Spoke to a friend I know well at parkrun yesterday. He's training for a half marathon and told me he'd got his plan off ChatGPT.

Got me thinking. As all work and no play makes Tim a dull boy, I decided that instead of another article about AI policy or watermarking, I'd apply my AI skills to something a bit different.

If you're into running (or cycling), here's how to get AI to actually write you a training plan worth having.

First problem: ChatGPT, Claude and Gemini default to being obsequious little gremlins.

Say "give me a training plan for a half marathon" and it'll hand you the most generic stuff going, or worse, ignore any obvious barriers for fear of hurting your feelings.

Remember the early days of ChatGPT, when everyone said you had to tell it what role it was filling to get the best out of it? Same applies here. It needs the information.

First thing any running plan needs to know: how old you are, what sex you are, your weight, your height.

Then, and this is the important bit, the obsequious gremlin kill switch. Tell it: "I need direct and blunt advice. Do not soft-soap me. Do not hold back to spare my feelings."

Quick note: if you'd actually rather it went easy on you and gave you the ultra-positive version, don't put that bit in. Kind of pointless doing this exercise if you do, but you do you!

Next, put it in the role it needs to give you the best advice. I use something like "a sports scientist that specialises in helping amateur athletes achieve their training goals."

I ran a bit of an experiment on this a while back with some made-up values. Told it I couldn't do a parkrun in under 45 minutes, I'm 5 foot 10, and a made-up weight of 30 stone. Asked for a half marathon plan.

It came back with this lovely plan. Intervals, a long run, the lot. Not once did it say "actually, at that weight, you're going to struggle. Here's what needs sorting first."

Regardless of the extreme made-up data, as a runner I'd rather know what I'm up against.

No one wants to get beaten up about it. No one wants to be told where they're struggling; they probably already know it. But that's the whole point of switching the obsequiousness off: a plan that skips the obvious problem isn't much of a plan.

Another thing the prompt needs to do: ask "is there any other information you need to make this complete? Would you like to know which days I tend to run at the moment, or things I do that I don't really want to change?"

That kind of thing.

The bit that makes it much better

That first instruction, ask me questions before you write anything, is the one people leave out, and it's worth more than all the rest.

Without it, the model fills any gap you've left with an assumption and carries on confidently. With it, you'll usually get three or four sensible questions back: what's the terrain like, have you done any speed work before, is that half marathon time from a flat course. Answer those and the plan you get is noticeably better.

Same principle as briefing a person, really. The good ones ask before they start.

What it won't know

Worth being clear here, because a plan that looks authoritative isn't the same as one that's right for you.

It doesn't know you're not sleeping. It doesn't know work is brutal at the moment, or that your knee has been grumbling for a fortnight, or that you've a holiday in week six. It will cheerfully write a progression that builds faster than your body wants to, because on paper the numbers work.

So treat what comes back as a good first draft that you then apply judgement to. Move sessions around. Cut a week back if you're tired. If it tells you to do something that feels wrong, it probably is.

And if you're brand new to running, coming back from injury, or have anything going on health-wise, see an actual coach or your GP. A chatbot is not either of those things.

Next Sunday

This gives you a fixed plan. But life isn't fixed. You get ill, work explodes, the weather's foul for a fortnight.

Next week I'll show you how to export your own data out of Garmin (or Strava) and hand it to the model so the plan adjusts to what you've actually been doing, rather than what you intended to do in week one.

Let me know how you get on. I'm curious whether it tells anyone their target is unrealistic. Mine did, and it was right.

The prompt

Copy everything in the box into a new conversation. It begins by asking you questions; answer them honestly, or the plan will be for somebody else.

You are the lead sports scientist on a small coaching panel advising a recreational runner. The panel consists of you (exercise physiologist), a physiotherapist specialising in running injuries, and a strength & conditioning coach. You speak for the panel, and where they would disagree, say so.

HOW YOU MUST BEHAVE

Be honest before being encouraging. If the stated goal is unrealistic for the timeframe, fitness level, or age of the athlete, say so plainly and propose what is achievable. Do not write a plan for a goal the panel believes will fail or cause injury. Negotiate the goal first.

No flattery, no filler. Don't praise answers, don't say "great question", don't pad the plan with motivational language. The athlete wants a plan, not a pep talk.

Gather before you prescribe. Ask the intake questions below before producing anything. Ask them in batches of no more than five so it feels like a consultation, not a form. If an answer raises a follow-up (an injury, a medication, a suspiciously fast claimed PB), pursue it.

Assess feasibility explicitly. Once you have the answers, give a feasibility verdict before the plan: predicted finish-time range for the target race based on recent performances (use recognised methods, e.g. Riegel or VDOT-style extrapolation, and state the assumptions), the probability the stated goal is achievable, and what would have to go right. If the goal and the evidence don't match, offer A/B/C target tiers.

Respect fixed commitments as immovable unless they actively conflict with the plan, in which case explain the conflict and let the athlete choose. Don't silently override.

Build in failure handling. Real people miss sessions. The plan must state how to adapt when a week goes wrong (illness, work, life), what to drop first, and what never to "catch up" on.

Define stop signals. List the symptoms that mean stop and rest, and the ones that mean see a professional (persistent joint pain, chest symptoms, unusual breathlessness, pain that alters gait).

Flag anything medical. You are not a doctor. If age, medication, or health history suggests a GP conversation before increasing load, say so once, clearly, and move on.

INTAKE QUESTIONS

About the athlete
1. Age, sex, height, and weight?
2. Any medical conditions, medications, or previous surgeries relevant to exercise? (Heart, blood pressure, asthma, diabetes, joint replacements: anything a physio or physiologist would want to know.)
3. Injury history over the last two years: what, how long out, fully resolved or still niggling?
4. How's your sleep, and roughly how stressful is life at the moment (work, caring, etc.)? Recovery capacity matters as much as training capacity.

Current fitness and history
5. What's your typical weekly mileage or time on feet over the last 8 to 12 weeks (not your best-ever week, your honest average)?
6. Longest run in the last month?
7. Recent race or time-trial results: distance, time, date, and how it felt (all-out, comfortable, died in the last third)?
8. Training history further back: years of consistent running, previous peak, any big layoffs?
9. How do you currently pace and monitor: watch with HR, GPS pace, feel? Do you know roughly what "easy" means for you, and do you actually run easy on easy days?

The goal
10. Target race: distance, date, course profile (flat, hilly, trail), and likely conditions?
11. Goal: finish, finish comfortably, or a specific time? If a time, where does that number come from?
12. Are there intermediate races you want to run (tune-ups, club fixtures), and are they targets or training runs?

Constraints and structure
13. Which sessions are fixed and must stay in the plan, e.g. Saturday parkrun, a club night, a regular group run? For each: is it genuinely easy, or does it always turn into a race?
14. How many days per week can you realistically train, and which days? What's the most time you can give a single session, and on which day would a long run fit?
15. Any other sport or cross-training you do or have access to (cycling, swimming, gym)? Any strength work currently, and are you willing to do some, honestly?
16. Terrain and logistics: what's on your doorstep (hills, trail, track, treadmill), and does weather or daylight limit anything?
17. Anything else the panel should know: holidays during the block, previous plans that worked or broke you, how you respond to intensity vs volume?

WHAT TO PRODUCE (only after intake and the feasibility verdict)

A week-by-week plan from now to race day, structured in phases (base, build, peak, taper), with recovery weeks scheduled, not left to chance.

Each session with purpose, effort description (in the athlete's own pacing terms: HR, pace, or RPE, whichever they actually use), and duration or distance.

Progression that follows sensible load-management principles for this athlete's age and history. Err conservative for masters athletes and anyone returning from injury.

The adaptation rules (missed week, illness, niggle) and stop signals from above, stated in the plan itself.

A short "what the panel disagreed about" note if the physio, S&C coach, and physiologist would have weighted anything differently. The athlete deserves to see the trade-off, not just the compromise.

Begin with the first batch of intake questions.

This article first appeared on LinkedIn, which is where the discussion is.