Training with ChatGPT in 2026 (1): The Planning Session

Published on By Jinwoo
AIrandonneur

Over the past two years, I’ve trained with both an AI-driven training platform and a human coach.

I started with TrainerRoad. Later, I switched to working with a human coach after being introduced through TrainingPeaks.

Both approaches worked. Both helped me improve.

But in 2026, I want to try something different.

This year, I’m running an experiment: using ChatGPT as the primary tool to design my training plan, while I remain fully responsible for execution, recovery, and judgment.

This is the first post in a series. Series 1 focuses on the planning session. How the plan is created, what information goes in, and what comes out.

Why Change at All?

I genuinely enjoyed structured training with a coach. Having a clear plan, accountability, and long-term direction removed a lot of mental overhead.

So why change?

First: cost. Coaching is expensive. While it can be worth it, it’s not something I want to commit to indefinitely.

Second: understanding the why. When you work with a coach, most of your job is execution. You trust the plan and follow it. That works—but over time, I found myself wanting to understand why certain workouts existed, why weeks were structured a certain way, and how trade-offs were made.

I want to be able to design, adjust, and own my training more independently.

What I Learned from Each Phase

TrainerRoad Phase

Strengths

  • Strong consistency
  • Clear progression
  • Minimal thinking required

Limitations

  • Black-box decision making
  • Limited adaptability to real life
  • Hard to ask “why” a workout exists

TrainerRoad was excellent at keeping me consistent, but it often felt like following instructions without fully understanding the logic behind them, and sometimes it suggests me super hard sessions after another.

Human Coach Phase

My experience with a human coach was very positive. My coach was responsive and willing to answer questions, but flexibility was often tied to pricing tiers. Full adaptability came at a higher cost.

The biggest advantage of a human coach was empathy.

On days when I felt exhausted or mentally drained and couldn’t complete a session, the coach understood. TrainerRoad would simply adjust future workouts based on performance metrics. A human coach could acknowledge the struggle, provide reassurance, and offer emotional support.

There’s also accountability. Knowing a real person is paying attention matters.

But the contrast was clear:

  • ~$22/month vs ~$230/month

That gap between automation and human insight is where I wanted to experiment.

Why ChatGPT, and Why Now?

After two years of structured training, I feel confident in the fundamentals:

  • Weekly structure
  • Training load and recovery
  • Workout intent

I’m no longer starting from zero.

At the same time, large language models have become capable of reasoning across context, asking clarifying questions, and explaining decisions. That makes them interesting—not as a replacement for coaches, but as a planning tool.

So instead of asking ChatGPT to comment on my training, I asked it to create the plan.

How ChatGPT Fits In (Accurately)

ChatGPT doesn’t just react to my training. It actively designs the plan.

I provide a structured prompt that allows ChatGPT to assess:

  • My training history
  • Current fitness and fatigue
  • Time availability
  • Goals and constraints

Based on that assessment, ChatGPT proposes a complete training plan:

  • Weekly structure
  • Key workouts
  • Progression logic

I don’t follow the plan blindly.

I review it, question it, and sometimes modify it—but the initial plan comes from ChatGPT, not from me.

In that sense, ChatGPT plays a role similar to a coach during the planning phase, with two important differences:

  1. The assessment process is explicit and transparent
  2. I can ask why every decision was made

Listening to my body still overrides everything. But instead of starting from a blank page, I start from a plan that has already been reasoned through.

The Planning Session

Before asking ChatGPT to create a plan, I let it interview me.

In the next section, I’ll share:

  • The exact prompt I used
  • Outcome

The Plan ChatGPT Created

To generate the plan, I didn’t start by asking for workouts.

Instead, I asked ChatGPT to behave like a cycling coach and interview me first before producing anything.

Here is the initial prompt I used:

You are a fitness coach focused on cycling. I want to create a training plan for 2026, with a primary focus on endurance cycling and a secondary goal of hill strength. The actual plan should start in mid-January. I will have 3–4 hours total availability across Monday to Friday. On weekends, I have time for at least one long ride of 12 hours or more. The main goal is two A events:

1) Coulee Challenge (August 6–9), a 1200k ride 2) Progressive brevets (200k, 300k, 400k, 600k) leading up to the event

I don’t race or compete with others—this is randonneuring-focused training.

Before creating the plan, ask me questions like an interview so you can build a better plan.

ChatGPT then asked a series of follow-up questions about:

  • My recent training history
  • Weekly consistency
  • Strengths and weaknesses
  • Fatigue tolerance
  • Recovery habits
  • Non-negotiable life constraints

Only after that assessment did it propose a plan.

The Resulting Plan

After answering the interview questions, ChatGPT generated the following high-level plan:

Training Plan

What stood out to me immediately:

  • Clear phase separation (base → specificity → brevets → peak)
  • Conservative recovery after long brevets
  • Explicit rules around not resuming intensity too early

This felt less like a generic plan and more like something designed for the constraints I described.

What I Did With the Plan

I didn’t follow this plan verbatim inside an app.

Instead, I:

  • Copied the plan into Google Sheets as a reference
  • Use it as a north star for weekly decisions
  • Manually create weekly workout blocks in TrainingPeaks based on this structure

TrainingPeaks

This keeps the plan flexible while preserving intent.

ChatGPT designs the framework. I handle execution, adjustments, and recovery decisions.

What Comes Next

This post covers the planning session.

The next posts in the series will cover:

  • Month 1: execution vs reality
  • The first setback
  • Mid-block adjustments
  • End-of-season reflections

I don’t know if this approach will work better—or at all.

But I’m curious to find out, and I’ll share what I learn along the way.

Comments

Add a Comment