, and it automatically reroutes you back down after each banner crossing.