The Football Manager Data Report 2026
footballgpt.co/fm/data-report
24,099 Football Manager questions against 24,759 real-world coaching questions, all time. From March to August 2026 the game overtook the pitch. This is what 22,787 of those questions were about.
22,787
FM questions in the window
11.4%
asked in a language other than English
4-2-3-1
the shape named most
Aggregate only. No player, save, club or account is identifiable. Questions are counted, never quoted.
Finding 01 · The game overtook the pitch
n = 42,555 mode-tagged questions · March to August 2026
1,302 FM questions in March to 6,245 in August. Counts are questions sent in each mode, not people. Coaching questions grew too.
Finding 02 · What FM players actually ask
n = 22,787 FM questions · fm-taxonomy-v1
Each question is assigned to the first matching theme by a rule-based classifier (fm-taxonomy-v1). 32.4% everything else and 8% asked in another language, not classified. Shares are of all FM questions in the window, so the bars do not add to 100.
Finding 03 · The formations they build
n = 1,624 questions naming a shape · March to August 2026
7.1% of FM questions name a shape that adds up to ten outfield players. Shares are of those questions. Shapes named fewer than 50 times are not shown. A question naming two shapes counts once for each.
Finding 04 · The styles they chase
n = 2,961 questions naming a style · March to August 2026
13% of FM questions name a recognisable style. Shares are of those questions and a question can name more than one style. Styles named fewer than 50 times are not shown.
Finding 05 · The clubs they manage
n = 1,857 questions naming one of 65 clubs · March to August 2026
Ranked by how often each club is named, bars relative to the top club. Only 8.1% of FM questions name a club at all, and the check covers 65 well-known clubs, so this is the shape of a thin sample, not a league table. Club names are shown as text; no crest or badge is used.
Finding 06 · When they play
n = 22,787 FM and 19,768 coaching questions · March to August 2026
Share of each lane's questions by day of the week, UTC. FM: Sunday 15.8%. Coaching: Monday 16.6%, Saturday 10.2%. By time of day FM leans to the afternoon and evening (32.2% and 31.1%), coaching spreads more evenly. These are timestamps of questions, not of matches played.
Finding 07 · Which game, and in what language
n = 22,787 FM questions · March to August 2026
10.1% of FM questions name FM26 and 1.5% name FM24; the rest name no version. Language is detected by function words, so it is coarse, and only languages above the 50 floor are named. Shares are of all FM questions.
For FM players · code FM50
Ask it about your own save, your tactic and your squad. It is the same FM Mode that answered the 22,787 questions above. Normally $4.99 a month. New customers only, with code FM50 at checkout. Free to try without a card.
How this report was built
Source. Questions sent to FootballGPT in Football Manager mode between 1 March and 31 August 2026, compared with real-world coaching mode over the same window. The all-time split on the cover is every mode-tagged question since records began. Computed on 6 September 2026.
Method. A rule-based classifier (fm-taxonomy-v1) assigns each question to its first matching theme. Formations are strings that add up to ten outfield players. Styles, clubs, game versions and languages are matched by published word lists. The rules live in the FootballGPT codebase and are versioned; a rule change means a recompute, never a hand edit.
Floors. No ranked row is shown below 50 questions. Accounts that opted out of the data portal are excluded upstream. No question text, save, club setup or account appears anywhere in this report.
What it does not prove. This is what people asked, not what they were told and not what happened in their save. It says nothing about the quality of FootballGPT's answers. It is not a survey of Football Manager players in general: it describes the people who brought their save to this product.
The coaching side of the same dataset is at footballgpt.co/data, with the full methodology.