Artificial intelligence is moving into professional sport, but access to a chatbot does not mean that every coach is equally prepared to use one. A new international study of 213 professional football coaches suggests that a basic level of digital competence may sharply separate occasional or reluctant users from coaches who incorporate AI tools more extensively into their work.
The study found a particularly striking threshold. Among coaches scoring below 4.5 on a seven-point digital-literacy scale, 28% had adopted AI chatbots. Above that threshold, adoption reached 78%. The difference was statistically significant, with a chi-square value of 48.7 and p below 0.001.
Age initially appeared to work against adoption, as might be expected in a rapidly changing technology environment. Yet mediation analysis suggested that the age relationship operated through digital literacy. Once differences in digital competence were considered, age itself no longer explained adoption in the same way. A similar pattern appeared among highly experienced coaches who perceived AI as culturally difficult to integrate into football practice.
The results do not show that improving digital literacy will automatically cause a coach to adopt AI, nor do they establish that chatbot use improves team performance. The study was cross-sectional and relied on self-reported behaviour and outcomes. What it does provide is a detailed snapshot of where adoption appears to stall, which capabilities distinguish heavier users, and what coaches say they gain when AI becomes part of a broader working process.
AI adoption in coaching is not simply a question of age
Generative AI can potentially support a wide range of coaching activities. A coach might use a chatbot to organise information, generate ideas for training sessions, prepare communication, analyse or summarise material, or support planning around matches and player development. The practical value of those tools, however, depends on the user’s ability to formulate useful requests, assess the resulting information and integrate it into professional judgement.
That makes digital literacy more than a background demographic characteristic. It can function as part of the infrastructure needed to use AI effectively. The researchers therefore examined whether differences usually attributed to age, experience or professional culture might instead reflect differences in underlying digital competence.
This distinction matters for organisations deciding how to introduce AI. If age is treated as the principal barrier, older coaches may be assumed to be naturally resistant to new technology. If digital competence explains much of that relationship, the problem becomes more actionable. Training can be aimed at skills rather than demographic groups.
An international survey of 213 professional coaches
The researchers conducted an international cross-sectional online survey involving 213 professional football coaches. Adoption and digital literacy were assessed with a psychometrically evaluated 31-item index. Rather than relying on a single statistical test, the team used several complementary analytical approaches, including mediation models, receiver operating characteristic analysis, analysis of variance and cluster analysis.
This multi-method design allowed the researchers to ask different questions of the same dataset. Mediation analysis examined whether digital literacy statistically accounted for relationships involving age and experience. Receiver operating characteristic analysis was used to identify a potential literacy threshold associated with adoption. Cluster analysis then looked for distinct patterns of AI use among coaches, while analysis of variance compared reported outcomes between those user groups.
The approach is useful because technology adoption rarely falls along a single continuum. Two coaches may both say they use AI, while one uses a chatbot occasionally for isolated questions and another incorporates it across planning, analysis and communication. Grouping usage patterns can therefore reveal differences hidden by a simple user versus non-user comparison.
A 4.5 out of 7 threshold separated sharply different adoption rates
The clearest numerical result was the threshold identified for digital literacy. At a score of 4.5 out of seven, adoption probability changed substantially. Only 28% of coaches below the threshold had adopted AI chatbots, compared with 78% above it.
That is a 50 percentage-point difference and means adoption was about 2.8 times as common among coaches above the identified threshold. The association was strong enough to produce χ²(1) = 48.7, with p below 0.001.
The threshold should not be interpreted as a universal pass mark. It was derived from this particular sample, instrument and professional context. A coach scoring 4.4 is not meaningfully transformed by moving to 4.5, and a threshold found in football cannot automatically be applied to teachers, clinicians or office workers. Its value is instead as evidence that adoption may accelerate once a foundation of digital competence is in place.
For training design, this suggests that introductory AI programmes may be less effective if they begin with advanced prompting techniques while participants still lack broader digital confidence. The study points toward a staged model in which foundational competence comes first and more specialised AI workflows follow.
Digital literacy accounted for the apparent age disadvantage
Age showed a negative relationship with adoption before digital literacy was considered. The mediation analysis, however, indicated complete statistical mediation through digital literacy, with an indirect effect of ab = -0.31.
In practical terms, the model suggests that older coaches were less likely to adopt AI because they tended to have lower digital-literacy scores, rather than because chronological age independently prevented adoption. Once the literacy pathway was incorporated, the direct age relationship was no longer the central explanation.
Complete mediation is a statistical description, not proof of a causal chain. Because all variables were measured within a cross-sectional survey, the analysis cannot demonstrate that age caused lower literacy or that lower literacy subsequently caused lower adoption. Unmeasured factors could influence all three. Access to technology, organisational support, previous digital training, role seniority or personal interest could contribute to the observed pattern.
Even with that caution, the finding challenges a simplistic generational interpretation. It suggests that a skill gap may be more informative than an age gap when football organisations try to understand why AI uptake differs across their coaching staff.
Experience looked like a cultural barrier until literacy entered the model
A related result emerged among coaches with more than 15 years of experience. Highly experienced coaches were more likely to perceive a cultural barrier around AI, potentially reflecting established routines, professional norms or scepticism about fitting a new technology into coaching practice.
Yet here too, digital literacy statistically accounted for the relationship. The reported indirect effect was ab = 0.38, and the authors describe the mediation as complete. Lower digital literacy explained the apparent link between extensive coaching experience and perceptions of a cultural barrier.
This is an important distinction for professional development. Resistance can easily be framed as an attitude problem, especially when experienced practitioners question a new tool. The results suggest that at least some of what appears to be cultural resistance may coincide with uncertainty about how the technology works or how to use it effectively.
That does not mean every concern about AI disappears with training. Coaches can have legitimate objections involving accuracy, confidentiality, tactical information, overreliance, intellectual property or the replacement of contextual judgement with generic outputs. Digital literacy may in fact make some users more critical, because greater competence can improve their ability to recognise limitations.
A small group had integrated AI across a wider value chain
The cluster analysis identified a distinct group the researchers termed the AI Value Chain. This group represented about 10% of the sample. Rather than using chatbots for a narrow isolated task, these coaches appeared to integrate AI more broadly across their working processes.
The group reported substantially higher satisfaction, with a mean score of 8.9. Differences between user clusters were statistically significant, F(2,121) = 15.8 with p below 0.001. Members of the AI Value Chain group also reported time savings exceeding three hours per week.
Those figures are potentially important for the business of professional sport. Three hours per week, if genuine and sustained, could represent meaningful capacity for coaches whose work includes planning, meetings, video review, player communication, administration and training preparation. But the outcome was self-reported. The study did not independently time coaches’ workloads or experimentally compare equivalent tasks completed with and without AI.
Higher satisfaction among intensive users is also open to more than one interpretation. Effective integration may create satisfaction, but people who already enjoy and trust AI may be more inclined to integrate it extensively. Cross-sectional data cannot establish which direction dominates.
The findings favour capability building over demographic targeting
The practical message is not that every football coach should immediately use a chatbot. It is that organisations interested in AI adoption may get more value from measuring and building digital competence than from assuming that adoption problems belong to older or more experienced employees.
A threshold-based training strategy could begin by identifying whether coaches have the foundational skills required to evaluate digital information, navigate tools confidently and understand basic limitations. AI-specific instruction can then build on that foundation through realistic coaching tasks rather than abstract demonstrations.
The study’s cluster result also suggests that training should move beyond teaching isolated prompts. The coaches reporting the strongest satisfaction and time savings were characterised as a value-chain group, implying integration across multiple parts of the workflow. The organisational question is therefore not only whether staff can ask a chatbot a useful question, but whether they can decide where AI adds value, where human judgement must remain dominant and how outputs should be checked before they affect decisions.
That distinction is especially relevant in football, where decisions can involve sensitive player information, tactical confidentiality and high-stakes performance judgements. Digital literacy should include critical evaluation and responsible use, not merely technical fluency.
What the study cannot tell us
The strongest limitation is the cross-sectional design. The study identifies associations and statistically modelled pathways at one point in time. It cannot show that an intervention raising digital literacy from below to above 4.5 would cause adoption to jump from 28% to 78%.
The sample of 213 professional coaches is informative but modest for drawing conclusions about a global profession. Football systems vary widely by country, competition level, club resources, language, data infrastructure and access to specialist analysts. The same chatbot may be useful in a well-resourced professional environment and much less practical elsewhere.
Self-report creates another limitation. Adoption, satisfaction, perceived barriers and time savings may be affected by recall or enthusiasm. Objective measures of chatbot activity, task completion time, output quality and coaching decisions would strengthen future research.
Most importantly, the study did not test whether AI use improved match results, player development, injury outcomes or the quality of coaching decisions. Reported efficiency and satisfaction are relevant outcomes, but they are not substitutes for performance evidence. A tool can save time while producing poor advice if its outputs are not critically evaluated.
The next test is whether training changes behaviour
The study provides a useful hypothesis for future experiments: digital literacy may be a modifiable bottleneck in professional AI adoption. The next step would be to test that proposition prospectively.
A randomized or longitudinal study could measure coaches’ baseline competence, provide structured digital-literacy training, and then observe whether adoption, task efficiency and output quality change over time. Such work could also test whether the 4.5 threshold replicates in another sample or whether it is specific to the instrument and coaches studied here.
For now, the findings shift the conversation away from who coaches are and toward what they know how to do. Age and long experience may correlate with AI uptake, but in this sample digital literacy provided the more direct statistical explanation. That makes competence a more constructive target for organisations trying to introduce AI without reducing technology adoption to a generational stereotype.
Source Information
Study: Ghorbel, A., Yaakoubi, M., Trabelsi, O. et al. Digital literacy shapes AI chatbot adoption, barriers, and performance outcomes in football coaches: a multi-method analytical study.
Journal: Scientific Reports (2026).
Publication date: 27 September 2026.
DOI: 10.1038/s41598-026-72410-4.
Study design: International cross-sectional online survey using mediation analysis, receiver operating characteristic analysis, analysis of variance and cluster analysis.
Sample: 213 professional football coaches.
Measures: AI chatbot adoption and digital literacy assessed using a psychometrically evaluated 31-item index, alongside reported barriers, usage patterns and outcomes.
Key caution: Cross-sectional self-reported data cannot establish that digital literacy causes chatbot adoption or that AI use improves coaching or team performance. The identified 4.5 out of 7 threshold requires replication.
