Ten years ago I finished my first marathon in 4 hours and 52 minutes. I cried at the finish line, ate an entire pizza, and told everyone who would listen that I was going to run a sub-4 the following year. That did not happen the following year. Or the year after that. For most of the next decade I plateaued somewhere between 4:10 and 4:25, cycling through the same training mistakes on loop, wondering why I wasn’t improving despite putting in the miles. Sound familiar?
The thing nobody tells you about being an intermediate runner is that it is the loneliest place to be. Beginners have endless advice aimed at them. Elite runners have coaches. But if you have been running for three or four years, can comfortably knock out a half marathon on a weekend, and still can’t crack that target time, you are kind of on your own. Hiring a real running coach runs about 150 to 300 dollars a month. Premium app subscriptions stack up fast. And most generic training plans assume you’re either a complete beginner or already running 70 miles a week. None of them account for the fact that you had a terrible sleep Monday, your easy runs are actually not that easy, and your left knee starts complaining on anything over 16 kilometers.
That is the exact gap that AI coaching started filling for me, and after two years of experimenting with it seriously, I want to break down specifically what changed and why.
The Problem With Generic Training Plans 📋
Most marathon training plans are built around a fictional average person. They assume you will hit every session, recover on schedule, and live somewhere flat with predictable weather. Hal Higdon’s Novice 2 plan, for example, has you running 5 days a week with a long run that increases by about 1.6 kilometers each week. It’s a solid framework, but it doesn’t know that you work shifts, that you live in a city with hills that add 20 percent more effort to every outdoor run, or that your easy pace is actually 10 to 15 beats per minute above your aerobic threshold because nobody taught you about heart rate zones until recently.
The result is that a lot of intermediate runners spend months building volume without ever addressing the actual bottleneck in their performance. For me, that bottleneck turned out to be simple and embarrassing: I had been running my easy runs too fast for years. My so-called easy pace of 6 minutes per kilometer was still pushing me into zone 3, which meant I was accumulating fatigue without building the aerobic base that actually makes you faster. I only found this out when an AI coaching tool analyzed 14 weeks of my logged runs and flagged that my heart rate on recovery days averaged 158 beats per minute. For context, my true easy zone caps out at around 145.
No generic plan would have caught that. A human coach would have, but I wasn’t paying for one.
What AI Running Coaches Actually Do (And Don’t Do) 🤖
Let’s be clear about what AI coaching is in 2024, because there’s a lot of hype and some legitimate skepticism worth taking seriously. AI coaching in running apps is not magic and it is not a replacement for a certified coach who watches you move, assesses your mechanics, and adjusts your plan in a real conversation.
What it actually does well is pattern recognition across your own data at a scale and speed that would take a human coach hours. When you feed it consistent GPS data, heart rate readings, pace splits, elevation, and rest days, a good LLM-based coaching tool can identify trends you simply cannot see yourself. It can notice that your pace in the final 5 kilometers of your long runs has been slowing by an average of 45 seconds per kilometer over the last 6 weeks, which suggests you’re going out too fast or your fueling strategy needs work. It can flag that your Wednesday tempo runs consistently produce higher heart rate numbers than your Friday equivalents, possibly pointing to mid-week sleep debt.
What it struggles with is nuance and accountability. It can tell you to run a 25-minute easy jog but it cannot tell whether the shin tightness you mentioned briefly in a text input three weeks ago is getting better or worse. It cannot read your body language when you are clearly overtired and pushing anyway. And the quality of the output is heavily dependent on the quality of the data you put in. If you forget to log your sleep, skip GPS tracking, or run with your watch inside your jacket, the coach is working with an incomplete picture.
The sweet spot is using AI coaching as a consistent analytical layer that surfaces information you then apply with your own judgment.
The Specific Changes That Actually Moved My Numbers 📉
After two years of taking AI coaching seriously, here are the concrete adjustments that contributed to me finally running a 3:58 last spring.
First, slowing down my easy runs. Once the pattern analysis flagged my heart rate problem, I dropped my easy pace from 6:00 per kilometer to 6:45 to 7:00. For about six weeks this felt humiliating. I was being overtaken by people walking their dogs. But my heart rate on those runs dropped to a genuine zone 2, and within two months my tempo run paces improved by about 15 seconds per kilometer without any increase in perceived effort. The aerobic base was actually building.
Second, restructuring my weekly layout. My instinct had always been to cluster my hard days together because I thought I was making them count. The AI analysis showed a consistent performance dip in the second half of every week, suggesting I wasn’t recovering between quality sessions. Spreading the hard days further apart and adding a true rest day on Thursday instead of Sunday changed my energy levels noticeably within three weeks.
Third, addressing my long run pacing. My AI coach flagged that I was running my long runs at about 80 percent of marathon goal pace, which is too fast for the aerobic adaptation you’re trying to trigger at that distance. Pulling back to 85 to 90 percent of goal pace and extending the distance by 15 minutes instead felt counterintuitive but the data supported it.
None of these were revolutionary insights. A good human coach would have told me the same things. But I had never had a good human coach, and I had never been able to see these patterns in my own data because I didn’t know what to look for.
Replacing Premium Subscriptions Without Losing Functionality 💸
Here’s the part that matters practically if you’re trying to do this without spending a lot of money each month.
For years I paid for a premium running analytics subscription that gave me detailed pace zone breakdowns, segment analysis, and monthly progress summaries. It cost around 8 dollars a month, which sounds low but adds up to nearly 100 dollars a year for a runner who is, at the end of the day, just a person who runs for fun and personal goals. When Strava increased its premium pricing, I started looking for alternatives seriously.
The honest answer is that free tools have genuinely caught up for most of what intermediate runners actually need. Detailed pace zone analysis, elevation-adjusted splits, heart rate zone tracking, monthly and annual progress summaries, and AI-generated training suggestions based on your personal performance history are now available without paying a monthly fee.
One app I’ve been using, Geowill, offers free analytics that cover all of those functions, plus an AI coaching layer that analyzes your pace history and generates personalized training suggestions. It also does something I genuinely enjoy and didn’t expect to care about: it auto-generates a 3D flyover video of your route after each run, which sounds gimmicky until you run somewhere beautiful and want to share it. But more practically, the free analytics are competitive with what I used to pay for.
The point isn’t any specific tool. The point is that if you are paying a monthly subscription primarily for analytics features and you haven’t checked what free alternatives now offer, it is worth 30 minutes of your time to do that comparison honestly.
Building the Habit That Makes the Coaching Work 🔄
Data analysis and personalized training suggestions are only useful if you are consistent enough to generate meaningful data in the first place. This sounds obvious but it is the piece most people skip over when they talk about AI coaching.
The minimum threshold for useful pattern recognition is roughly 8 to 10 weeks of regular, consistently logged runs. Before that point, the AI doesn’t have enough signal to distinguish your patterns from normal variation. A lot of runners try an AI coaching feature for two or three weeks, find the suggestions generic or slightly off, and give up. The suggestions are generic because the data is insufficient. Keep going.
Practically, this means logging every run even when it goes badly, wearing your heart rate monitor consistently even when it feels annoying, and being honest in your notes about how a run actually felt rather than how you wanted it to feel. The more honest context you provide, the more accurate the analysis becomes.
One thing that genuinely helped my consistency was gamification. Leaderboards, streaks, and treasure-hunt style location challenges sound juvenile until you realize you have run 8 extra kilometers this month chasing a prize marker on a map. Motivation doesn’t need to be sophisticated to be effective. If a notification telling you there is a rare item three kilometers away gets you out the door on a cold Tuesday evening when you otherwise wouldn’t have gone, the method worked.
The Real Lesson After 10 Years 🏅
Running improvement is rarely about running more. For most intermediate runners, it is about running smarter, which means understanding what your body is actually doing during training rather than what you assume it is doing.
The reason AI coaching made a difference for me after a decade of plateauing is not that the AI knew something magical. It’s that I finally had a consistent, honest record of my training analyzed by something that had no ego investment in the conclusions. It told me my easy runs weren’t easy. My hard days were too clustered. My long runs were too fast. I had heard vague versions of all of these suggestions before and ignored them because I couldn’t see the evidence clearly enough to believe them. Seeing the patterns visualized in my own data over 14 weeks of runs made them impossible to dismiss.
If you’re an intermediate runner who has been following generic plans for a few years and wondering why your times aren’t moving, the answer is almost certainly in your data. You just need the right analytical layer to surface it. That layer is now free, which is honestly kind of remarkable. Use it.