Gameweek 6 Verdict: Our AI Beat Sutton and the Guests, but the BBC's AI Won by One
Published 13 October 2026 · FootyPulse AI Analysis
6 min read
FootyPulse's AI called 5 of 10 Gameweek 6 results correctly. That put it ahead of Chris Sutton (3) and the BBC's guest panel (2), and one behind the BBC's own AI (6). It is the first week we have finished ahead of two of the three BBC sources at once. It is also a ten-match sample, which is why we are going to be careful about what it proves.
Gameweek 6 scoreboard: who called the most results?
Scoring is the same as on our AI vs Pundits leaderboard: one point for each correct 90-minute result (home win, draw or away win).
| Source | Correct | Out of |
|---|---|---|
| The BBC's AI | 6 | 10 |
| FootyPulse AI | 5 | 10 |
| Chris Sutton | 3 | 10 |
| BBC guests | 2 | 10 |
The BBC's calls come from BBC Sport's Gameweek 6 predictions. Every fixture was played and scored, and every source made a call for every match.
Every call, hit and miss
| Fixture | Result | FootyPulse | BBC's AI | Sutton | Guests |
|---|---|---|---|---|---|
| Arsenal v Leeds United | 2–1 | ✅ Arsenal | ✅ Arsenal | ✅ Arsenal | ❌ Draw |
| Aston Villa v Brentford | 2–2 | ❌ Brentford | ✅ Draw | ❌ Villa | ❌ Villa |
| Chelsea v Bournemouth | 5–1 | ✅ Chelsea | ✅ Chelsea | ✅ Chelsea | ✅ Chelsea |
| Ipswich Town v Fulham | 2–1 | ✅ Ipswich | ❌ Draw | ❌ Draw | ❌ Fulham |
| Sunderland v Brighton | 0–2 | ✅ Brighton | ✅ Brighton | ❌ Draw | ❌ Draw |
| Manchester United v Tottenham | 1–1 | ❌ Man Utd | ❌ Man Utd | ❌ Man Utd | ❌ Man Utd |
| Crystal Palace v Nottingham Forest | 3–1 | ❌ Forest | ❌ Draw | ❌ Forest | ✅ Palace |
| Hull City v Everton | 2–2 | ❌ Hull | ✅ Draw | ✅ Draw | ❌ Everton |
| Liverpool v Manchester City | 0–3 | ✅ Man City | ✅ Man City | ❌ Draw | ❌ Liverpool |
| Coventry City v Newcastle United | 4–1 | ❌ Newcastle | ❌ Newcastle | ❌ Draw | ❌ Draw |
Two results beat everyone. All four sources backed Manchester United against Tottenham, and all four missed Coventry's 4–1 win over Newcastle.
Where the one-point gap came from
The BBC's AI and ours made the same call in six of the ten matches, and scored the same in them: four right, two wrong. Both backed Arsenal, Chelsea, Brighton and Manchester City correctly, and both got Manchester United and Newcastle wrong.
The whole difference sits in the four games where the BBC's AI chose a draw: Aston Villa v Brentford, Ipswich v Fulham, Crystal Palace v Forest and Hull v Everton. It was right in two (Villa–Brentford and Hull–Everton), we were right in one (Ipswich), and both were wrong in Palace v Forest.
That is a one-match swing across four games. It is not evidence that either model reads football better. It does show where the real divergence lives, which is draws.
Three draws, and a model that rarely calls one
Three of the ten results were draws: Aston Villa 2–2 Brentford, Manchester United 1–1 Tottenham and Hull 2–2 Everton. FootyPulse's AI named no draws, as we said in our Gameweek 6 preview, so none of those three could have counted for us.
The probabilities were not wildly off. Our draw probability across the ten matches averaged about 27%, and 30% of the results ended level. The problem is the format: we publish the single most likely outcome, and a draw is rarely the single most likely one. Across the season, 19 of 60 matches (32%) have ended level, and our AI has made a draw its call once in 57 recorded predictions.
Calling draws is not a free fix. The BBC panel named 12 draws in total in Gameweek 6 (the BBC's AI four, Sutton five, the guests three) and landed three of them: two for the BBC's AI, one for Sutton and none for the guests. Draw-hedging earned the BBC's AI its lead and left Sutton and the guests behind us.
Take the draws out and the picture is kinder to us. We called 5 of the 7 non-drawn results correctly. The two we missed were Crystal Palace beating Nottingham Forest and Coventry beating Newcastle.
The two misses we own
Coventry 4–1 Newcastle is the clearest failure. We forecast Newcastle at 58%, our joint-highest confidence of the week, and gave Coventry an 18% chance. According to our match-event data, Coventry led at half-time, and our half-time re-prediction still leaned towards Newcastle.
Crystal Palace 3–1 Nottingham Forest was a lower-confidence call. We forecast Forest at 38% and gave Palace 28%, but a call with 38% probability is wrong more often than right. Two of our five 58% calls failed this week (Newcastle and Manchester United). The other three, Arsenal, Ipswich and Brighton, came in. Five matches is far too few to say anything about whether our confidence levels are well calibrated.
The season picture: still behind the BBC's AI and Sutton
Three Gameweek 1 fixtures (Everton v Crystal Palace, Ipswich v Sunderland and Nottingham Forest v Leeds) have no recorded FootyPulse prediction. We have left them out of every comparison below rather than estimate them. That leaves 57 matches that all four sources predicted.
| Source | Correct | Out of 57 | Accuracy |
|---|---|---|---|
| The BBC's AI | 29 | 57 | 50.9% |
| Chris Sutton | 27 | 57 | 47.4% |
| FootyPulse AI | 23 | 57 | 40.4% |
| BBC guests | 22 | 57 | 38.6% |
We are ahead of the guests and behind the BBC's AI and Sutton. In 18 weekly head-to-heads (three sources over six gameweeks) we have won 4, tied 4 and lost 10. Gameweek 6, with two wins, is the best return so far, and Gameweek 4, when we scored 2 from 10, remains the worst. Our Gameweek 5 review looked at the confidence calls.
What the half-time numbers can and cannot tell us
Our half-time re-predictions got 6 of 10 right in Gameweek 6, against 5 of 10 before kick-off. Over the season the figures are 35 of 57 (61%) at half-time and 23 of 57 (40%) pre-match.
Those two numbers are not like for like, because the half-time version has seen 45 minutes of play. A basic rule of "whoever leads at the break wins, and level means a draw" gets 32 of 57 on our match-event data. So most of the improvement comes from knowing the score (how the half-time re-prediction works). The model's contribution above that rule is three matches in 57, which is well within the range of chance.
What happens next
Gameweek 6 was a better week for the model than most, and the season record is still below the BBC's AI and Sutton. Every call and result is on the accuracy page, and the full leaderboard is on AI vs Pundits. We will keep publishing the misses alongside the hits.
FAQ
How accurate was FootyPulse's AI in Gameweek 6?
FootyPulse's AI called 5 of 10 Premier League results correctly in Gameweek 6 (50%). Across the 2026/27 season it has called 23 of 57 recorded predictions correctly (40.4%).
Who was most accurate in Gameweek 6: the AI or the BBC's pundits?
The BBC's AI was most accurate with 6 of 10. FootyPulse's AI scored 5, Chris Sutton 3 and the BBC's guest panel 2.
Does FootyPulse's AI predict draws?
It publishes the single most likely outcome for each match, and a draw is rarely the most likely one. It has named a draw as its call once in 57 recorded predictions, while 19 of 60 matches this season have ended level.
Why does the season comparison use 57 matches rather than 60?
Three Gameweek 1 fixtures have no recorded FootyPulse prediction. We excluded them from every source's total rather than estimate a result.
Share this article: