Premier League · Gameweek 4 Verdict
Gameweek 4 Verdict: 2 from 10 as upsets dominate the weekend
Published 15 September 2026 · FootyPulse AI Analysis
The AI managed just 2 correct calls from 10 fixtures in Gameweek 4, marking a difficult weekend where favourites stumbled across the division. Season accuracy now stands at 35% (13 from 37), as unpredictability continues to define the early campaign.
The scoreboard
Gameweek 4 delivered a sobering 2 from 10 correct predictions, dragging the season accuracy to 35% after 37 fixtures. Four draws and multiple away upsets exposed the model's limitations when form lines blur and underdogs find their rhythm.
Biggest upset
Liverpool 0-0 Fulham stood out as the weekend's most surprising result. The model gave the hosts a commanding 68% win probability, expecting Anfield to prove the difference against promoted opposition. Instead, Fulham held firm for a goalless draw, overturning the model's strongest conviction of the round.
Best correct call
Brighton's 5-0 demolition of Coventry City validated the model's least confident correct prediction. With just 42% backing for the away win, this was effectively a coin-flip call that the data marginally favoured. Brighton's clinical performance turned uncertainty into emphatic vindication, though the model can claim only modest credit given the narrow pre-match edge.
Every fixture
- Crystal Palace 2-3 Ipswich Town – Predicted home win (48%), actual away win ❌
- Bournemouth 2-2 Brentford – Predicted away win (38%), actual draw ❌
- Liverpool 0-0 Fulham – Predicted home win (68%), actual draw ❌
- Chelsea 2-2 Hull City – Predicted home win (68%), actual draw ❌
- Aston Villa 1-2 Nottingham Forest – Predicted home win (48%), actual away win ❌
- Tottenham Hotspur 0-0 Everton – Predicted home win (38%), actual draw ❌
- Sunderland 0-2 Arsenal – Predicted away win (55%), actual away win ✅
- Coventry City 0-5 Brighton & Hove Albion – Predicted away win (42%), actual away win ✅
- Manchester United 0-1 Manchester City – Predicted home win (38%), actual away win ❌
- Leeds United 4-1 Newcastle United – Predicted away win (38%), actual home win ❌
Gameweek 4 highlighted the Premier League's resistance to pattern recognition in its opening phase. Six fixtures produced draws or away wins where the model favoured the home side, suggesting the data hasn't yet captured the nuances of current form. See the pre-match preview for how these probabilities were framed before kick-off. Recalibration ahead of Gameweek 5 will focus on refining home advantage weighting and incorporating early-season momentum shifts.
Half-time re-predictions this gameweek
At half-time the AI re-reads each match and re-predicts. Here is how those calls did against the pre-match calls, on the same set of Gameweek 4 matches.
Pre-match
2/10
20% correct
At half-time
6/10
60% correct
Where half-time changed the call
Crystal Palace 2–3 Ipswich Town
Pre-match: Crystal Palace ✗ · Half-time: Ipswich Town (1–4) ✓
Why the call changed
- Red card at 34' — Disasi
- Shots on target 2 v 2
- Possession 45% v 55%
Chelsea 2–2 Hull City
Pre-match: Chelsea ✗ · Half-time: Draw (3–3) ✓
Why the call changed
- Shots on target 2 v 2
- Possession 66% v 34%
Bournemouth 2–2 Brentford
Pre-match: Brentford ✗ · Half-time: Draw (2–2) ✓
Why the call changed
- Shots on target 2 v 1
- Possession 57% v 43%
Leeds United 4–1 Newcastle United
Pre-match: Newcastle United ✗ · Half-time: Leeds United (4–0) ✓
Why the call changed
- Shots on target 5 v 1
- Possession 59% v 41%
This is a single gameweek — far too small a sample to read anything into, in either direction. See the running pre-match vs half-time comparison on our accuracy page for the season picture.
Share this verdict:
Prediction figures in this article reflect the data available at the time of publication and may have shifted since. See the live Gameweek 4 fixtures for the current numbers on each match.
See how the AI called it before kickoff:
Read the Gameweek 4 preview →