Let’s See What’s Still Possible

30 August 2026

There’s a lot of talk about AI at the moment. It seems you can ask it to do almost anything. Write something. Plan something. Explain something. Analyse something. And apparently, it can even write you a running training plan.

But can it actually help an ordinary runner train for a half marathon?

I decided there was only one way to find out. Give it a proper test.

Not a two-week experiment. Not asking it for a plan and then deciding whether it looked good on paper. I gave it 14 weeks.

And in two weeks' time, on 13 September 2026, I'll find out how it went when I line up for the Great North Run.

But before we get to the AI, there’s a bit of history to explain.

I’m not chasing PBs anymore

If you've spent any time on Keith Is Running, you'll probably know that I'm not really a PB chaser these days. That wasn't always the case.

Back in 2019, I ran my all-time half marathon PB of 1:43 at the Great North Run. I'd spent around three years chasing that time. And when I finally got it, that was pretty much it as far as chasing a faster half marathon was concerned.

For the next few years, I just ran. I trained consistently, enjoyed running and didn't really have any expectations about what time I should run at the end of it. And I was perfectly happy doing that.

Then came last year's Great North Run.

I'd decided to mix things up a little and put some quality training back into my running. I wasn't expecting anything spectacular, but I ended up running 1:56

That surprised me. It was my quickest half marathon in six years. And it planted a little seed.

Maybe there was still something there.

Then came London

I carried that consistency through the UK winter and put another good training block together in 2026.

That took me to the London Landmarks Half Marathon. The training had gone to plan. The race execution went to plan. And I crossed the line in 1:54 — my quickest half marathon in seven years.

It wasn't until later that day that I started thinking about it.

I hadn't suddenly discovered some magic formula. I hadn't run anywhere near my 1:43 PB. But I'd just run my quickest half marathon for seven years.

And I thought:

I could probably train to give a little more.

That wasn't me suddenly becoming a PB chaser again. It was something different. It was curiosity.

After more than a decade of running, what is still possible for an everyday runner in his 50s?

That was the question.

And I already had the perfect opportunity to find out. I was entered for the Great North Run again.

So I decided to give myself something to aim for. I settled on a target range of around 1:51 to 1:53, with 1:52 sitting nicely in the middle.

My 1:43 PB wasn't the target.

I wasn't trying to become the runner I was in 2019 again. I'm interested in what this version of me can still do.

Let's see what's still possible.

So where does ChatGPT come in?

This is where the experiment started.

I've been running for more than a decade. Over that time I've tried quite a few different approaches to training.

I've used free online training programmes. I've paid for online training plans. I've read books. I've followed a Garmin training plan. I've taken part in online group coaching, including a marathon training programme with weekly online chats. And I've written my own training plans over the years.

The one thing I haven't done is have a one-to-one running coach. I'd actually like to experience that one day, even if it was online, just to see what that personal coaching relationship is like.

But for this particular training block, I wanted to try something different.

There's so much discussion around AI, training plans and coaching that I wanted to find out what all the fuss was about. And I thought there might be other runners wondering exactly the same thing.

Could you use ChatGPT instead of paying for an app? Could it produce a sensible training plan? Could it understand an experienced runner? And could it give you something genuinely useful rather than just throwing a generic plan at you?

So I decided to find out.

And I thought 14 weeks was long enough to give it a proper test.

What did I tell it?

I didn't just type: "Give me a half marathon training plan."

I gave it some context.

I told it I'd been running for more than a decade. I gave it my all-time half marathon PB of 1:43 and the year I achieved it. I gave it my more recent race performances.

I told it how many days I could realistically run. I told it I wanted my long run on a Saturday. I told it that I wanted to incorporate strength training once a week.

And I gave it my target.

1:52.

I wanted to see what it would come up with.

Before asking ChatGPT, I'd actually scribbled down the sort of plan I thought I would write myself. That gave me something to compare it against.

And when the AI produced its plan, I was surprised. Actually, I was slightly amused.

My reaction was basically:

"Bloody hell, the robot knows its running!"

Because the plan was almost exactly what I would have written myself.

The plan

The basic structure was four or five runs a week:

Monday: Easy
Tuesday: Intervals
Wednesday: Rest
Thursday: Tempo
Friday: Easy
Saturday: Long run
Sunday: Rest

It also gave me training paces for the different sessions. And again, they were pretty much spot on compared with what I would have calculated myself.

There wasn't anything particularly revolutionary about the plan. And I don't think there needed to be.

There were easy runs, intervals, tempo work, long runs, recovery and progression.

One thing I particularly liked was the use of a reduced recovery week every fourth week. That made perfect sense to me.

It's the sort of thing I'd expect from a sensible training plan, whether it had come from me, a running coach, a book, an app or AI.

The mileage also increased and decreased in what I thought was a sensible pattern.

And the long runs were good.

In fact, the whole thing looked remarkably familiar.

The interesting thing is that I didn't ask AI to coach me

This is where I think there is an important distinction.

I didn't use ChatGPT as my virtual running coach.

I didn't feed it the result of every run. I didn't tell it that I'd missed a pace and ask it to rewrite the following week. I didn't ask it to constantly analyze my training.

I stuck to the plan.

If a session didn't go to plan, I moved on.

That's something I've learned over 15 years of running. Not every session is going to be perfect. You won't hit every pace. Some days your legs aren't going to cooperate. Some days you're tired.

That's training. Training is the time to make mistakes.

You don't need to panic because you missed a target pace on a Tuesday morning. You take the lesson, move on and concentrate on the next session.

There were times I made my own adjustments

There were a couple of weeks when I felt overly tired. On those weeks, I knocked the Friday easy run on the head.

I didn't ask ChatGPT what I should do.

I made the decision myself.

And that's one of the important differences between what I did and what you might get from an adaptive running app or a coach.

An adaptive app may alter your training based on the data it receives. A coach can look at your training, talk to you and make a judgement about how you're doing.

ChatGPT wasn't doing that.

And because I've been running for 15 years, I was comfortable making that judgement myself.

This is where experience matters

There's something else I think is really important if you're considering using AI to create your own running plan.

AI is only as good as the information you put into it.

If you give it sensible information, realistic goals and an accurate picture of your current running, you're much more likely to get something sensible back.

But if you put silly numbers in, you can get nonsense out.

And that could have consequences.

Tell it you're capable of running a pace you aren't actually ready for and it can build a plan around that assumption. Then you could end up overtraining, burning out halfway through the programme or arriving at race day completely cooked.

You might then conclude:

"ChatGPT doesn't work."

But was that actually an AI problem?

Or was it user error?

That's something I'd be very wary of if I were a relatively inexperienced runner. Because how do you know what information to give it if you haven't yet learned what your body is capable of?

I wouldn't recommend AI for your first half marathon

This is probably where my advice is slightly different from what you might expect after reading this.

If you're reading this and thinking:

"I'm a recreational runner and I've never run a half marathon. Should I ask ChatGPT to write me a plan?"

I'd probably say no.

At least, not as my first choice.

When I ran my first half marathon, AI obviously wasn't around in the way it is now. But even if it had been, I don't think I'd have used it.

I'd have bought an inexpensive running book with a sensible training plan in it. I'd have learned what the different types of runs were for. I'd have followed the plan.

And then I'd have enjoyed the experience.

Enjoy, enjoy, enjoy.

Your first half marathon should be about experiencing what it's like. The atmosphere. The nerves. The excitement. The crowds. The feeling of running further than you've run before. And, most importantly, crossing that finish line.

No expectations. No obsession with pace. Just experience it.

Then, if you cross the finish line and think:

"I want to give this another go."

That's when I'd start thinking about improvement.

That's pretty much what I did.

The runner still has to be part of the equation

One of the things I think gets lost in the conversation about AI is that the technology doesn't actually do the running.

It can give you a plan. It can explain a session. It can calculate training paces. It can help you understand why you're doing something.

But it doesn't know your body unless you tell it. And even then, it only knows what you've told it.

That's where experience comes in.

After 15 years of running, I've learned when something feels like normal training fatigue and when something doesn't feel right. I've learned that one bad session doesn't ruin a training block. I've learned that sometimes the sensible thing is to skip a run. And I've learned that you can't judge a training block halfway through it.

Which brings me nicely to where I am now.

Fourteen weeks later

I'm now into the taper.

And I'm not going to pretend I know yet whether the plan worked. I don't. That's the whole point.

The training has pushed me harder than I've trained for a half marathon in around seven years. But that's what I wanted.

I was deliberately trying to see what might happen if I put a proper quality block together again.

The first eight weeks of effort sessions were hard, Very hard.

But gradually, as my fitness improved, those sessions started to feel more manageable.

Interestingly, my relationship with those sessions has changed over the years too.

Back when I was chasing that 1:43 in 2019, speed sessions were probably my favorite. I would go out, smash the workout, look at the pace afterwards and think:

"Yeah, that was fast." Done.

The easy miles and longer runs? I found those a bit boring.

These days, it's almost the opposite. I don't particularly enjoy pushing my body over and above what it wants to do. Give me a long, slower run and I'm much happier.

Perhaps that's age. Perhaps it's experience.

Probably a bit of both.

And if you gave the 2019 version of me this training plan?

He'd have nailed it.

The 2026 version of me still got through it, just with a little more respect for recovery.

I also failed one part of the plan

Strength training. I said I'd do it. ChatGPT included it. I knew it was important. And then I hardly did any. That's one thing I regret.

Because who knows? Maybe the strength work would have made an even bigger difference.

But that's not an AI failure. That's a Keith failure. And perhaps that's another lesson.

A training plan can tell you what you should do.

It can't make you do it.

So, am I fitter? 

I don't know and I'm quite happy saying that.

I certainly don't feel dramatically fitter at the moment.

But that's also pretty normal for me at this stage of a training block.

It's only now, during the taper, that my legs are starting to come back.

Yesterday I had a shorter effort session and nailed the paces. That gave me a little more confidence.

But confidence isn't evidence. The race will be the evidence.

And that's why I'm not going to give you the answer in this article.

Because I don't have it yet.

The real test is 13 September 2026

The Great North Run is coming.

I've got my target.

1:52.

I've got 14 weeks of training behind me.

I've got a few missed paces, some tired legs, plenty of easy miles, some tough intervals, long Saturday mornings and — let's be honest — nowhere near enough strength training.

I've also got more than a decade of running experience behind me.

And I've got an AI-generated training plan that turned out to be remarkably close to the plan I would have written myself.

But none of that tells me what will happen on race day.

The Great North Run will.

And that's what makes this experiment interesting.

I'm not trying to prove that ChatGPT can replace a running coach.

I've never had a one-to-one coach, so it wouldn't be fair for me to make that claim anyway.

I'm not saying AI is better than a paid training app.

I'm not even saying that everyone should use ChatGPT to write their next training plan.

What I am saying is that, for an experienced everyday runner, AI has produced a sensible plan that I was happy to follow for 14 weeks.

It saved me the time of writing it myself. It gave me training paces that were very close to what I'd have calculated. It explained the sessions well when I asked it questions. And it gave me something to work towards.

But the responsibility for the running was still mine.

And perhaps that's the most important thing I've learned.

AI can give you the plan.

It can't give you the experience.

It can't know what you haven't told it.

And it certainly can't run the 13.1 miles for you.

So what's still possible?

That's the question I've been asking myself since London Landmarks.

I'm not trying to be the runner I was in 2019.

I'm interested in what this version of me can still do.

I've spent 14 weeks finding out whether AI can help an experienced everyday runner put together a sensible training block.

I've done the training. I've made a few adjustments when my experience told me I needed to. I've probably neglected the strength work more than I should have. And now I'm tapering.

But the biggest question remains unanswered.

What is still possible?

I'll find out on 13 September.

Let's see what happens.

Part 2 is coming...

The training is done. The taper is underway. Now there's only one thing left to do.

Run the Great North Run.

And then we'll find out whether those 14 weeks with AI actually made a difference.