Golf’s problem is not a shortage of data. It is that almost none of it accumulates into a picture of a player. Stand behind a launch monitor for an hour and you will watch something strange. A player hits 60 shots. Each one produces a dozen numbers, displayed instantly, to two decimal places. At the end of the hour the screen is cleared. The numbers are gone, or they are in a session file nobody will open, and what survives is whatever the player and the professional remember about how it felt.
Golf has more measurement per shot than almost any sport and less continuity than most. That is the gap this paper is about, and it is not closed by adding a 13th number.
Key findings at a glance
- The Rules already govern this. Data may be recorded during a round for use afterwards. Processing or interpreting information from the current round is not allowed, and using physiological data recorded during the round is prohibited outright.
- Spin rate is not a basis for a decision. Between-session reliability for spin ranged from 0.02 to 0.60 with a standard error of 241 to 455 revolutions per minute, while clubhead and ball speed were highly reliable.
- Devices are not interchangeable. Against a reference monitor, a consumer device showed a mean spin bias of up to 1,238 revolutions per minute with an interval from minus 2,628 to plus 5,103.
- Flexibility and balance show no association with clubhead speed. In meta-analysis, jump impulse was the strongest correlate at 0.82, while flexibility and balance were indistinguishable from zero.
- Putting is not a stroke problem. Face angle accounts for 80 per cent of stroke direction consistency, and a tour player’s stroke variability is good enough to hole about 95 per cent of four-metre putts. In tournaments they hole 17 per cent.
- There is no systematic review of AI in golf. None exists. Nor is there peer-reviewed validation of any consumer launch monitor, of any green-reading method, or of dispersion-based strategy as a coaching intervention.
Continuity rather than diagnosis
The useful contribution of artificial intelligence in golf is memory. A system earns its place by remembering what a player was working on and why, by comparing them against their own history rather than a model swing, by stating the error on the numbers it reports, and by respecting a line the Rules of Golf have already drawn.
This paper set out expecting to be about swing analysis and computer vision, which is where the market’s attention sits. The evidence sent it somewhere else, towards the far duller question of what gets written down at the end of a practice session.
The numbers most confidently displayed have the weakest support
Spin rate is the clearest case. It appears on every launch monitor screen to the nearest revolution, and its between-session reliability ranged from 0.02 to 0.60, with a standard error of 241 to 455 revolutions per minute. Clubhead and ball speed, displayed with identical confidence, are highly reliable.
Devices compound this. Against a reference monitor, a consumer device showed a mean spin bias of up to 1,238 revolutions per minute, with an interval running from minus 2,628 to plus 5,103. A player who changes device has changed their own history, and nothing on the screen says so.
Where effort is misallocated
Two examples stand out. In the gym, flexibility and balance work is widely prescribed for distance, and in meta-analysis both were indistinguishable from zero as correlates of clubhead speed, while jump impulse was the strongest at 0.82.
On the green, the gap between 95 per cent of four-metre putts holed in testing and 17 per cent in tournaments is not explained by stroke mechanics, since face angle already accounts for 80 per cent of direction consistency. The variance lives in green reading and aim, which is not where most putting practice goes.
What follows for a coach, an academy or a union
Keep the session, not the screen. Record every practice session with its intention, because a launch monitor number without an intention cannot be interpreted six months later, and most of them are not. Compare a player against their own history. Report error alongside every number that carries one. Measure rounds available, which is countable and which the injury evidence supports.
Refuse four things. No in-round processing or interpretation of data from the round in progress, which the Rules prohibit. No use of physiological data recorded during a round, which they prohibit outright. No model swing presented as a target. And no claim that a swing characteristic causes injury, which no review this paper could verify supports.
Free download
Get the full paper
Read The Augmented Golfer in full, with every claim carrying an evidence grade and the full reference list. Give your name and email and the PDF opens straight away.
Frequently asked questions
Do the Rules of Golf allow AI or data analysis during a round?
Only within limits, and the limits are tighter than most products assume. Data may be recorded during a round for use afterwards. Processing or interpreting information from the current round is not allowed, and using physiological data recorded during the round is prohibited outright. Nothing here is a ruling, and any question of a player’s compliance in competition is a matter for the relevant Committee and for qualified advisers.
How accurate are consumer launch monitors?
Accurate for some numbers and poor for others, without saying which. Clubhead and ball speed are highly reliable. Spin is not, with between-session reliability from 0.02 to 0.60 and a standard error of 241 to 455 revolutions per minute. Against a reference monitor, a consumer device showed a mean spin bias of up to 1,238 revolutions per minute with an interval from minus 2,628 to plus 5,103. There is no peer-reviewed validation of any consumer launch monitor.
What actually increases clubhead speed?
In meta-analysis, jump impulse was the strongest correlate at 0.82. Flexibility and balance were indistinguishable from zero. That is an uncomfortable result for a great deal of what is prescribed in golf-specific gym programmes, and it points training towards force production rather than mobility work.
Why do tour players hole so few four-metre putts?
Because putting at that distance is not primarily a stroke problem. A tour player’s stroke variability is good enough to hole about 95 per cent of four-metre putts in testing, and in tournaments they hole 17 per cent. Face angle already accounts for 80 per cent of stroke direction consistency. The missing variance sits in green reading and aim rather than in the stroke.
Is there evidence for AI improving golf coaching?
Not yet in any systematic form. No systematic review of AI in golf exists, and there is no peer-reviewed validation of any consumer launch monitor, of any green-reading method, or of dispersion-based strategy as a coaching intervention. That does not mean the tools are useless. It means claims about them should be read as marketing until the evidence catches up.
About this series
This is the last of seven papers on the responsible use of artificial intelligence in performance sport. Golf closes the series because it is the sport with the most measurement and the least continuity, and because its own rules have already answered a question the rest of sport is still arguing about. The governance framework is set out in The Responsible Performance System. The series also covers elite athletics, grassroots athletics, talent identification in football, triathlon and elite football.
Legal, regulatory and rules notice. This article and the paper it summarises describe the Rules of Golf, data protection legislation and regulatory guidance. Nothing here constitutes legal advice or a ruling on the Rules of Golf. The provision of legal advice sits outside the terms of any engagement with the author or with Athleet.AI, and any question of a player’s compliance in competition is a matter for the relevant Committee and for qualified advisers. Every player, coach and academy used as an illustration is synthetic.