Training guides

How to choose an AI cycling coach

Check what an app uses to choose your next workout. Sleep, HRV and ride notes give it information beyond your power data. Then check whether it can use what you tell it to change the training.

RestOrTrain plans your season and builds the workouts on your calendar. It explains those decisions and sends the sessions to your devices or plays them on your indoor trainer.

How training apps use AI

Competitor features and access terms below were checked on October 9, 2026.

• TrainerRoad builds cycling plans in base, build and specialty phases, selecting and tuning workouts with its own machine-learning models.
• JOIN builds a cycling plan from fitness, goals and availability, then reshuffles it around missed and extra rides.
• Xert uses a three-number fitness signature and an adaptive daily workout advisor.
• Athletica provides adaptive plans with an AI chat that explains and suggests but "does not automatically modify your training".
• ChatGPT, Claude and Gemini are general-purpose AI assistants. They answer training questions, remember details across chats when memory is on, and read supplied files and screenshots.

TrainerRoad, JOIN and Xert have no chat.

What it looks at

Does it use workout data or the whole athlete?

TrainerRoad uses power, heart rate, effort surveys, training history and your calendar. Sleep, HRV and resting heart rate are not inputs. JOIN uses Oura Readiness and WHOOP Recovery scores, without reading raw sleep or HRV. Athletica reads HRV and wellness data through Intervals.icu. RestOrTrain reads sleep, HRV and resting heart rate alongside your rides. It uses them with your training load, plan and calendar.

General-purpose assistants can also read connected data. Claude's Strava connector gives Strava subscribers activity history, including per-second power and heart rate, but no sleep or HRV. In the US, ChatGPT Health reads Apple Health through an iPhone; Claude offers that access in its paid health beta. Both health connections and the Strava connector are read-only.

Does it learn from what you tell it and act on it?

TrainerRoad's effort surveys and JOIN's manual readiness check let you report how training feels. General-purpose assistants can remember details from conversations. RestOrTrain learns from every interaction, including ride notes and post-workout feedback. It keeps what it learns organized for endurance training and uses it in later decisions. Every answer draws on that history, including your strengths, weaknesses and power profile.

Tell it you're away for work Thursday and Friday. It marks the trip on the calendar and immediately reworks the week with your approval. Ask why a session changed: the explanation uses your history and recovery data as well as the trip. Facts remembered from chats are visible in the Memory list on the Profile tab, where you can add or delete entries.

How it builds and explains training

Does it build each workout for you or select from a library?

TrainerRoad selects and tunes workouts from its library. JOIN builds plans from more than 400 workouts.

RestOrTrain sets out season phases with dates, weekly hour ranges and a reason for each phase. It creates each session to fit you and where it falls in that plan. You can request a different session or make an existing one harder or easier in the chat.

Does it calculate your numbers or work them out again each time?

General-purpose models work training numbers out again each time. The results can vary, and they make mistakes. They have no guardrails built specifically for endurance load.

RestOrTrain calculates every number it can without AI, including training load, FTP from your power curve, fitness and fatigue. It answers from those numbers, leaving far less room for error. It also has endurance load guardrails. To inspect a training block, ask for a chart of threshold power across it. The chart appears in the reply using your computed data.

Is the AI tuned for endurance training decisions?

Language models are conservative by default. In training, that means weeks that are too light and key sessions that are too short. RestOrTrain's own tests showed this bias. It is tuned to make the right training call instead.

Do you have to write the prompts?

With a general-purpose assistant, you give the training instructions and bring the context through messages, files or connections.

With RestOrTrain, you can ask in ordinary words. The prompting is done for you, with instructions tuned for world-class endurance coaching performance. It answers with concise, concrete recommendations geared to your performance.

Devices and access

Where does the workout end up?

TrainerRoad delivers workouts to the main bike computers and Apple Watch, and its workouts run in Zwift. JOIN sends them to the main bike computers, Zwift and Rouvy. Both have an ERG player. General-purpose assistants cannot send workouts to devices on their own; some third-party plugins can.

RestOrTrain sends planned workouts to Garmin Edge and Garmin watches, Wahoo ELEMNT, Hammerhead Karoo and COROS, among its workout connections. Zwift receives each workout on its day; Apple Watch receives planned rides and runs when paired with an iPhone. On iPhone and Android, its indoor player runs workouts on a Bluetooth FTMS smart trainer. Each step needs a power target. You can turn ERG off or adjust intensity by 1% per tap. A heart-rate strap is optional.

Does it run on your devices?

TrainerRoad runs on iOS, Android, Windows and macOS. Athletica runs on iOS, Android and the web.

RestOrTrain runs on iPhone and iPad with iOS 18 or later, and Android 8 or later. Its iPad app also runs on Macs with Apple chips. There is no web version. Apple Health and Apple Watch connections require an iPhone. Subscription costs are in the AI training app ranking.

Frequently asked questions

RestOrTrain plans your season and puts today's session on your trainer or Garmin.

A few questions a month are free.

Published September 15, 2026, updated October 10, 2026