In August we published an analysis asking whether players on OnCourtAI actually improve. We have now re-run it on a substantially larger dataset — 2,602 analyses from 891 players across 93 countries, covering 16,872 individual shots.

One of our findings got stronger. One did not survive. Both are below, because publishing only the half that flatters us would make the first half worthless.

The Finding That Held

The relationship between how much you use the platform and whether you improve is clear, and it has strengthened with more data.

Taking every player with at least three analyses and comparing their earliest sessions against their most recent:

  1. 3 to 4 sessions: 50% improved — a coin flip
  2. 5 to 9 sessions: 56% improved
  3. 10 to 19 sessions: 68% improved
  4. 20 or more sessions: 73% improved

The average gain follows the same curve. Players in the 20+ group gained roughly seven times as many points as players in the 3-4 group.

Grouping more simply: players with ten or more sessions improved 70% of the time, against 52% for players with fewer. That gap is the single most reliable thing we know about how people use this platform.

We Tried Hard to Break It

There is an obvious way this could be an illusion. If scores drift upward for any reason — noise, seasonal form, players warming up to the camera — then anyone with more sessions would show more drift, and it would look like improvement without anything real happening.

So we tested it the direct way. We took every player's set of scores, randomly shuffled the order, and re-ran the entire analysis. If the effect were just an artefact of having more data points, shuffled data would reproduce it.

It did not. On shuffled data the heavy-use group improved 50% of the time, with an average change of almost exactly zero. We ran this 500 times; only 7 of those 500 shuffles produced a result as strong as the real one, giving a probability of around 1.4% that this is chance.

The ordering of sessions matters. These players genuinely got better, in sequence.

The Finding That Did Not Replicate

In our September article we reported that spacing your videos mattered more than volume — that filming across several separate weeks beat cramming videos into one afternoon.

On the larger dataset, that does not hold up, and we are correcting it.

Grouping players by the typical gap between their uploads now gives an almost opposite picture:

  1. Same-day filming: 57% improved
  2. 1 to 6 day gaps: 57% improved
  3. 1 to 4 week gaps: 50% improved
  4. Gaps over a month: 39% improved

The players with the longest gaps now show the worst outcomes. The tidy "spacing beats cramming" story was, on reflection, a small-sample result that we should have hedged harder at the time.

What we think is happening is simpler than a motor-learning story. A long gap between uploads is mostly a marker of disengagement, not of patient deliberate practice. Someone filming twice a year is not carefully letting changes bed in — they have largely stopped, and they come back a little rustier each time.

What Actually Survives: Volume, Not Timing

Here is the clean test. We split players by whether they filmed everything on single days or spread uploads across different days, then looked at volume within each group:

  1. Same-day filers, 3-9 sessions: 54% improved
  2. Same-day filers, 10+ sessions: 70% improved
  3. Spread filers, 3-9 sessions: 49% improved
  4. Spread filers, 10+ sessions: 71% improved

The volume effect is almost identical in both groups — about 70% either way. Spacing barely moves the needle once you control for how much someone uses the platform.

So the honest advice is simpler than what we said in September: keep going. Whether you film three strokes in one evening or across three weeks matters far less than whether you are still filming in month four.

Why Showing Up Beats Being Clever

None of this should be surprising, and it mirrors what the underlying improvement data showed about which faults respond to video.

Video feedback is excellent at faults you can correct once they are pointed out — where you meet the ball, where the racket finishes. Those corrections are not single events. You see the fault, you work on it, you drift back, you see it again, you work on it again. That cycle needs iterations, and iterations are exactly what the ten-session threshold is measuring.

A player who films three times sees three snapshots. A player who films fifteen times has run the loop repeatedly, which is how technique actually changes.

The Caveat We Will Keep Repeating

This data cannot separate two things, and we would rather say so than let you find it later.

Players who film twenty sessions are, on average, more committed players. They are more likely to be taking lessons, playing more matches and practising deliberately. We cannot fully untangle "filming consistently causes improvement" from "people who improve are the sort of people who film consistently."

What the shuffle test establishes is narrower but still worth having: the improvement is real and ordered in time, rather than a statistical mirage from accumulating readings. Whether the camera caused the gain or faithfully recorded it, consistent measurement reliably marks out players who are getting better.

Our honest reading is that it is both. The feedback helps, and the habit of measuring keeps you honest about whether the work is landing.

What We Would Tell You To Do

Film whenever you get the chance, and do not agonise about timing. Pick one component to work on and ignore the rest until the next video. Film the same stroke from the same angle so comparisons mean something — our guide on filming for AI analysis covers the setup.

And get past the tenth session. Almost everyone who stops, stops before that point, which is precisely where the data says the returns begin.

Half the players on this platform have filmed once or twice. On our numbers that is a coin flip. Fifteen sessions is not.