Garmin Daily Suggested Workouts: Three Weeks of Following Them Exactly

Garmin’s daily suggested workout is the most-used AI coach in running and almost nobody uses it as a coach. It sits on the watch face every morning, offers a number and a benefit label, and most people glance at it, think “sure, roughly,” and then run what they were going to run anyway. So I stopped doing that. For 21 days, from 8 to 28 September, I did exactly what my Forerunner 965 told me to do, on the day it told me to do it, at the targets it gave me, with a marathon loaded into the Garmin Connect calendar 14 weeks out.

The short version: it is an excellent easy-run manager and a poor race build. Those are not the same job, and the gap between them is structural, not a bug that a firmware update fixes.

The rules of the test

One rule, really. Whatever the watch showed at 06:30 is what I ran. If it said rest, I rested. If it said 26 minutes under 136 bpm, I ran 26 minutes under 136 bpm, even on a Sunday when every instinct said otherwise. I did not scroll ahead and pick the suggestion I liked better. I did not swap a Thursday threshold onto a Saturday because of childcare.

Baseline going in: VO2max estimate 52, Connect race predictor giving 3:21:14 for the marathon, and three prior weeks at 58, 61 and 55 km. Training Status read Productive. HRV Status read Balanced, 7-day average 61 ms. Nothing exotic, a runner in ordinary shape at the start of a build.

Week one: it is genuinely good at the boring part

Here is the card from the second morning, transcribed as the watch presented it:

TUE 9 SEP · DAILY SUGGESTION
Primary benefit     BASE
Duration            48:00
Target              HR 139-152 bpm (Zone 2)
Training readiness  71  (Good)
HRV status          Balanced
Acute load          412   (optimal 388-612)
Load ratio          0.9

That is a well-judged easy run and I would not have written anything different. Across the full 21 days the easy-run prescription was consistently sane: durations between 38 and 72 minutes, heart-rate ceilings that kept me honest, and a recovery day inserted the morning after every hard session without fail. If your actual problem is that your easy days are too fast and your hard days are too soft, obeying this thing for a month will fix you.

Week one totalled 44 km across five runs and two rest days. Against the 62 km my build called for, that is a 29% shortfall in the first week of a marathon block, which is where the trouble starts.

The long run never arrived

Over 21 days the longest run the watch ever offered was 1:34, 18.2 km, labelled Base. It came on Sunday 14 September. The following Sunday it offered 1:12. The Sunday after that, 58 minutes.

For a 3:20 marathon you need long runs progressing toward 2:45 to 3:00, with chunks at goal pace stitched in from about week six. Daily Suggested Workouts does not build toward anything on a weekly cadence, because it has no concept of a week. It has yesterday, your acute load, your HRV, and a race date. A long run is a deliberate overreach whose payoff arrives in a month. A model optimising this morning’s readiness score will almost never choose it.

Zero marathon-pace segments appeared in three weeks. Not one. Race-aware suggestions do get more specific as the date closes in, which is precisely the issue: the specificity arrives during the sharpening phase, after the aerobic build it was supposed to support has already been under-prescribed.

Week two: the load ratio quietly took the week off my hands

Week two was meant to be the step-up. What happened instead:

51 km against a planned 64. The Wednesday threshold was good work, well-targeted, the right duration. Then Thursday’s card showed acute load at 561 and a load ratio of 1.3, flagged “High,” and it put me on 26 easy minutes. By Sunday the system had been nudging me back toward its optimal band all week, and the long run that was supposed to be the point of the week got shaved to 72 minutes.

Garmin’s Load Focus chart is the closest thing the platform has to structure, and it is a rear-view mirror: four weeks of accumulated low aerobic, high aerobic and anaerobic load with target ranges. It tells you what you did. It does not commit to what you will do, so it cannot deliberately spend three weeks digging a hole in order to climb out of it in the fourth. Supercompensation requires a plan that tolerates feeling bad on purpose. Readiness-driven suggestion is constitutionally incapable of that, and if you want the longer comparison of which coaching apps actually periodise versus which just react, that is the thesis of our head-to-head of the main AI coaching apps.

Week three fell into the readiness hole

Thursday 18 was a short-sleep night. Here is where that led, four days on:

MON 22 SEP · DAILY SUGGESTION
Primary benefit     RECOVERY
Duration            24:00
Target              HR < 136 bpm
Training readiness  31  (Low)
Sleep score         48   (4h 52m)
HRV status          Unbalanced (46 ms, 7-day avg 59 ms)
Recovery time       19:40 remaining

Fair enough for one day. The problem is the feedback loop. Low readiness produces an easy day, the easy day does not restore HRV (because sleep does that, not running), readiness stays low, and the watch offers another easy day. Week three came in at 37 km with three Recovery-labelled runs and two rest days. Training Status had flipped to Maintaining by the Friday, and the race predictor had drifted to 3:24:51.

A human coach reading those same numbers says: you slept badly on Thursday, you are not injured, go and do your long run on Sunday and get to bed earlier. The watch has no way to distinguish “under-recovered from training” from “had a bad week at work,” because both present as a depressed HRV baseline.

The pace targets move under you

On 10 September the threshold target was 4:14-4:20/km. On 20 September I ran a parkrun inside the week’s base allocation, the VO2max estimate ticked 52 → 53, and on 24 September the card read:

WED 24 SEP · DAILY SUGGESTION - THRESHOLD
Warm-up         10:00  easy
4 × 6:00        4:05-4:11 /km   (2:00 jog recovery)
Cool-down       10:00
Est. load       168

Nine seconds per kilometre faster, from a single 5K effort, with no change in my actual lactate threshold. I ran it as given. Rep four went 4:16 with a heart rate 6 bpm above the previous session’s average for the same work, which is the signature of a threshold session that has quietly become a VO2max session. Firstbeat’s VO2max estimate is a good population-level number and a noisy individual one, and anchoring tomorrow’s prescription to it means your interval paces wobble with GPS quality, wind, and whether you did a hilly route on Saturday.

What I’d actually keep

Three weeks, 132 km against a planned 178. Four quality sessions: two threshold, one VO2max, one anaerobic. Eleven base runs, three recovery runs, three rest days. No long run over 18.2 km, no goal-pace work, and a race predictor pointing the wrong way.

But the easy days were immaculate, and the morning-after recovery call was right every single time. That is the trade. Use Daily Suggested Workouts as a veto layer rather than a plan: keep your Runna block, your Garmin Coach plan or your ChatGPT spreadsheet as the skeleton, and let the watch’s readiness score decide whether Tuesday’s intervals happen today or Wednesday. When it tells you to take the day easy, it is usually right. When it tells you what to build toward over the next fourteen weeks, it isn’t telling you anything, because it hasn’t looked that far.

Next thing I’m testing: the same 21-day protocol with the race set four weeks out instead of fourteen, to find out whether the race-aware logic earns its name once the taper window opens.