Which health and fitness connectors do people actually use?

If I build a connector for my fitness app, what should it do?

Let people ask about data they already have. Every fitness connector near the top of ChatGPT or Claude reads runs, sleep, heart rate or lab results the user has already collected. The calorie counters and workout loggers rank much lower.

Some context: in Claude, the top fitness connector is Strava, at #33 of 6,444 ranked connectors. In ChatGPT, OpenAI's own Health plugin sits at #16 of 7,403. The next three in ChatGPT are Tredict (#41), Fitness AI Connector (#68) and COROS (#72). All of them read your own data.

Which fitness connectors rank highest?

In ChatGPT, the five health plugins in the top 200 are all readers. The first logger, Calorie Tracker, comes in at #265. In Claude, the six health and fitness connectors in the top 10% of the directory (Strava, Health Data Avatar, Function Health, COROS, Alma and HealthEx) all read the user's own records: activities, health history, lab results or nutrition logs.

OpenAI's Health is a first-party plugin, so its rank likely reflects OpenAI's own placement as much as demand. The independent ones are more telling. Fitness AI Connector says on its own site that it has "60,000+ users and growing."

Does fitness do better than other categories?

Not as a whole. The median health and fitness plugin in ChatGPT ranks #3,550, a little better than the median plugin (#3,711). In Claude it ranks clearly worse: #4,246 against #3,238. Most of the 327 apps in the category are small calorie counters, gym logs and meditation timers. So "fitness does well in AI" is not true in general. Readers of your own data do.

What do the top fitness connectors actually do?

The tool lists tell the story. Here is what the leaders offer, quoted from the store listings:

• COROS: sleep, recovery and training load. querySleepOverview "Returns sleep score, main sleep duration, deep/light/REM ratio, awake time." queryRecoveryStatus is for "when the user asks how recovered they are or how much energy they have left." It also writes back: createScheduledWorkout puts a workout on the user's COROS schedule.
• Fitness AI Connector: Garmin data in plain language. get_health_summary returns a "daily Garmin health snapshot," for "'how am I doing today?' queries." create_workout sends a workout "to their Garmin Connect calendar so it syncs to their watch."
• Tredict: the raw time series. Its activity tool returns speed, altitude, cadence, power and heart rate for a single run, and show-activity-ui draws charts in the chat.
• Strava: activities and streams. Tools such as list_activities and get_activity_streams, and a promise to "help spot training patterns, suggest improvements, cheer you on." Strava is in Claude only.
• freddy: one place for every device. connect_source will "Connect a wearable, ring, CGM, training platform, gym log, Apple Health, or Health Connect."

Tredict runTredict heart rateLiftTrack weekAI Endurance recovery

Why does "analyze my data" work so well in chat?

The data already exists. A watch records every run and every night of sleep with no effort. A logger asks you to type in meals or sets, and it competes with an app built for exactly that job. A reader asks for one sign-in.

The questions are open-ended. People ask "Why was my run slower than last week?" or "Is my HRV down because of travel?" A chat assistant with your raw data can answer questions that no dashboard has a screen for.

People come back. You train again tomorrow, so you ask again tomorrow. That repeat use is what moves a connector up the rankings.

The assistants push in this direction too. When we typed "i want to analyze my workout data" into Claude, it searched its connector directory and offered Strava.

Both AI companies have also built health data into their own assistants. Anthropic's healthcare announcement says US Claude Pro and Max subscribers "can choose to give Claude secure access to their lab results and health records." OpenAI's Health plugin connects "supported medical records and Apple Health data."

What if my wearable brand isn't in these stores yet?

You are not alone. We found no Oura, Whoop, Fitbit or Garmin connector in ChatGPT or Claude. Their data still reaches chat, through other people's connectors: Fitness AI Connector reads Garmin, and aggregators such as freddy and Wellness Project read Oura, Fitbit, Withings, Polar and Apple Health.

We think the wearable brands are waiting too long. When a Garmin owner asks ChatGPT about their training today, the answer comes through a connector that Garmin does not control. (More on this pattern in who builds AI connectors.)

So, what should you do?

Start with read tools. If your app already holds workouts, sleep or body data, let people ask about it. That is what every leader does.

Name tools after the question. "Query sleep overview" and "recovery status" tell the model what each tool is for. COROS also says in each description when the model should use it.

Add one write tool people want. COROS and Fitness AI Connector both send a planned workout back to the watch. That closes the loop from question to training.

Ship your own before someone else does. If a third party already reads your users' data, it also owns the answer they get.

Methodology

We read the ChatGPT and Claude directories every day and give each app one category. All numbers here are from October 20, 2026. ChatGPT rank is the position in ChatGPT's plugin directory, and Claude rank is Anthropic's popularity rank; we never mix the two. "Reads existing data" is our own reading of each connector's tools. More on our method page.