Gameweek 7 Verdict: Our AI Called 7 of 10, Beating Two BBC Pundits but Not the BBC's AI

FootyPulse's AI called 7 of 10 Gameweek 7 results correctly, our best week of the season. It still finished second of the four, behind the BBC's AI, which got 9.

One gameweek, 10 matches. One result moves a score by 10 points.

That is ten matches, so a single result moves a score by ten points. Read it as one week's record, not a ranking. Our best previous weeks were 50%, and the season record now stands at 30 from 67 (44.8%). There have been 70 completed matches: three Gameweek 1 fixtures had no recorded prediction and are excluded, not counted as hits.

Every call, fixture by fixture

Show every miss. HT calls use in-match information. Not like-for-like with pre-match. BBC = BBC AI; S = Sutton; G = Guest. H/D/A = home/draw/away percentages.

Where we beat the pundits

Five of our seven hits came in fixtures where at least one BBC predictor missed:

• Brentford v Liverpool: we and the BBC's AI backed Liverpool. Sutton and the guest both went for a draw.
• Fulham v Hull and Man City v Ipswich: the guest called the away side in both, and we were right.
• Bournemouth v Sunderland: Sutton called a draw. We had the home win.
• Forest v Arsenal: the guest called a draw. We had Arsenal at 58%.

The BBC panel made five draw calls across the week, and none landed. Only one match was drawn (Leeds v Man United), and none of the four predictors had it.

Where we lost: two away wins we had down as home wins

We agreed with the BBC's AI in seven fixtures and were right in all seven. The two-point gap comes from two other fixtures:

• Everton v Chelsea. We had Everton at 48% and Chelsea at 23%. Chelsea won 2-0, and the BBC's AI and the guest both called it.
• Newcastle v Aston Villa. We had Newcastle at 38%, the draw at 34% and Villa at 28%. Villa won 1-0.

Our third miss, Leeds v Man United (48/29/23, drawn 1-1), was missed by all four predictors.

We have not diagnosed these three from the data we hold, so we are not going to offer a made-up reason. What the numbers do show is that all three were calls where our top probability was under 50%. Four away wins arrived in total. Our away-win probabilities add up to 2.8 expected away wins, so the model was slightly light on them this week.

The honest angle: what one gameweek can and can't tell us

• A good week, not yet a trend. Our own probabilities imply about 5.0 correct calls from these ten. We got 7. A week of 7 or better would turn up roughly one time in six if our numbers are right and match outcomes are independent, so this is well within normal variation.
• The BBC's AI had a strong week too. It called all four away wins and missed only the draw. Nine from ten is a high score for a ten-match sample, and we would not read it as a new baseline for them any more than 7 is one for us.
• Confidence lined up this week, on tiny numbers. Our three calls of 58% or more all landed (Man City, Brighton and Arsenal). Of the seven calls under 50%, four landed. Three matches is far too few to call that a pattern.
• The season record is still modest. 30 from 67 is 44.8%. We will publish the head-to-head season table on the AI vs BBC page once all weeks are scored there.

Single gameweeks are 7–10 matches. Week-to-week swings are normal noise.

Three matches in one band. Too small to call a pattern.

Half-time re-predictions: 8 from 10, a one-match lift

Our half-time re-predictions went 8 from 10 on the outcome, against 7 from 10 pre-match. The whole difference is one match: at Newcastle v Villa we were wrong pre-match, and the half-time call (a Villa win) was right. Everton v Chelsea and Leeds v Man United were wrong both times.

Two re-predictions matched the final score exactly, Brighton 5-0 Palace and Brentford 1-2 Liverpool. Brighton's call was recorded about 50 minutes after kick-off; Brentford's about 60 minutes after kick-off. Both meet the site's half-time lock rule. We would not read much into two exact scores: a re-prediction uses information from part of the game already played, which makes it a different task from calling it before kick-off. Across the season so far, 43 right from 67 at half-time compares with 30 from 67 pre-match, with both records scored on the same 67 eligible fixtures. The live accuracy page also notes one excluded, unlocked re-prediction. The two tasks are not like-for-like because the half-time call has in-match information. Our view is that it is too early to tell how much the re-prediction adds. Half-time re-predictions are a FootyPulse Pro feature.

What's next

Gameweek 8 starts on Friday 23 October. Our predictions for it will be on the fixtures page, and we will run the same scorecard after the final game. Track the running record on our accuracy page.

BBC predictions are factual attributions of publicly published forecasts. Source: BBC Sport's Gameweek 7 predictions: Chris Sutton v Fabio Wardley and AI, published 15 October 2026. We score the 90-minute win/draw/loss outcome, not the BBC's exact-score points system.

Club analysis

Everton · Chelsea · Newcastle · Aston Villa · Leeds · Man United · Arsenal · Nottingham Forest

FAQ

How accurate were the Gameweek 7 predictions?

FootyPulse's AI got 7 from 10 (70%). The BBC's AI got 9, Chris Sutton 6 and the BBC guest 4.

Did the AI beat Chris Sutton in Gameweek 7?

Yes, 7 to 6, a margin of one match. One gameweek is too small a sample to say who is better.

Which Gameweek 7 results did the AI get wrong?

Everton v Chelsea, Newcastle v Aston Villa and Leeds v Manchester United.

Do half-time re-predictions improve accuracy?

In Gameweek 7 they went 8 from 10 against 7 from 10 pre-match, a one-match difference. They use information from after kick-off, so the comparison is not like-for-like, and one week is too small to establish a pattern. The full scored record stays public; live half-time re-predictions are a FootyPulse Pro feature.