AI scouting report

What an AI scouting report is

An AI scouting report is a written assessment of a footballer, drafted by an artificial intelligence model from the player’s data —minutes, performance, injuries, contract, market— and ending in a signing recommendation. At Gemsy it is written by Vega, the platform’s AI model.

It does not replace the scout at the stadium or the video analyst. It replaces the hours spent collecting, cross-checking and organising data before that trip: you get a document that already says what to watch for, what still needs confirming and what price makes sense. What makes it useful is not the amount of data but that every sentence can be checked, and that the conclusion dares to say “sign”, “monitor” or “pass”.

What a Gemsy report contains

Every report follows the same structure, designed for a sporting director who has to decide quickly: conclusion first, arguments after.

A real report, untouched

This is a Gemsy report exactly as Altair delivered it. We requested it from an internal test account on 19 October 2026, with high reasoning, on Alberto Moleiro, a winger at Villarreal CF. It is in Spanish because it was requested in Spanish: Gemsy writes each report in the language you ask in, and we prefer to show a real report rather than translate it. Figures and the player’s situation are as of that date, which is why the report opens by stating how recent its data is.

What it shows, in short:

• It separates data from opinion. Figures carry their date and a generic origin (the database, the club, the press); interpretations are marked as such.
• It states what it does not know. When two sources disagree on last season’s minutes, it refuses to give a figure and explains why. When there is no physical data, it says how to get it.
• It commits. The verdict is “MONITOR, do not sign today”, with the exact conditions that would change the decision (“sign him if…”, “rule him out if…”), a reasoned maximum price and dated next steps.

Report quality: Gemsy against the best general-purpose AIs

Update of 11 November 2026: against Claude Fable 5.2

On 11 November 2026 we repeated the test against Claude Fable 5.2, the most recent general-purpose model, with the same 23 players, the same request and the same criteria. Gemsy’s report, now with the Vega model, is on a par with Claude Fable 5.2: we generated it twice and judged it in two separate rounds, and in both the difference is not statistically significant.

• Round 1: +3.0 points for Gemsy [−0.5, +6.4]. Gemsy scores higher on 17 players and lower on 6: statistical tie.
• Round 2: −0.5 points for Gemsy [−5.6, +4.4]. Gemsy scores higher on 13 players and lower on 10: statistical tie.

What you should know about this update: in both rounds Gemsy scores higher on usefulness and on being up to date, and Claude Fable 5.2 on not stating what it cannot back up. There was a single judge, Claude Opus 5.5, from the same company as Claude Fable 5.2; the reports were anonymised and in random order. All the reports were generated that same day. The other systems in the October table were not measured again: their figures are those of 13 October and cannot be compared with those of this date.

Full test of 13 October 2026: five systems (Gemsy with the Altair model)

We measured the quality of Gemsy’s reports with a blind test: 23 players, the same request to every system and two judges who check the facts on the web before scoring. Gemsy’s report is on a par with the best general-purpose model tested: GPT-6.1 Sol scores 87.9 out of 100 and Gemsy 85.5, a difference that is not statistically significant. Claude Fable 5.1, Claude Opus 5.5 and Gemini 3.1 Pro High come next.

Gemsy against each one (mean difference per player and 95% confidence interval):

• GPT-6.1 Sol: −2.4 points [−6.3, +0.6]. Gemsy scores higher on 10 players and lower on 12 (1 tied): statistical tie.
• Claude Fable 5.1: +6.4 points [+1.2, +10.9]. Gemsy scores higher on 18 players and lower on 5: Gemsy ahead.
• Claude Opus 5.5: +26.2 points [+19.7, +31.7]. Gemsy scores higher on 22 players and lower on 1: Gemsy ahead.
• Gemini 3.1 Pro High: +34.7 points [+28.4, +40.3]. Gemsy scores higher on 22 players and lower on 1: Gemsy ahead.

By type of player (average score out of 100):

What the figures say, including what does not favour Gemsy: it has the best up-to-date score of the five (17.9 out of 20), because it checks the player’s current situation before writing, and it ranks first in two of the four groups of players, including European professional leagues. It still has room to improve on accuracy and on not stating what it cannot back up: the judges flagged 138 errors in its 23 reports, against 87 for GPT-6.1 Sol. In women’s football and in leagues outside Europe it is behind GPT-6.1 Sol.

How it was done

• Date: 13 October 2026, with the players’ situation as of 12 October 2026.
• Players (23): 4 from European professional leagues, 3 from the Americas and Asia, 5 women from four women’s leagues and 11 from Spain’s Primera and Segunda Federación (those in the table above).
• Request: the same for all: a scouting report in eight sections (summary, fit, physical, technical and tactical profile, risks, contract and market, verdict) to decide whether to sign the player that day, searching the web for the latest information. The request was written in Spanish.
• Gemsy: the high-reasoning report of the Altair model, in its version of that date.
• Judges: two AI models, Claude Opus 5.5 and Claude Fable 5.1, separately and with the reports anonymised and in random order. Each judge first checks each player’s key facts on the web (injuries, minutes, contract, market) and then scores: accuracy 40, up to date 20, usefulness for a signing decision 25 and no fabrication 15.
• No self-judging: since both judges are also contestants, Claude Fable 5.1 is scored only by the Opus judge and Claude Opus 5.5 only by the Fable judge; the rest get the average of both judges.
• Statistics: paired differences per player with a 95% confidence interval (bootstrap). If the interval includes zero, it is a tie.
• Who ran it: the test was designed and run by Gemsy; it is not an independent audit. Only the five systems in the table were measured.

How to request one

1. Create a free account on Gemsy. No card needed.
2. Type what you need in the chat, the way you would tell a colleague: “Write me a report on Alberto Moleiro”. If you do not know who yet, ask for candidates first (“right backs under 23 with plenty of minutes in the Eredivisie or the Liga Portugal”) and then for the report on the one you like.
3. Choose the reasoning level. Low for a report in seconds; high when the decision matters and you want it checked against the latest public information.
4. Download it as PDF, Word, Excel or PowerPoint for the meeting, or save the player to a shortlist.

Low and high reasoning

With low reasoning, the report is written from the Gemsy database: season, career, injuries, contract and comparison with players in the same position. It is the quick report for filtering a list of candidates.

With high reasoning, Vega spends a few minutes checking that data against recent public information before writing: the last match, what the club and the coach say, the contract situation or interest from other clubs. When something does not add up, it says so. The example above is a high-reasoning report. What each plan includes is on the plans page.

What the report does not do

• It does not watch matches. That is why its next steps almost always include which games to watch and what to confirm with the club.
• It does not know the dressing room or the player’s personal situation beyond what is public.
• It does not turn market value into price. Value is a dated reference; price is set by negotiation.
• It does not decide for you. It lays out the arguments for and against; the decision belongs to the sporting director.

Who it is for

For the sporting director who needs a clear recommendation before a meeting; the scout who wants to arrive at the stadium knowing what to look for; the agent who has to present a player with arguments; and the analyst who would rather start from an organised document than from a blank page. If you want to write your own reports by hand, our guide on how to write a football scouting report walks through the structure step by step, with a template.