GPHG 2026 - can an AI model predict the shortlists?

The Grand Prix d'Horlogerie de Genève (GPHG) website is not very user friendly, at least if that user is interested in researching a brand's history at the event, or comparing two brands head-to-head, or being able to perform a basic search.

The GPHG Index website, on the other hand, has converted 25 years of GPHG history into a fully searchable database that includes over 5,000 watches and all the prizes awarded. You can list the winners for each year, see a brand's historical performance at a glance and even see a head-to-head record for two different brands.

More interesting than that, though, is the site's attempt to use AI models to predict whether a watch will make the shortlist as well as the probability of a shortlisted watch winning its category.

The models that underpin this were first produced over ten years ago, primarily to see whether there was any commonality between shortlisted and winning watches. They have recently been updated using the state of the art in generative artificial intelligence. While you cannot hope to predict winners in a competition that has human involvement, the results are nevertheless interesting. It turns out that a brand's pedigree at the competition is the most important factor in whether or not a watch is selected (which makes the one-shot brands - those who enter once and win - all the more remarkable).

One basic thing the model had not considered, however, was the number of entrants in a category. A watch entered in a category with only ten contenders obviously has a higher percentage of being picked than one in a category with fifty watches. You don't need AI to tell you that, but you do need to pass it on to the AI. This made a considerable difference to the predictions in some categories, shifting the favourite's Index score (a measure of its probability of making the shortlist) up by 26 percentage points in the Mechanical Clock category (seven entrants) and down by 36 percentage points in the crowded Challenge category.

Despite this, the individual predictions for each category remained the same, just with a higher score. So how did the model do?

Although its overall predictions weren't great, its first pick was shortlisted in seven of the categories. See the full predictions here.