He's done everything by the book. Race morning, 200 mg of caffeine an hour out — roughly 3 mg/kg for a 70 kg rider, the exact number every fuelling article on the internet agrees on. He's done it in training, felt sharp, felt nothing to worry about. Today it's a club crit he's circled for months. He warms up feeling jittery in a way that doesn't quite read as "ready" — more like a low hum of unease sitting under his legs. The gun goes. He can't find the top end. Not sore, not empty — just flat, like someone quietly turned his dial down two notches before he even clipped in.

He rides home convinced it was the taper, or the course, or bad legs. It might have been none of those. There's a real, published possibility that the exact dose he trusted was working against him — not because he did it wrong, but because of three letters in his DNA that decide whether caffeine sharpens a rider or blunts him.

Section 01What the data actually shows

Guest and colleagues ran the test that most caffeine advice quietly assumes never needs running (Guest et al., 2018). 101 competitive male athletes completed three 10-km cycling time trials — one on placebo, one on 2 mg/kg caffeine, one on 4 mg/kg — and every rider was genotyped for CYP1A2, the gene that governs how fast the liver clears caffeine. The rs762551 variant sorts people into three groups: AA (fast metabolisers), AC (heterozygous), and CC (slow metabolisers).

Averaged across the whole group, 4 mg/kg caffeine improved 10-km time by about 3% versus placebo (Guest et al., 2018) — the kind of number that gets printed on a supplement label and called a day. It's also the number that hides the story. Split by genotype, the picture looks nothing like a uniform 3% win.

AA homozygotes — the fast metabolisers — improved by 6.8% at 4 mg/kg (P < 0.0001), and by 4.8% even at the lower 2 mg/kg dose (P = 0.0005) (Guest et al., 2018). AC heterozygotes got nothing detectable at either dose (Guest et al., 2018). CC homozygotes — the slow metabolisers — didn't just fail to improve. At 4 mg/kg, they were 13.7% slower than on placebo (P = 0.04) (Guest et al., 2018). Same protocol, same dose, same "textbook" pre-race routine. Three different races.

Womack and colleagues ran a related test over 40 km, at a higher 6 mg/kg dose, splitting riders into AA versus C-allele carriers (AC and CC combined) rather than three separate groups (Womack et al., 2012). AA homozygotes improved by 4.9%; C-allele carriers still improved, just by a much smaller 1.8% — and the treatment-by-genotype interaction was statistically significant (P = 0.005) (Womack et al., 2012). Womack's carriers didn't cross into "slower than placebo" the way Guest's CC group did at 4 mg/kg — but the direction of the story is the same one both papers are telling: genotype decides how much of the advertised benefit you actually get, and at the sharper end of that spectrum, the benefit can flip into a cost.

Section 02Why the same pill helps one rider and hurts another

CYP1A2 is the liver enzyme responsible for clearing roughly 95% of the caffeine you drink. The rs762551 variant sits in a regulatory region that controls how much of that enzyme your liver actually makes. One copy of the A allele and you build the fast version — caffeine gets broken down quickly, cleared before it lingers into anything unhelpful. Carry the C allele instead, and CYP1A2 activity drops — caffeine, and the compounds it breaks down into, stay active in your system for longer.

That's the mechanism the "one dose fits all" advice skips over entirely. For a fast metaboliser, a 3-4 mg/kg dose does its job — mild central-nervous-system stimulation, perceived-exertion dulling, a genuine ergogenic bump — and clears out before it becomes a liability. For a slow metaboliser, the same dose sticks around longer, at higher effective exposure, for the full length of the effort and beyond. Guest's CC group wasn't simply failing to respond. At the higher dose specifically, they were measurably worse than they'd have been on nothing at all — consistent with caffeine tipping past a useful stimulant dose and into a jittery, poorly-regulated one for a rider whose body can't clear it at the rate the "standard" protocol assumes.

It's worth being precise about what "slower" looks like from the saddle, because it isn't the same failure mode as bad legs or an empty tank. Riders in this data weren't running out of fuel — they were running an effort with an unhelpful level of stimulation sitting on top of it: harder to settle into rhythm, harder to hold a smooth cadence, more of the ride spent managing a wired, unsettled feeling than producing power. That's a genuinely different problem to solve than "eat more" or "rest more," and it's one most riders never think to test for because the pre-race caffeine dose is the one part of the routine nobody questions.

This genotype isn't a rare curiosity either. In the population Sachse and colleagues studied, roughly 1 in 10 carried the CC (slow-metaboliser) genotype (Sachse et al., 1999) — common enough that on any given club run, a handful of riders around you are quietly running a genotype the "standard" caffeine protocol was never built to help.

Section 03What the data doesn't yet answer

Guest's cohort was 101 competitive male athletes — the response magnitude in women hasn't been characterised to the same depth, and it would be a mistake to assume the exact percentages transfer unchanged (Guest et al., 2018). Womack's trial ran at a higher 6 mg/kg dose than Guest's top dose of 4 mg/kg, and grouped C-allele carriers together rather than separating AC from CC (Womack et al., 2012) — so it can't tell you whether a heterozygote gets closer to the AA result or the CC one at that higher dose. Habitual caffeine intake matters too: someone who drinks three coffees a day before a race is running a different experiment than someone who's caffeine-naive on race morning, and neither trial isolated that variable.

None of that changes the core finding. It does mean the honest version of this article isn't "avoid caffeine if you're a CC genotype" — it's "the standard advice was tested on a mixed population and it doesn't apply evenly, so test your own response before you trust the label." A slow metaboliser who's used caffeine for years, at a lower dose, timed earlier, may see a completely different result than Guest's CC group saw at 4 mg/kg cold.

Section 04The protocol that actually respects your genotype

1. Start lower than the label suggests

   Guest's data shows a real benefit at 2 mg/kg in fast metabolisers — you don't need to jump straight to 4-6 mg/kg to find out if caffeine works for you.
2. Run your own matched test, not a race-day guess

   Two comparable efforts — same course or trainer session, same warm-up — one with your normal caffeine dose, one without. Compare how the effort actually went, not just how you felt walking in.
3. Treat "jittery but not sharp" as real data

   A wired, unsettled feeling that doesn't translate into a faster effort isn't just nerves — in this data, it's what a slow-metaboliser response can look like from the inside.
4. Never trial a new dose on the day that matters

   If you haven't tested a caffeine protocol in training, race day is the wrong place to find out which genotype you are.
5. Time it with the whole day in mind

   Slower clearance means a race-morning dose lingers longer into the hours after you finish — factor that into how late in the day you're comfortable dosing.

Section 05Where riders get this wrong

Section 06Applying it with THRIVE

You can't genotype yourself from a training log, but you can build the evidence a genotype test would give you. Marco, THRIVE's AI coach, and the readiness layer in the app already track how a ride actually went against your rolling baseline — so if you're trialling a new caffeine dose, that's not a memory you're relying on, it's a data point sitting next to the day's readiness and effort notes. Run the test twice, look at the trend, and you'll know a lot faster than the label ever will whether your version of the "standard" protocol is helping you or quietly working against you.

Section 07Bottom line

The 3-6 mg/kg caffeine advice printed everywhere is real for a lot of riders — and actively wrong for some of them. Guest et al. (2018) found fast metabolisers (AA) gained a genuine 6.8% at 4 mg/kg, heterozygotes (AC) gained nothing detectable, and slow metabolisers (CC) were 13.7% slower than on placebo at that same dose. Womack et al. (2012) found the same genotype-gated pattern over 40 km. If your best efforts feel worse after your usual pre-race dose, that's not necessarily bad legs — run your own matched test before you rule it out.