Why Do AI Workout Plans Ignore Custom Exercises?

Direct answer: Custom exercises are often ignored because the app lacks structured information about their movement pattern, muscles, equipment, loading, and progression. A name alone is not enough for safe automated programming.

This guide is based on recurring patterns in recent App Store reviews, public community discussions, official product documentation, and relevant research. Individual reports do not prove that every user will experience the same problem, but repeated complaints reveal where a fitness product needs clearer design and safer defaults.

The complaint behind the question

Experienced lifters report importing routines only to see duplicate custom entries, blank demonstrations, forced supersets, or a generator that never selects the movements they created. Manual logging works, but the supposedly intelligent planning layer treats the exercise as unknown.

The practical standard is simple: an app should make an important fitness decision easier to understand, perform, or verify. When automation adds a polished result without showing its assumptions, users inherit the uncertainty while the software receives the credit.

Why this happens

Automation depends on metadata. Without a stable exercise identity and meaningful tags, an app cannot compare a custom movement with alternatives, estimate fatigue, or decide where it belongs. Guessing from the title can create more errors than leaving it manual.

Fitness apps operate with incomplete information. They may know the workout history but not the exact machine; see a camera frame but not pain; recognize a food but not its hidden oil; or receive wearable data without knowing the user’s full context. Strong products expose those limits instead of presenting every output with equal confidence.

A five-point quality check

1. Custom exercises have editable movement and equipment fields. Ask whether the app makes this condition visible and easy to correct before it affects the next recommendation.
2. Duplicates can be merged without losing history. Ask whether the app makes this condition visible and easy to correct before it affects the next recommendation.
3. The user controls whether the generator may select them. Ask whether the app makes this condition visible and easy to correct before it affects the next recommendation.
4. Video or notes remain attached across devices. Ask whether the app makes this condition visible and easy to correct before it affects the next recommendation.
5. Progression stays separate from superficially similar exercises. Ask whether the app makes this condition visible and easy to correct before it affects the next recommendation.

Run this check using a real session, not a marketing screenshot. Set up the equipment you normally use, interrupt the expected flow once, and inspect what is saved afterward. Reliability appears in recovery from ordinary messiness: a busy gym, a changed exercise, a weak connection, or an imperfect meal.

A safer decision rule

Use custom exercises manually until the app can represent them accurately. The right product behavior is transparent: it should say when a movement is loggable but not eligible for automated planning or camera coaching.

The important distinction is between assistance and authority. Software can organize evidence, recognize patterns, reduce tapping, and make a reasonable suggestion. The user still needs a visible path to inspect the evidence, change the choice, and recover from an error without losing history.

What good product behavior looks like

A trustworthy app should communicate three things at the moment they matter:

• What it observed. The exercise, set, image, sensor signal, or history used.
• What it inferred. The recommendation or classification produced from that evidence.
• What remains uncertain. Missing context, unsupported movements, weak visibility, incomplete data, or a result that needs confirmation.

This separation prevents a guess from looking like a fact. It also helps users report better feedback: instead of saying “the AI is bad,” they can identify a wrong observation, an unreasonable inference, or a missing control.

Where Boddy fits

Boddy’s live camera guidance is bounded by supported exercises. A custom movement can still belong in a user-controlled workout, but the interface should not promise recognition or correction where the model has not been designed and tested for that movement.

Boddy combines workout planning, supported phone-camera form guidance, meal logging, recovery, and progress on iPhone. That wider context is useful only when each feature stays within its real capability. Live feedback can improve the moment of a set; it does not replace sound programming, reliable records, user judgment, or qualified medical care.

A quick test before you depend on the feature

Try the feature with a low-stakes example. Compare the output with information you already know, deliberately make one correction, close and reopen the app, and confirm that the corrected result survives. Then test the edge condition most likely in your routine—different equipment, poor connectivity, partial camera visibility, a mixed meal, or a second connected device.

If the app explains the limitation, preserves your correction, and gives you control, the feature may be dependable enough for its intended job. If it silently rewrites history, hides the input, or blocks correction, the problem is not merely accuracy. It is recoverability.

Bottom line

Choose fitness technology that behaves predictably when the real world is inconvenient. The most useful AI is not the one that produces the most outputs. It is the one that reduces friction while making uncertainty, control, and safety easier to see.

Sources checked in September 2026

• ACSM progression models for resistance training
• Apple Health and fitness developer guidance
• Boddy product and coaching scope