Starting Monday, June 22, a machine helps decide who's ranked where in the UFC.

The promotion confirmed the rollout during Saturday's UFC Vegas 119 broadcast, where commentator Dan Hellie told viewers the UFC would "begin the transition from the traditional media panel towards a new objective, data-driven approach," rewarding "measurable performance" over "opinion or popularity," according to BloodyElbow's report. The factors he listed: who you beat, strength of competition, activity and consistency. Full details land at launch.

The traditional media panel isn't dead, though. Dana White confirmed both systems will run side by side. "I'm gonna have both," White said, per BloodyElbow. "I'm still gonna do human rankings. I think neither will be perfect, but it will get us closer." He framed the AI numbers as one more opinion in the room, with the fans as the real sanctioning body.

This is our exact corner of the sport, so we'll skip the cheerleading and the doom. An algorithmic ranking can do some things a voting panel cannot, and it will fall on its face in places a panel never would. We know, because we've been building and grading prediction models on this sport for a while, and the lessons transfer.

What an algorithm actually fixes

Start with the honest case, because it's real.

Voter politics and absenteeism go away. The media panel has long drawn criticism for inconsistency, for lapsed ballots, and for the slow creep of name recognition into a list that's supposed to measure merit. White himself has been blunt about disliking the current format. A system that scores fights the same way every week, with no favorites and no missed votes, removes that human drag entirely. That's a genuine gain.

Recency and activity get weighted on purpose, not by mood. A panel tends to remember the last great performance and forget the fighter who's been hurt or sidelined. An algorithm can encode "activity" as an explicit input, so a champion who hasn't competed in 18 months drops by design rather than by whim.

Strength of schedule becomes a number instead of a vibe. This is where the idea has legs, and the UFC isn't first to it. Tapology recently added a strength-of-schedule metric that grades a fighter's last six opponents on a 1-to-99 scale, BloodyElbow noted. Quantifying who someone actually beat, rather than how famous those wins felt, is the right instinct. Done well, it's the single most useful thing a ranking model can contribute.

Where it's going to struggle

Now the part the launch announcement won't dwell on.

Small samples wreck math models. A division leader might have three UFC fights. A debuting contender might have one. There's no clean statistical way to rank a fighter on a handful of bouts without the model leaning hard on priors, which is a polite term for guessing. Our own work runs into this constantly: thin data is thin data, and confident outputs on thin data are how you embarrass yourself.

Cross-division comparison is close to unsolvable. Pound-for-pound is the obvious trap. Comparing a flyweight to a heavyweight requires a value judgment that no amount of fight data settles, because the fighters never share an opponent pool. An algorithm will produce a P4P number, and that number will look precise. Precision isn't the same as being right.

Style mismatches don't show up in the inputs. Records and strength-of-schedule can't tell you that Fighter A is a nightmare matchup for Fighter B specifically. The "who you beat" input flattens a sport that's deeply rock-paper-scissors. The eye test still catches things the spreadsheet misses, which is precisely why White is keeping humans in the loop.

And then the unglamorous one: garbage in, garbage out. A performance-based system is only as good as the fight data feeding it. Inconsistent striking and grappling stats, scoring quirks and missing inputs all propagate downstream. We've spent real time cleaning UFC fight data, and "the numbers are clean" is a claim that has to be earned, not assumed.

Why we're not throwing stones

It would be cheap to critique a model without putting our own record on the table, so here it is.

FightIQ's live 2026 prediction record is 67.9% across all picks. On our high-conviction calls, the ones we tag LOCK or HIGH, it's 81.4% (57-13). A leak-free backtest on held-out fights lands around 70% all-picks and 84.3% on locks. Those are the verified numbers, and we don't blend the live and backtest figures to make either look better.

We also own the misses, which is the whole point. We made Ilia Topuria a HIGH-confidence pick at 77.4% against Justin Gaethje at the White House. The model only had him at 56.9%, the tag oversold a coin flip, and Gaethje knocked him out in the fourth round to take the undisputed lightweight title, one of the bigger upsets in the sport's recent history. Our gold dataset records it as a Round 4 KO/TKO for Gaethje. We backed it at HIGH confidence. It wasn't. That's the cost of putting numbers in public.

So we'll say the quiet thing plainly. Prediction accuracy and ranking quality are different problems. We're good at one and we still get individual fights wrong. A ranking model isn't forecasting Saturday's main event. It's compressing years of results into an ordinal list, which carries its own failure modes around sample size, cross-division math and data hygiene. Being right about who wins doesn't automatically make you right about who's third.

The bottom line

An AI ranking will be more consistent than a panel of humans and worse at the things humans are quietly good at. Keeping both, as White is doing, is the sane move, even if it's partly a hedge.

We'll be watching closely on Monday, with one specific question: when the algorithm and the panel disagree, who's right? That's a falsifiable question, and answering it honestly is the entire job. We'll grade it the same way we grade ourselves.