How to get better results from your ChatGPT workout plan

Hands up if you’ve done this: opened ChatGPT, Claude, or Gemini, and asked it to build you a workout plan.

I’m not asking because I think it’s a bad idea — that’s exactly where I started too, and my first AI-built program was surprisingly good. Solid split, sensible sets and reps, close to what I’d get from a coach. It helped me increase my squat 3rpm by nearly 20% in 8 weeks and get a new PB.

The plan wasn’t the problem. Tracking it was. I knew from experience that if I wasn’t logging my weights, I’d end up repeating the same numbers every week without noticing — no progress, no pushing myself.

So I tried everything. The Notes app on my phone, which meant scrolling back every session to remember what I’d done last time. A spreadsheet, which I’d genuinely bring on my laptop to the gym — useful for seeing the numbers side by side, ridiculous to actually lug around. A few different tracking apps, which either locked me into a prebuilt program that didn’t fit my goals, or made me rebuild everything manually, one line at a time, on my phone.

None of it worked. So I spent the next four months building Zena.

I still think AI-generated programs are a good place to start for people who can’t access (or afford) a coach. But I’m the kind of person who wants to see the data — I want to know, objectively, whether what I’m doing is working. And after pushing too hard on a previous program and actually injuring myself — which stalled my training for 12 months — I wanted something that could tell me when to push and when to hold back, too.

So I built the thing I actually needed: somewhere to bring the plan you already have — from ChatGPT, a trainer, anywhere — and get real tracking and coaching layered on top of it.

If you’ve got a plan like that (AI-written or otherwise) and you’re not seeing results, here’s what I’ve learned from a year of experimenting with this.

How to get more out of the plan itself

Be specific, not vague. “Write me a strength program” gets you something generic. “I can squat 60kg for 5, I train 3 days a week with a barbell, rack, and dumbbells, and I want to add 10kg to my squat in 12 weeks” gets you something you can actually use. The more precisely you describe your real situation, the less generic the output.

Ask for progressive overload to be built in, explicitly. A plan that repeats the same sets, reps, and weight every week isn’t really a program — it’s one workout copy-pasted. Ask directly: “build in progressive overload across the 12 weeks, and tell me exactly how the load or reps should increase week to week.” (This is exactly what tripped me up early on — if you want the mechanics of why it matters, I wrote a full guide to progressive overload.)

Ask it to sanity-check your timeline. A good coach tells you if what you’re asking for is unrealistic before building anything. ChatGPT won’t do that unless you ask it to — so ask: “is this goal realistic in this timeframe, and if not, what would be?”

Feed your actual results back in. It has no idea what you lifted last Tuesday unless you tell it. If you want the plan to actually adjust over time instead of staying static, you have to manually paste your logged weights, reps, and how they felt back into the conversation — and keep doing that every week. This is the part I gave up on eventually, if I’m honest.

Here’s roughly the prompt I’d use if I were still doing this in ChatGPT — combining all of that into one message.

I want you to build me a strength training program. Here's my situation:

- Goal: [e.g. add 10kg to my squat, or build general strength]
- Current numbers: [what you can currently lift, for your main lifts]
- Schedule: [how many days a week you can train]
- Equipment: [what you actually have access to]
- Timeframe: [how many weeks you want this program to run]

Please:
1. Sanity-check whether this goal is realistic in this timeframe — and if it's not, tell me what would be.
2. Build in progressive overload across the whole block, and be explicit about exactly how the weight or reps should increase week to week.
3. Ask me anything else you need to know before building it.

Ps. Zena can also create programs for you, and it takes all of that into consideration (as well as all the data on your previous numbers).

What no AI chat plan can actually do for you

I found both of these out the hard way.

It has no memory of what actually happened. Every adjustment depends on you manually re-entering your own training history, every single time. There’s no log, no persistent record, no auto-fill of last week’s numbers — just a chat thread you have to scroll back through, doing the job a training log should be doing for you. I’m still arguing with it all the time because it keeps forgetting PBs I’ve told it about.

It has no idea how you’re actually recovering. It can’t see your sleep, your heart rate, or your Apple Health data — it only knows what you type. So the plan it wrote in week one has no way to adjust itself when week six is a rough recovery week, unless you notice that yourself, describe it accurately, and remember to ask.

Neither of those is something you can prompt your way out of. A chat window just isn’t built to be a training log or a recovery-tracking system, no matter how well it writes the program in the first place.

If any of this sounds familiar

You don’t need to throw out a plan you like to get real tracking on top of it. That’s the whole premise behind Zena — import the program you already have, from wherever it came from, and get the logging, the recovery-based coaching, and the honest answer to “is this actually working” that a chat thread was never going to give you.

Let me know if you try it, I’d love to know what you think.