Eye2.AI asks several popular AI models the same question before a big game or event, then shows what they say. People often ask the obvious next question: are they any good? This post answers it with our own numbers, including the parts that do not look good for us.
How we counted
- Top pick. The option with the highest average win chance across the AI models that answered. It is the tallest bar on each Eye2.AI page.
- Hit. The top pick was the actual winner. We checked results against public sources.
- “Other” counts as a pick. On golf and some F1 pages, “other player” or “other driver” is an option. When it was the top pick, we counted it, and we mark those hits with an asterisk.
- One exclusion. The first game of the Yankees at Rays series listed the wrong teams on its page, so we could not score it.
- Data window. As of October 12, 2026, we were tracking 71 queries. 21 had a result and 20 could be scored. The first result is from August 23 and the latest from October 11.
Every result
Here is each event we scored. The tiles at the top of the page show the same 20 results at a glance.
| Event | Date | AI top pick | Actual winner | Result |
|---|---|---|---|---|
| Dota 2 The International | Aug 23 | Team Spirit 49.8% | Team Spirit | Hit |
| US Open, women's singles | Sep 12 | Aryna Sabalenka 30.2% | Elena Rybakina | Miss |
| Spanish Grand Prix | Sep 13 | Max Verstappen 22.0% | Andrea Kimi Antonelli | Miss |
| US Open, men's singles | Sep 13 | Carlos Alcaraz 44.4% | Alexander Zverev | Miss |
| Man City vs Sunderland | Sep 20 | Manchester City win 78.4% | Manchester City win | Hit |
| NFL Week 2, Colts at Chiefs | Sep 20 | Kansas City Chiefs win 64.0% | Kansas City Chiefs win | Hit |
| UFC 331, Van vs Pantoja 2 | Sep 20 | Joshua Van wins 57.0% | Joshua Van wins | Hit |
| NFL Week 3, Rams at Broncos | Sep 26 | Los Angeles Rams win 51.2% | Denver Broncos win | Miss |
| Azerbaijan Grand Prix | Sep 26 | George Russell 34.6% | George Russell | Hit |
| Laver Cup | Sep 27 | Team Europe 58.8% | Team Europe | Hit |
| Presidents Cup | Sep 27 | United States Team 52.4% | United States Team | Hit |
| NFL Week 4, Lions at Panthers | Oct 4 | Detroit Lions 60.1% | Carolina Panthers | Miss |
| Bank of Utah Championship | Oct 4 | Other player* 58.6% | Other player | Hit |
| NHL, Capitals vs Penguins | Oct 7 | Washington Capitals 49.5% | Washington Capitals | Hit |
| MLB ALDS, Yankees vs Rays | Oct 8 | Tampa Bay Rays 79.3% | Tampa Bay Rays | Hit |
| MLB ALDS, White Sox vs Guardians | Oct 8 | Chicago White Sox 70.2% | Cleveland Guardians | Miss |
| UFC Fight Night, Allen vs Duncan | Oct 10 | Brendan Allen 55.6% | Brendan Allen | Hit |
| Baycurrent Classic | Oct 11 | Other player* 60.7% | Other player | Hit |
| NFL Week 5, Ravens at Falcons | Oct 11 | Atlanta Falcons 56.7% | Baltimore Ravens | Miss |
| Singapore Grand Prix | Oct 11 | Other driver* 38.3% | Other driver | Hit |
Source: Eye2.AI prediction review, data as of October 12, 2026. Percentages are the average win chance across the AI models that answered. * The top pick was “other player” or “other driver”. Download the data (CSV).
What stood out
A high-confidence pick was not more likely to be right
Picks of 60% or higher were right 4 of 6 times. Picks below 60% were right 9 of 14 times. That is a small difference on small numbers, so we should not read a bigger percentage as a safer prediction.
Three hits came from “other”
The Bank of Utah Championship, the Baycurrent Classic and the Singapore Grand Prix were won by players or drivers outside our contender list. “Other” was the top pick each time, so it counted as a hit. Without them the record is 10 of 17, or 59%. We show both numbers because both are true.
Some sports were harder, but the groups are tiny
NFL games were the hardest, with the top pick winning 1 of 4. UFC went 2 of 2. Golf also went 2 of 2, though both were “other” picks. With groups this small, one game changes the picture, so we read this as a place to look, not a conclusion.
The misses
Seven events went against the top pick. Three stand out.
The most confident miss was the White Sox against the Guardians. The top pick had Chicago at 70.2%. Cleveland won, after being rated at 29.8%.
The one where every AI model agreed and was wrong was the Lions at the Panthers. All six picked Detroit, the average was 60.1%, and Carolina won after being rated at 36.4%.
The closest was the Spanish Grand Prix. Max Verstappen was the top pick at 22.0%, and Andrea Kimi Antonelli, who won, was at 21.6%. A 0.4 point gap in a race with many drivers is close to a coin toss.
How each AI did
We also looked at how often each AI’s own top option won. The counts are not equal, because not every AI answered every question, and ties count as half a hit. Read it as a rough picture, not a ranking.
| AI | Events | Hits | Hit rate | Note | |
|---|---|---|---|---|---|
| Grok AI | 20 | 13 | 65% | ||
| ChatGPT | 20 | 12.5 | 62% | ||
| Gemini | 20 | 12 | 60% | ||
| Claude | 19 | 9 | 47% | Declined the Laver Cup question | |
| Smart | 9 | 6 | 67% | Answered only 9 events | |
| Amazon Nova | 19 | 5.5 | 29% | Gave 10 to 15% to a tie or postponement on NFL games | |
Source: Eye2.AI prediction review, data as of October 12, 2026. “Events” is how many events that AI answered. Half hits are ties.
Some answers were extreme. Claude gave 100% to one golfer who did not win, and Claude and Grok both gave 100% to Team Spirit at The International, who did. A 100% answer leaves no room for being wrong, which is why we do not treat any single AI’s number as the answer.
What we got wrong
Scoring the AI models also meant checking our own work, and we found problems on our side. Here is what we found and what we are changing.
| Priority | What we found | What it looked like | What we are changing |
|---|---|---|---|
| High | The AI models copied our example numbers. | Identical answers appeared on UFC 331 (55/40/2/3), the Spanish GP and the Azerbaijan GP. Copied numbers are not predictions. | Take the numbers out of the example in our prompt. |
| High | One page showed the wrong matchup. | The first ALDS game page listed the White Sox and Guardians for a Yankees at Rays game, so the title and the chart did not match. | Check the first prompt's output against the page heading before the page goes live. |
| High | The contender list missed some winners. | Winners such as Verstappen and golf field winners fell outside the list, which gave “other” too much weight. | Use a longer contender list for golf and F1. |
| High | Test entries and a cut-off question were still in the system. | Two test rows and one truncated Spanish GP question needed cleaning up, and two US Open pages used an older template. | Delete the test rows, restore the full question, and move the US Open pages to the newer template. |
| Medium | Some AI answers were extreme or missing. | Claude declined the Laver Cup question and left out “Other driver” on the Azerbaijan GP. Amazon Nova gave 10 to 15% to a tie or postponement on NFL games. A few answers were 0% or 100%. | Under review. We will flag answers that look extreme before they reach the average. |
Source: Eye2.AI prediction review, data as of October 12, 2026. Priorities are the review’s own.
Limits of this review
- Small sample. Twenty events over about seven weeks, spread across many sports. Groups of two to four events cannot show a pattern.
- A win chance is not a promise. A 60% pick should lose about four times in ten, and some of these misses are exactly that.
- We scored the prediction shown on each Eye2.AI page and checked results against public sources. Nobody outside Eye2.AI has audited the review.
- The totals depend on how we count. Treating “other” as a pick helps the record, so we show both versions.
- This is not betting advice.
