The digital transformation of primary care in Europe has accelerated in recent years, driven by policy imperatives, technological advancements, and the necessity for remote care during the COVID-19 pandemic [1,2]. Digitalization is reshaping general practice, with general practitioners (GPs) increasingly using electronic health records (EHRs) or electronic medical records (EMRs), telemedicine, and AI-based tools, such as clinical decision support systems (CDSS). While these technologies promise greater efficiency and access, their effects on everyday practice and professional roles vary across contexts [1,3,4]. Several reviews have examined digital health in primary care, including scoping and systematic reviews of digital health interventions and integrated care models in European primary care, as well as global reviews of telehealth, e-consultations, patient portals, and electronic CDSS tools.

Digital health technologies present both promise and paradox in general practice, with studies reporting improved patient access to care through e-consultations [5-7] and AI-driven triage tools [4,8], but their implementation is often problematic. For instance, while electronic CDSS can reduce diagnostic errors [4,9], they may simultaneously increase administrative workloads due to fragmented interfaces and alert fatigue [2,3,9,10]. Similarly, telemedicine has expanded access to underserved populations [6,11], yet GPs report disruptions to workflow continuity [12] and concerns about depersonalized care [5]. Though empowering for patients, symptom checkers and patient-operated triage tools risk misdiagnosis and generate additional consultation requests, thus amplifying GP workloads [8,13].

Integrating digital tools into primary care workflows has introduced significant challenges to workflows and the workforce. Pre–COVID-19 studies highlight that eHealth tools, such as EHRs, often failed to reduce administrative burdens, instead creating “double documentation” tasks [2]. Postpandemic, complex integrated care models supported by digital interventions have further strained GPs, requiring new competencies in data management and patient communication [1,14,15]. Machine learning applications that automate administrative tasks, such as coding and billing, show promise [16-18]. Still, their adoption is hindered by interoperability issues and a lack of trust in algorithmic outputs [4,18,19].

Regional disparities and competency gaps continue to shape the uneven impact of digitalization across Europe, reflecting differences in infrastructure, funding, and training [7]. Northern and Western European countries report higher adoption rates of EHRs and telemedicine, whereas Eastern European systems lag due to resource constraints [1,11]. Even in technologically advanced settings, GPs often lack the digital literacy needed to leverage AI tools effectively [16], exacerbating inequities in care quality [4,7,14]. Furthermore, ethical concerns about data privacy and algorithmic bias persist, particularly in AI-driven diagnostics, where transparency remains limited [4,9,13,17,18].

Despite growing research on patient outcomes and system efficiency, important gaps remain in the literature, particularly regarding GPs’ subjective experiences [3,20]. Several prior reviews have examined digital health in primary care, but largely from perspectives other than European GPs’ subjective experiences. Mold et al (2019) [5] synthesized evidence on patient-provider e-consultations in primary care, focusing on usage, access, and service outcomes rather than GP attitudes. Jimenez et al (2020) [14] mapped digital health competencies for primary care professionals, identifying skills and training needs but not providing an in-depth account of the day-to-day experiential consequences of digitalization. Fletcher et al (2023) [3] scoped the workload and workflow implications of electronic CDSS across high-income primary care systems, highlighting perceived time pressure, alert fatigue, and workflow disruption, but did not isolate European GPs’ wider views on digital transformation. Susanto et al (2023) [17] reviewed machine-learning-based CDSS and mainly reported effects on decision-making, care processes, and patient outcomes, while Radionova et al (2023) [13] and Gottliebsen and Petersson (2020) focused on symptom checkers and triage tools, emphasizing safety, demand, and mixed professional views rather than GPs’ broader subjective experience [8]. Sørensen et al (2023) [16] described machine learning applications and administrative automation in general practice, with attention to task support and efficiency [18], whereas Badr et al (2024) [21] examined digital health and inequalities at the population level, emphasizing structural and equity dimensions over frontline GP experience.

Other reviews have targeted specific domains such as telemedicine use barriers (Tabaeeian et al, 2024) [11], access systems in general practice (Eccles et al, 2024) [20], integrated digital care models in Europe (Mezzalira et al, 2024) [1], the impact of eHealth on general practice workload (Keuper et al, 2024) [2], and AI-based or conventional CDSS in primary care (Gomez-Cabello et al, 2024 [4]; Meunier et al, 2023 [9]; Susanto et al, 2023 [17]), typically prioritizing effectiveness, usage, or mixed professional samples rather than European GPs as a distinct group. Finally, Tun et al (2025) [19] and related work on trust in AI-CDSS, as well as Ambrosi et al (2025) [22] on digital interventions for chronic conditions, address acceptability and outcomes but not systematically the cumulative, everyday implications of digitalization for European GPs’ workload, workflow, autonomy, and doctor-patient relationships. As far as we are aware, no existing review has yet provided a focused synthesis of European GPs’ subjective experiences across digitalization. This review offers a dedicated synthesis of European GPs’ experiences across a broad range of digital tools (EHRs, telemedicine, digital health applications [DHAs] such as health portals, triage, and symptom-checker tools, and AI-based systems).

Therefore, the objective of this systematic review is to synthesize European GPs’ subjective experiences of digital health technologies between 2014 and 2024, using a combined effect-direction approach to characterize adoption and normalization processes and the “implementation paradox” in everyday general practice.

This systematic review explores GPs’ perspectives on digital health integration, focusing on the adoption trends of telemedicine, AI decision support, patient portals, and digital therapeutics; GPs’ experiences with benefits, workflow changes, and patient interactions; impacts on workload, efficiency, and ethical concerns (eg, data privacy and equity); barriers to implementation (technical, organizational, and patient-related); and policy recommendations for equitable, sustainable adoption. By synthesizing these insights, this review aims to support evidence-based policy and practice decisions for health care providers, policymakers, and digital health innovators.

This systematic literature review was conducted following the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) 2020 guidelines. All PRISMA checklists and the Synthesis Without Meta-Analysis reporting checklist [23] are available in Checklists 1 and 2 [23,24]. The review aimed to synthesize studies that explore GPs’ perspectives on digitalization in European primary care settings. The review protocol was registered retrospectively with the OSF (Open Science Framework; KE25C) on May 31, 2025 [25], after the initial search, but before data extraction, data synthesis, and analysis.

We screened for papers that met the following criteria:

A comprehensive search was conducted across 8 electronic databases: PubMed, Scopus, SpringerLink, Wiley Online Library, Cochrane Library, Web of Science, Google Scholar, and Ovid. In addition, we used Elicit [26,27], a browser-based AI-powered literature review assistant, to identify additional gray literature records and to export structured bibliographic and study-level data under full human supervision. During the search period (November 2024 to January 2025), we accessed Elicit [26] via a Pro subscription through a standard web browser. The interface did not expose a model identifier or software version, so we describe the tool by provider, URL, subscription tier, and time window of use. No additional sources, such as reference lists of included studies, manual journal browsing, or direct contact with authors or organizations, were used to identify studies beyond the specified databases and Elicit [26]. We did not search study registries, set up citation alerts, or perform backward or forward citation chasing beyond what is reported in Multimedia Appendix 1. The primary search was completed on January 5, 2025. A combination of free-text terms and Boolean operators was used, including: ("digital" OR "digitalization" OR "eHealth" OR "telemedicine" OR "artificial intelligence") AND ("primary care" OR "general practice" OR "general practitioners"). The search strategy was developed iteratively by JMF and was not prospectively peer-reviewed using a standardized checklist; instead, we conducted a retrospective PRESS (Peer Review of Electronic Search Strategies) 2015 assessment after the searches were completed [28]. We evaluated the full strategy against the core PRESS domains, including the translation of the review question into search concepts, the choice of subject headings and text words across databases, the use of Boolean and proximity operators, the application of limits and filters (time period, language, and study design), and the overall structure and syntax of the search strings. This assessment did not identify any major errors or omissions in these domains; only minor wording refinements were suggested, which did not materially alter the sensitivity or specificity of the original strategy. The entire search strategy and a detailed PRESS assessment table are provided in Multimedia Appendix 1 [29-33] and Checklist 3. Our search strategy was structured around the PICO (Population, Intervention, Comparator, and Outcome) elements of the review question (European GPs in primary care; digital health technologies; GP-reported experiences and impacts) and implemented using extensive synonyms and Boolean operators. Database-specific field tags (for example, PubMed [Title]/[pdat], Scopus TITLE-ABS-KEY, Cochrane ti,ab,kw, Web of Science topic fields) and subject headings were applied where available, but heterogeneous interfaces and uneven indexing for digital health meant we relied primarily on text-word searching and phrase searching rather than systematic use of MeSH/EMTREE and proximity operators. A search update was performed later on October 7, 2025, by rerunning the entire search using identical search strings and databases to identify studies published between January 1, 2025, and October 7, 2025. The updated search results are provided in Multimedia Appendix 2, but were not included in the present synthesis.

The initial search yielded 1580 references. After automatic and manual deduplication in the Rayyan web application, 251 duplicate entries were removed, yielding 1329 unique papers. Title screening excluded 929 papers that did not meet the inclusion criteria. Abstracts of the remaining 400 papers were assessed, and 118 studies were selected for full-text review. Following full-text evaluation, 65 studies met all eligibility criteria and were included in the final synthesis. A total of 53 full-text papers were excluded from the review. A detailed list of these excluded studies, along with the specific reasons for their exclusion, is provided in Multimedia Appendix 3. One primary reviewer (JMF) conducted the screening process. A second reviewer (LN) independently screened 20% (266/1329) of the total records during title, abstract (80/400), and full-text screening (24/118). Interrater agreement was excellent across all screening phases: 95.86% (255/266) at title screening (Cohen κ=0.84), 95% (76/80) at abstract screening (κ=0.87), and 95.83% (23/24) at full-text screening (κ=0.92). As all disagreement rates were below the prespecified 5% threshold, no additional screening was required. Discrepancies were resolved through discussion and consensus during team meetings. The final decision on inclusion or exclusion was made by a third reviewer (EZ). Data were stored in Zotero, and screeners’ decisions were noted in the Rayyan web application.

Elicit AI [26] and JMF performed data extraction. The preset extraction sets from Elicit AI [26] were used (eg, study metadata, study design, methodologies, study participants’ sample sizes, digital health intervention type, GPs’ perceptions or outcomes, quantitative results [if available], etc) to extract study data. In the first step, training and calibration were conducted using 5 studies, where extractions were performed by Elicit AI [26] and independently by the first author. A 100% overlap of extracted information was found. JMF cross-checked all finalized data extracted by Elicit AI [26]; in case of discrepancies, the reviewer used the manually extracted data. No authors were contacted for missing or unclear data.

Extracted data included the following:

For each included study, all reported findings that were compatible with the defined outcome domains were extracted. No selective collection of results was performed within eligible domains. The complete data-extraction table for all 65 included studies, covering every extracted variable, is provided in full in Multimedia Appendix 4 [29-93], and a condensed evidence summary (author, country, and key findings for each study) is provided in Multimedia Appendix 5 [29-93]; owing to their length, both are presented as Multimedia Appendices rather than in the main text, in line with JMIR house-style guidance for oversized tables.

Included studies were subject to critical appraisal using the Mixed Methods Appraisal Tool (MMAT) 2018 version [94]. No formal quality threshold was applied at the inclusion stage; all 65 studies contributed to the narrative synthesis irrespective of MMAT score. MMAT ratings were used descriptively to characterize overall study quality and to avoid overinterpreting isolated findings from lower-scoring studies. A total of 4 studies scored 60%, and the remaining 61 studies scored ≥80%; no study scored below 60%. We did not apply a formal GRADE (Grading of Recommendations Assessment, Development and Evaluation) approach. The papers’ quality was assessed manually by JMF and checked for disagreements by the other authors. The MMAT score table and a vote-counting analysis for the four 60% MMAT studies are reported in Multimedia Appendix 6 [29-93]. Design-specific JBI (Joanna Briggs Institute) critical appraisal tools were used to assess the risk of bias for all included studies. Qualitative studies were appraised using the JBI Qualitative Research checklist [95]; quantitative descriptive or cross-sectional studies, the JBI Analytical Cross-Sectional checklist [96]; and mixed methods studies, their qualitative and quantitative components were appraised separately using the relevant JBI tools. Study-level risk-of-bias results are reported in the Results, and a color-coded version of the appraisal is provided in Multimedia Appendix 6. Limited sensitivity analyses for respondent overlap and study quality (MMAT) are reported in Multimedia Appendix 7 as side-by-side comparison tables showing percentage distributions with and without the potentially overlapping or lower-MMAT studies. Given the marked heterogeneity of this literature, we did not undertake a formal meta-analysis. Instead, a narrative synthesis approach, following the framework of Popay et al (2006) [24], was used. Within this framework, we applied two complementary theoretical lenses. The technology acceptance model (TAM) [97] was used to analyze individual GP adoption decisions, focusing on perceived usefulness and ease of use. The normalization process theory (NPT) [98] was used to examine implementation processes, with coding for coherence, cognitive participation, collective action, and reflexive monitoring. Studies were grouped into four mutually exclusive categories (AI, digital health tools [DHTs], EHRs, and telemedicine). No changes were made to the planned groupings after protocol registration. Studies covering multiple technologies without a specific focus on AI, EMRs, or telemedicine were classified as DHTs. Each study was assigned to exactly one category, with subcategories providing additional granularity within the main classification scheme. The topic coding, along with all extracted data, is shown in Multimedia Appendix 4. Descriptive statistics, such as frequencies and percentages, were used to characterize the included studies (eg, methodological approaches, temporal trends, and geographic distribution) and to summarize quantitative findings from individual studies where available. We also created an effect-direction table to visualize heterogeneous findings from TAM and NPT analyses. We transformed study-level results into direction-of-effect categories (strong positive, positive, mixed, negative, and strong negative) using prespecified percentage thresholds, and then summarized the proportions of studies in each category. The included studies reported different but partly overlapping constructs across TAM and NPT domains, combined qualitative, quantitative, and mixed methods designs, and operationalized key outcomes using noncomparable, study-specific measurement scales, with only very small numbers of directly comparable studies per technology-country combination, often from different time periods. Under these conditions, a quantitative meta-analysis would risk producing a misleading “fruit-salad” summary [99]. No further statistical transformations of effect estimates were undertaken. JMF conducted the data synthesis, which the other authors further developed.

The review departed from the registered protocol (OSF; KE25C) in the following respects. The protocol was registered retrospectively, as noted under the Study Design section. Screening was conducted primarily by one reviewer (JMF), with a second reviewer (LN) independently screening a 20% sample at each stage and a third reviewer (EZ) making the final inclusion decisions, rather than full dual screening. In addition to the planned MMAT appraisal, design-specific JBI tools were added for a formal risk-of-bias assessment. Beyond the planned narrative and thematic synthesis, a combined TAM-NPT effect-direction approach was applied, and the prespecified rule of treating a theme as robust when it appeared in ≥30% of studies and across at least three countries was operationalized through the effect-direction thresholds reported above. Post hoc sensitivity analyses (respondent overlap; exclusion of studies scoring 60% on the MMAT) were added in response to peer review (Multimedia Appendix 7 [35,38,46,48,64,65,68]). A search update to October 2025 was conducted but not synthesized (Multimedia Appendix 2). Finally, one pan-European survey was retained in which 1 of the 31 countries surveyed was non-European (1/31, 3.2%), as the aggregated data could not be disaggregated. None of these deviations changed the review question or the eligibility criteria or altered the overall conclusions.

The studies were clustered into topics; a complete list of all extracted parameters, including key findings for each study, is available in Multimedia Appendices 4 and 5. DHTs comprised 55.4% (36/65) [29-31,34-66] of the studies, with general information and communication technology tools prominent (18/36) [29,31,34-38,40-42,44,47,48,50,53,59,64,66], and DHAs were explicitly examined in 6 of 36 studies [54-56,61,63,65]. Telemedicine accounted for 27.7% (18/65) [32,67-83], including studies on COVID-19 impacts (4/18) [72,73,78,81], video consultations (VCs) (4/18) [68,69,76,80], and email consultations (4/18) [67,71,79,83]. AI applications comprised 10.8% (7/65) [33,84-89], and EMR or EHR studies represented 6.2% (4/65) [90-93].

The screening process is depicted in Figure 1.

Appraised with the design-appropriate JBI tools described above, most of the 65 studies were at low to moderate risk of bias; only three—all short conference-proceeding papers, in which the publication format rather than the design limits methodological transparency—were rated high risk.

The most frequent limitations were incomplete reporting of researcher reflexivity in the qualitative and mixed methods studies and unclear handling of confounding and measurement validity in the cross-sectional studies. These did not invalidate any study; high-risk studies were interpreted with caution, were never the sole basis for a key conclusion, and their exclusion did not change the overall results (Multimedia Appendix 7). The full item-level appraisal for each study is shown in Tables 3-5.

Frequently mentioned barriers to implementation or adoption were the complexity of implementation and limited integration or compatibility with existing systems [34,41,43,49,57,82,84,88], poor applicability and design [57,60,77], technical and organizational obstacles [31,38,62,78,85,88], workflow adjustments and time constraints [34,49,71], costs and inadequate reimbursement [55,66,78,85,88], and training efforts [66]. Adoption decisions were also influenced by data protection as a critical barrier [50,55,70,90]. System interoperability problems were noted by a majority of UK and German GPs in 2 studies [50,66]. Another barrier mentioned was insufficient network coverage [82] and a lack of adequate broadband internet access [75]. Organizational barriers centered on a lack of skills [78], inadequate training (n=1044, 91.5%) of GPs who lacked formal digital health education [39], and change resistance, particularly in mixed-age practices [57]. Digital literacy was identified as a barrier to implementation, as it varied dramatically by age, with younger GPs showing a higher affinity for using digital tools [43,49,57,74,78,82]. In another study, staff felt that the provided training was not tailored to their individual needs and did not adequately support them in planning and understanding the intervention’s implementation [62]. Furthermore, observed barriers included perceived passive roles or uncertainty among GPs in delivering the intervention and the absence of a feedback loop [62]. Some GPs also reported a lack of comprehensive information about DHAs, with only a minority feeling competent to advise patients on their use [55]. Patient-related barriers reflected disparities in capability and preference, including digital affinity, motivation, age (younger patients demonstrated greater trust in digital consultations than older patients [60,93]), and health literacy [43,49,56,78,82]. Furthermore, systemic barriers included liability uncertainties (eg, GPs worried about telemedicine malpractice) [70,76] and evidence gaps, with many digital tools lacking robust outcomes data [41,57].

This systematic review synthesized evidence from 65 studies involving 24,994 GPs across 14 European countries and found a consistent pattern of high perceived usefulness but low perceived ease of use for digital health technologies in general practice. In the provided overall TAM analysis, all effect directions for perceived usefulness were positive (56/65, 86.2%). In contrast, responses regarding perceived ease of use were predominantly negative (43/65, 66.2%) or neutral (22/65, 33.8%), and behavioral intention showed only a modest positive trend (3/65, 4.6%). At the same time, the overall NPT analysis indicated that although GPs often engage with digital tools, the effect directions for cognitive participation (3/65, 4.6%, positive) and collective action (65/65, 100%, partial) were frequently negative, implying low buy-in to specific tools and limited embedding in everyday work routines. These findings align with the “promise and paradox” of digitalization described in the broader primary care literature, where digital technologies are widely deployed but often increase complexity and cognitive load rather than simplifying practice [105-108]. At the same time, the overall high actual use observed (63/65, 96.9%) in this review, despite low perceived overall ease of use (22/65, 33.8%, neutral, and 43/65, 66.2%, negative) and modest behavioral intention (84.6% partial and 10.8% negative), suggests that digitalization in European general practice is now largely system-driven rather than purely clinician-driven and that it is pervasive and largely mandatory but only partially normalized from the perspective of frontline GPs. The observed pattern of high overall reflexive monitoring (43/65, 66.2%) is consistent with this interpretation.

The temporal distribution of studies (52/65, 80%, published between 2020 and 2024) demonstrates that the COVID-19 pandemic period dominates the empirical evidence base on GP experiences with digitalization. This concentration mirrors broader analyses of telehealth and digital primary care, which report accelerated digital adoption during the pandemic, followed by heterogeneous trajectories of stabilization or partial rollback [108,109]. The fact that 55.4% (36/65) of all included studies focused on generic DHTs and that only a minority addressed AI (7/65, 10.8%), EHRs (4/65, 6.2%), or telemedicine (18/65, 27.7%) suggests that, for most GPs, digitalization is experienced as a diffuse, system-level change rather than as isolated innovations. This is consistent with cross-national surveys showing that basic eHealth functions are nearly universal in European primary care, while more advanced functions and AI-based tools remain unevenly implemented [38,106,110]. Geographically, 47.7% (31/65) of the included studies originated from Germany and the United Kingdom, with additional clusters in the Netherlands (7/65, 10.8%), and Scandinavian countries (17/65, 26.2%). This pattern parallels multicountry digital maturity assessments that identify Northern and Western Europe as early adopters with higher digital readiness, while Eastern and some Southern European countries lag [105,111]. The fact that this review found relatively few studies from Eastern Europe likely reflects this underlying maturity gap and underscores that current knowledge about GP perspectives is biased toward more digitally advanced health systems [112]. Our assessments on digital maturity in general practice also found that interoperability and evaluation of best practices were the weakest dimensions, and that rural settings and smaller practices were consistently associated with lower digital maturity. These patterns are consistent with a general finding that group practices and digitally “mature” practices report more positive experiences and higher use, whereas rural, resource-constrained settings struggle with connectivity, integration, and sustained adoption [105].

Across technologies, this review found that while digital technologies may shorten individual encounters, GPs more often reported increases in workload than reductions, particularly with telemedicine, patient portals, and electronic messaging. In the effect direction analysis, most studies on workload reported neutral or adverse effects, indicating perceived time pressure, additional documentation, and “hidden” tasks such as managing patient portal messages and reconciling duplicate documentation across parallel systems. These results are consistent with a pre–COVID-19 systematic review of eHealth use in general practice, which found that digital communication and portal functions frequently increased overall workload despite potential gains in individual encounters [2]. They are also consistent with modeling work on digital-first access in UK general practice, which estimated that online and VCs would increase total GP workload by 3%‐31% in typical scenarios unless consultation length and demand are carefully managed [113]. In addition, the TAM-NPT pattern identified high perceived usefulness (56/65, 86.2%), very low ease of use (0/65, 0%), low buy-in (3/65, 4.6%), but near-universal use (63/65, 96.9%), which fits with these workload findings. It suggests that GPs often experience digitalization as an obligation embedded in regulatory, reimbursement, and organizational requirements, rather than as a voluntary efficiency gain. This interpretation aligns with broader analyses of “work as imagined vs work as done” in digital health, which emphasize the gap between policy narratives of simplification and clinicians’ experiences of increased complexity and administrative “shadow work” [108]. The dominance of negative or neutral effect directions for workflow and workload in this review adds empirical precision to the more general concerns reported in umbrella reviews of digital health technologies for health workers, which describe increases in mental workload, multitasking, and documentation burden associated with EHRs and related tools [12,15]. At the same time, a subset of studies in this review reported positive workload effects for well-integrated asynchronous tools and for specific use cases, such as chronic disease monitoring or asynchronous specialist consultations [39,51,52,57,60]. This mirrors external evidence that digital interventions can reduce workload when they replace, rather than add to, existing processes and when they are matched to appropriate clinical tasks [6,107].

This review also found that digital tools improved access and continuity for specific patient groups but raised concerns about depersonalized care and uneven benefits. Numerically, most studies addressing continuity and access reported positive effects, whereas those focusing on relational quality and communication nuances more often reported neutral or adverse effects. This nuanced pattern aligns closely with qualitative studies of digital primary care in Scandinavia and elsewhere in Europe, where clinicians describe digital platforms as enhancing convenience and follow-up while limiting nonverbal communication and making complex conversations more challenging [114]. Furthermore, this review found that GPs consistently perceived older patients, people with low digital literacy, and those in rural or socioeconomically deprived areas as less likely to benefit from digital services. Throughout studies, GPs explicitly linked digital access channels to “digital queues” that are easier to navigate for younger, better-resourced patients, raising concerns about a widening inverse care law. These perceptions are strongly supported by a scoping review of digital health technologies and inequalities, which concludes that digital tools tend to reinforce existing social gradients in access and outcomes unless counterbalanced by targeted inclusion measures [21]. National-level analyses of remote consultation uptake in general practice likewise show higher digital use in less-deprived populations and lower uptake among older and more disadvantaged groups, even where services are universally available [7,115]. Against this background, the observed effect-direction patterns strengthen the argument that equity considerations must be integral to digital strategy in European general practice, rather than an afterthought.

Only a small fraction of the included studies in this review focused on AI or CDSS (7/65, 10.8%). The effect-direction analysis for these technologies showed a distinctive profile: GPs frequently rated perceived usefulness as high (6/7, 85.7%) but expressed negative or neutral directions for perceived ease of use (3/7, 42.9%, neutral and 4/7, 57.1%, negative), behavioral intention (6/7, 85.7%, neutral and 1/7, 14.3%, negative), coherence (2/7, 28.6%, positive and 5/7, 71.4%, neutral), and collective action (7/7, 100%, neutral). This asymmetry mirrors findings from recent scoping and systematic reviews of AI-based CDSS in primary care, which report that while such systems can improve process measures or diagnostic accuracy in controlled settings, clinicians express persistent concerns about transparency, responsibility, and the potential for over-reliance [4]. A systematic review of the benefits and harms of algorithmic decision-making systems in health care similarly concludes that evidence for clear clinical benefit remains limited and heterogeneous [18]. In contrast, perceived risks related to bias, opacity, and workflow disruption are repeatedly documented [18].

Moreover, these results indicate difficulties in understanding when and how to use these systems, as well as in integrating them with existing decision-making practices. This pattern aligns with studies on trust in AI-based CDSS, which find that lack of explainability, misalignment with clinical reasoning, and inadequate training are significant barriers to adoption [17,19]. The numerical imbalance between high perceived usefulness and low perceived ease of use, and the coherence in this review, therefore, reinforce the interpretation that AI in primary care remains at an early stage of normalization: GPs see potential value but lack the infrastructure, governance, and experiential reassurance needed to embed these tools in routine care.

The cross-cutting barriers identified include poor interoperability, fragmented systems, lack of training, inadequate reimbursement, and limited local support, and are described in multiple implementation-focused reviews and policy reports [9,10]. The combined TAM-NPT analysis reveals that 100% (65/65) of implementations remain partial, regardless of technology type, setting, or perceived benefits, which indicates systemic issues in how digital health is conceived, deployed, and supported in primary care settings. In combination with the high actual overall usage (63/65, 96.9%), this indicates that technologies are being adopted without full integration. Across all technologies, a lack of organizational support and training was found. GPs in the included studies described limited formal training, lack of protected time to learn new systems, and inadequate local support as key barriers to effective use. These findings resonate with European-level surveys showing that a majority of health care professionals report little or no structured digital skills training, and that available training is often perceived as insufficient or poorly aligned with practical needs [116]. They are also consistent with conceptual and policy papers arguing that digital competencies should be embedded into undergraduate and postgraduate curricula and accompanied by continuous workplace-based learning [117].

Regarding system conditions, this review found fragmented platforms and poor interoperability as significant sources of frustration, which negatively affected workflow integration and perceived system quality. OECD (Organisation for Economic Co-Operation and Development) and World Health Organization reports on EHR and telehealth deployment identify similar issues, noting that lack of common standards, fragmented vendor landscapes, and unclear governance undermine both clinician acceptance and the realization of potential benefits [117-119]. The convergence between these policy-level analyses and the quantitative patterns of effect direction observed here strengthens the conclusion that many of the problems GPs report are systemic rather than individual: they arise less from resistance to innovation and more from the design, implementation, and funding of digitalization.

By combining TAM and NPT with effect-direction analysis, this review provides a structured, quantitative overview of how European GPs experience digitalization across multiple outcome domains. The predominance of positive effect directions for perceived usefulness, contrasted with negative or neutral directions for ease of use, behavioral intention, cognitive participation, and collective action, suggests that digitalization in its current form creates a “compliance without conviction” dynamic: GPs use digital tools extensively because they must, not because these tools reliably make their work easier or their care better. External evidence on workload, burnout, inequalities, and implementation barriers supports this interpretation and indicates that similar patterns are being observed in other high-income primary care systems [2,9,21,120]. Taken together, this review and related literature suggest that the central challenge for European general practice is no longer whether to digitalize, but how to redesign systems so that digital tools genuinely support, rather than strain, frontline clinical work [111,116,117,121]. This review suggests that, for the digital transformation of general practice, digital technologies should be embedded in person-centered, team-based models of care, with iterative adaptation informed by user feedback and real-world evaluation. For GPs, this implies prioritizing (1) interoperable, user-centered systems that reduce duplicate documentation and align with existing workflows; (2) reimbursement and staffing models that recognize additional digital workload and support roles such as digital navigators or dedicated triage staff; (3) structured education on digital health and AI at undergraduate and postgraduate levels, combined with ongoing protected time for practice-based training; and (4) governance approaches that balance innovation with privacy, safety, and equity, including explicit standards for teleconsultation quality and AI use in clinical decision-making.

Future research should build on these quantitative patterns by (1) conducting longitudinal and experimental studies that can disentangle short-term disruption from long-term adaptation; (2) expanding the evidence base to underrepresented regions and practice settings, especially in Eastern and Southern Europe; and (3) rigorously evaluating hybrid care models that purposefully combine digital and face-to-face contacts, including their equity and economic impacts. Such work will be essential to move from documenting the paradoxes of digitalization to identifying concrete design, policy, and training strategies that allow digital tools to genuinely support GP-centered, equitable, and sustainable primary care in Europe [6,107,122].

This systematic review provides a comprehensive understanding of digitalization in European general practice, supported by several key strengths. We adhered to the PRISMA 2020 guidelines and used a robust, innovative search strategy across 8 databases in combination with Elicit AI [26] to identify gray literature. A significant strength is our extensive and diverse evidence base, which synthesizes data from 24,994 GPs across 14 European countries and encompasses a rich mix of qualitative, quantitative, and mixed methods studies. The predominance of qualitative and exploratory designs is appropriate for examining emerging technologies, where understanding contextual factors and user experiences is critical. While large quantitative surveys (eg, the pan-European survey with 9196 respondents [38]) allow for the identification of broad trends, qualitative studies with smaller samples provide rich contextual understanding of the mechanisms underlying these trends. Mixed methods designs enabled both the validation of trends using quantitative data and the exploration of underlying mechanisms through qualitative inquiry, providing complementary perspectives. This methodological diversity and consistency across methods (interviews, focus groups, and content analysis) strengthen the comparability of findings, as many studies independently identified similar themes in their qualitative data, adding confidence to the common denominators and enabling triangulation of findings. The included multicountry study [38] contained data from Turkey, which falls outside our strict geographic eligibility criteria. However, this study was retained for pragmatic reasons: Turkey represented only 1 of 31 countries surveyed (approximately 1/31, 3.2%), and the aggregate data could not be disaggregated to exclude Turkey while preserving the remaining European data. As this large quantitative study strengthens the evidence base, its inclusion outweighs the minor geographic deviation. Importantly, the findings from this study align with patterns observed in studies conducted exclusively within our defined geographic scope, and its inclusion does not materially affect the review’s overall conclusions. It is also worth noting that there may be some overlap (eg, the pan-European survey [38] may include respondents who also participated in country-specific studies). To assess this, we repeated the TAM-NPT effect direction analysis after excluding the three largest DHT surveys [38,46,65]; category percentages changed by less than 2 percentage points, and the overall patterns remained unchanged, indicating robustness to plausible double-counting. We also repeated the qualitative synthesis after excluding the four studies with MMAT scores of 60%; no changes in the direction of conclusions were observed in any outcome domain, suggesting that our main findings are robust to quality restrictions as operationalized by MMAT. Given the heterogeneity of the included studies, a narrative synthesis was appropriately chosen to integrate findings, reinforcing credibility through thematic consistency across independent datasets.

Despite these strengths, our review has several limitations. A primary constraint is its Eurocentric focus, as all included studies are from Europe, thereby limiting the generalizability of the findings to global contexts, particularly the Global South. Besides, representation from Eastern European countries remains limited. Furthermore, the language restriction to English and German may have introduced language bias. Further, 6 records were excluded solely due to language (1 Spanish, 3 French, 1 Italian, and 1 Dutch study; identified in Scopus, Web of Science, and SpringerLink), which may have led to a slight underrepresentation of evidence from these countries and related themes. This pattern is summarized in Multimedia Appendix 3. Additionally, the review predominantly captures the perspectives of GPs, with patient views and those of other crucial health care stakeholders underrepresented. Furthermore, although the search was broad and informed by PRESS 2015, the limited and nonuniform use of controlled vocabulary (MeSH/EMTREE) and proximity operators across heterogeneous database interfaces may have reduced search sensitivity to some extent, despite our efforts to mitigate this through extensive synonyms and multiple databases. Additionally, the search strategy was not formally peer-reviewed, and the study selection and data extraction processes did not involve a fully independent assessment by reviewers, which may have introduced bias. Methodologically, the evidence base relies heavily on surveys and observational designs, with no randomized controlled trials, which limits the ability to infer causality. Data collected through self-reports are also susceptible to bias. Additionally, a notable limitation across the quantitative studies was low response rates typical of GP surveys, which may have introduced nonresponse bias. Finally, we used Elicit AI [26] to support literature searching and data handling. As the underlying models and software builds are not disclosed and may change over time, and because full use of our shared Elicit workspaces requires Pro-tier access, exact replication of the AI-assisted components of the search may be limited, although the core database searches are fully reproducible from the reported strings and sources.

This systematic review of 65 studies, including 24,994 GPs, is, to our knowledge, the first to offer a Europe-wide synthesis of GPs’ experiences with digitalization across EHRs, telemedicine, DHTs, and AI-based tools, using a combined TAM-NPT effect-direction framework. By quantifying an implementation paradox—high perceived usefulness and near-universal use in the absence of positive usability ratings or full workflow integration—it extends earlier reviews that primarily examined patient outcomes, system-level metrics, or single digital interventions. For policy and practice, the findings argue for investing in interoperable, user-centered infrastructures, reimbursing digital workloads, embedding protected-time digital-competency training, and establishing governance standards that ensure equitable, safe, and relationally sustainable digital primary care in routine European general practice. Future research should extend beyond early-adopter countries and short-term evaluations to examine the long-term effects of digital transformation on workload, continuity of care, and health equity in diverse European primary care settings.