JĘZYKOZNAWSTWO nr 3(28)/2026
ISSN 2391-5137
https://doi.org/10.25312/j.10337
Praveen Malik https://orcid.org/0000-0002-8769-6631 Medical University of Lodz
Artificial intelligence mediated communication in Polish primary healthcare: A linguistic perspective
Artificial intelligence (AI) tools are increasingly embedded in Polish primary healthcare (Podstawowa Opieka Zdrowotna, POZ), reshaping how clinical encounters are conducted and documented. Based on a narrative review (2015–2026) of applied-linguistics and health-communication literature – spanning Conversation Analysis, narrative medicine, genre theory, and plain-language standards – and illustrated with Polish- language clinical examples, this article argues that AI systems act as algorithmic interactants that introduce evidential “turns” into consultations, shifting epistemic authority away from patients and clinicians; compress or flatten illness narratives by favouring categorical over experiential detail; and push clinical documentation toward standardised, template-based genres at the expense of stance-marking and interpersonal nuance. It further identifies Polish-specific linguistic features – free word order, rich inflection, aspectual distinctions, dialectal variation, and dense medical lexicon – that create systematic risks of symptom misrecognition and misclassification in NLP/ASR-based tools. Drawing these strands together, the article proposes an implementation framework combining plain-language documentation standards, CA-informed repair strategies, genre-sensitive documentation models, and corpus-based sociolinguistic validation, with the aim of reducing communicative and equity risks as AI is adopted in Polish primary care.
Keywords: Artificial Intelligence, Polish primary healthcare, conversation analysis, health communication, Natural Language Processing (NLP), narrative medicine, interactional linguistics, Automatic Speech Recognition (ASR)
Sztuczna inteligencja (AI) jest coraz częściej wykorzystywana w polskiej podstawowej opiece zdrowotnej (POZ), zmieniając sposób prowadzenia i dokumentowania wizyt lekarskich. Na podstawie przeglądu narracyj- nego (2015–2025) literatury z zakresu językoznawstwa stosowanego i komunikacji medycznej, obejmującego analizę konwersacyjną (CA), medycynę narracyjną, teorię gatunków oraz standardy prostego języka, artykuł analizuje wpływ AI na komunikację kliniczną. Ilustrowana polskojęzycznymi przykładami klinicznymi analiza wskazuje, że AI funkcjonuje jako algorytmiczny uczestnik interakcji, wprowadzający do konsultacji dowodowe „wypowiedzi”, przesuwający punkt ciężkości w kwestii autorytetu epistemicznego z pacjentów i klinicystów na inne podmioty. Jednocześnie systemy te spłaszczają narracje chorobowe, faworyzując in- formacje kategoryczne kosztem szczegółów doświadczeniowych oraz sprzyjają ujednolicaniu dokumentacji klinicznej, ograniczając ślady postawy i niuanse interpersonalne. Artykuł identyfikuje ponadto właściwości języka polskiego – szyk swobodny, bogatą fleksję, kategorię aspektu, zróżnicowanie dialektalne oraz gęsty leksykon medyczny – generujące systematyczne ryzyko błędnego rozpoznania lub klasyfikacji objawów przez narzędzia NLP/ASR. Na tej podstawie proponuje ramy wdrożeniowe łączące standardy prostego języka w dokumentacji, strategie naprawy oparte na CA, modele dokumentacji wrażliwe na gatunek oraz walidację socjolingwistyczną opartą na korpusach, mające ograniczyć ryzyko komunikacyjne i nierówności w dostępie do opieki w miarę wdrażania AI w polskiej POZ.
Słowa kluczowe: sztuczna inteligencja, komunikacja medyczna, analiza konwersacyjna, narracja medyczna, NLP, POZ, Automatic Speech Recognition (ASR)
Language is central to how symptoms are articulated, interpreted, and transformed into clinical decisions. With the growing integration of artificial intelligence (AI) tools into Polish primary care (POZ), communication is no longer shaped solely by patients and clinicians but increasingly co-constructed with algorithmic participants such as triage bots, CDSS systems, and automated documentation platforms (Topol, 2019; Chen et al., 2023; Dryl, 2024; Qin et al., 2025). As these systems enter the communicative space of the consultation, they influence not only workflow efficiency or diagnostic pathways but the linguistic fabric through which care is enacted.
A review of the current literature reveals a striking gap: while numerous studies ex- amine the technical, organisational, and ethical aspects of AI in Polish healthcare, very few approach the subject from a linguistic or discourse-analytic perspective. Existing work tends to focus on system performance, data extraction, usability, or policy frame- works. What remains underexplored – and what this article aims to address directly – is how AI reshapes the very language of care, including patient voice, narrative structure, conversational organisation, documentation genres, pragmatic framing, and, ultimately, equity and accessibility in communication.
By adopting a linguistics-centred perspective grounded in Conversation Analysis (Sche- gloff, 2007; Barnes, 2019), narrative medicine (Charon, 2006), genre theory (Swales, 1990; Biber, 1995), and plain-language principles (ISO, 2023), this article examines how AI systems reconfigure institutional discourse in Polish family medicine. It aims to articulate
not only what AI does for clinical efficiency, but what it does to communication: how it alters epistemic dynamics, compresses or filters patient narratives, changes repair possi- bilities, and standardises genre features in ways that impact the therapeutic relationship and the patient’s ability to express experience.
In addressing this gap, the article contributes a new conceptualisation of AI as an emerg- ing communicative actor in Polish healthcare – one whose linguistic footprint demands systematic examination.
This study employs a narrative review methodology covering the period 2015–2025, drawing together insights from applied linguistics, medical informatics, communication studies, sociolinguistics, and Polish-language NLP research (OECD, 2021; Szetela, 2025). Rather than focusing on quantitative counts of publications, the review emphasises conceptual depth, theoretical relevance, and the diversity of linguistic approaches used to analyse AI-mediated healthcare communication.
The selection process prioritised materials that explicitly intersect language and health- care, particularly those addressing:
AI-driven changes in interactional organisation and turn-taking,
shifting epistemic authority between patients, clinicians, and algorithmic systems,
transformations of illness narratives,
the evolution of clinical documentation genres,
patient comprehension and health literacy in technologically mediated settings, and
challenges posed by Polish morphosyntax, dialects, and speech variability.
The narrative review approach allows for the integration of heterogeneous sources – scholarly articles, technical reports, policy documents, linguistic corpora studies, and applied NLP research – facilitating a multidimensional understanding of AI’s communi- cative footprint. This methodology supports theoretical synthesis across traditions such as Conversation Analysis, narrative medicine, genre theory, pragmatics, and sociolinguistics (Chinta et al., 2024; Świącik, 2024). It also enables the identification of emerging patterns in how AI systems influence the structure of clinical encounters, reshape communicative norms, and reproduce or mitigate linguistic inequalities.
Through this synthesis, the methods aim not only to describe the current state of AI-me- diated communication in Polish primary care but to construct a robust interdisciplinary framework for analysing future developments at the intersection of language, technology, and healthcare.
AI inserts algorithmic turns within clinician–patient talk, shifting epistemic dynamics and producing hybrid participation frameworks (Greatbatch et al., 2005). CDSS (Clinical Decision Support System) output often occupies a third position in sequences, reframing how clinicians interpret or ratify patient accounts.
AI systems tend to compress or flatten patient narratives, prioritising categorical answers over metaphorical, temporal, or emotional detail (Mishler, 1984; Hydén, 1997). This diminishes narrative richness and diagnostic nuance.
AI systems typically lack self-repair mechanisms and cannot engage in negotiated clar- ification. As a result, misunderstandings introduced by the AI often persist (Jefferson, 2004; Schegloff, 2007).
AI-generated documentation tends toward template-based standardisation, which reduces stance-marking, interpersonal nuance, and contextual detail (Swales, 1990; Biber, 1995).
Polish presents a range of linguistic properties that significantly complicate the perfor- mance of NLP and ASR systems in clinical settings. These challenges stem not only from grammatical structure but also from sociolinguistic diversity, patient communication norms, and the highly specialised medical lexicon used in primary care (Kaźmierczyk, Kupis, Maj, 2022).
Polish allows considerable syntactic flexibility, meaning that crucial clinical information can appear in different positions without altering meaning. For example:
“W klatce mnie czasem boli”.
“Czasem mnie boli w klatce”.
“Boli mnie czasem w klatce”.
Although functionally equivalent for a clinician, AI systems may segment or prioritise different parts of these utterances, misinterpreting what is symptom, what is temporal framing, and what is emphasis. In triage contexts, such variation can produce incorrect symptom extraction.
Polish inflection affects nouns, verbs, adjectives, and pronouns, creating many sur- face forms for the same lexeme (Grochola-Szczepanek, 2020). For example, the noun “gorączka” (fever) can appear as “gorączki”, “gorączce”, “gorączką”, “gorączek” de- pending on case and number. ASR systems often struggle with:
older adults’ reduced articulation,
regional phonetic variants,
context-dependent stress patterns.
For AI-driven documentation, misrecognition of inflected forms may lead to incomplete or incorrect symptom coding. Engineering approaches such as synthetic-speech-augment- ed reinforcement learning have been proposed specifically to improve ASR robustness for Polish medical terminology (Czyżewski, 2023a).
Polish relies on verbal aspect to encode temporal structure. The distinction between “boli” (is hurting) and “zabolało” (suddenly began to hurt) signals clinically relevant information about onset, duration, and episodicity. Many AI models trained on general corpora cannot infer these distinctions.
For example:
“Boli mnie brzuch”. → ongoing condition; might indicate chronic issue.
“Zabolało mnie w nocy”. → acute onset; may require urgent evaluation.
Inaccurate interpretation of aspect can lead to inappropriate triage outcomes or mis- leading clinical summaries.
Poland features notable regional varieties such as Silesian (Śląski), Highland Podhale speech (podhalański), Kashubian-influenced Polish (kaszubski), and Eastern borderland features (Wyderka, 2014). Patients may say:
“Godó sie, że żech je łogłupiała od tego bólu” (Śląski).
“Jo mom ciśnienie wysoki” (Podhale).
Standard NLP models frequently misrecognise these forms, particularly in speech.
This risks:
missing key symptoms,
misclassifying emotional stance,
or flagging irrelevant keywords.
Older speakers and individuals with lower health literacy often use idiosyncratic vo- cabulary or localised metaphorical expressions to describe bodily states, which AI systems frequently fail to decode.
Medical Polish is morphologically dense and often intimidating for patients. Terms like “niewydolność krążeniowo-oddechowa” or “zaostrzenie przewlekłej obturacyjnej choroby
płuc” overwhelm many speakers. AI-generated messages that adopt or reproduce these forms without simplification can hinder understanding.
Additionally, patients frequently use colloquial substitutes:
“zator” instead of “zakrzepica”,
“zakwasy” for various musculoskeletal pains,
“skurcz w sercu” for palpitations.
AI systems trained on formal medical datasets may interpret such expressions literally, leading to mismatches between patient intention and system categorisation.
Polish intonation and pragmatic markers carry substantial meaning. Utterances like:
“No… boli, ale to chyba nic takiego”. (mitigating stance),
“Strasznie mnie nasiliło!” (intensification),
“Jakoś tak mnie kłuje czasem…” (vague evidential marking),
encode uncertainty, fear, or minimisation strategies. These pragmatic nuances help clini- cians gauge severity and patient concern but are often lost in ASR–NLP pipelines.
Tab. 1. Polish-specific linguistic phenomena and their AI/NLP implications in primary care
Linguistic phenomenon | Example (Polish) | AI/NLP risk | Clinical consequence |
Free word order | Boli mnie czasem w klatce. / Czasem mnie boli w klatce. | Inconsistent segmen- tation of symptom vs. temporal framing | Incorrect symptom extraction in triage |
Rich inflection | gorączka → gorączki, gorączce, gorączką | Surface-form mi- smatch in ASR/entity recognition | Incomplete or incor- rect symptom coding |
Verbal aspect | boli (ongoing) vs. zabolało (acute onset) | Failure to encode onset/duration di- stinctions | Misleading triage urgency |
Dialect / regiolect | Śląski, Podhalański regional forms | Misrecognition, parti- cularly in speech | Missed symptoms; misclassified emotio- nal stance |
Medicalised lexicon / colloquialisms | zakwasy (colloquial) vs. formal musculo- skeletal terminology | Literal interpretation of colloquial terms | Mismatch between patient intention and system categorisation |
Prosody / pragmatics | mitigating, intensi- fying, and evidential markers (e.g. “No… boli, ale to chyba nic takiego”) | Loss of stance/uncer- tainty cues in ASR– NLP pipelines | Misjudged severity or patient concern |
Source: Author’s own elaboration, based on Grochola-Szczepanek (2020), Wyderka (2014), Malisz, Foremski, Kul (2024), and Pondel-Sycz et al. (2025).
AI’s entry into Polish primary care provides a novel testing ground for applied linguistics, extending multiple theoretical traditions:
AI tools can be understood as algorithmic interactants insofar as their outputs are routinely introduced into clinical encounters and treated as evidential contributions that influence subsequent human talk and decision-making. Unlike human participants, algorithmic outputs typically
make contributions without being accountable in interactional terms,
produce recommendations without direct access to the lived, embodied context of the consultation, and
reconfigure epistemic gradients by introducing external evidence that clinicians must interpret or ratify.
This set of dynamics challenges core Conversation Analysis assumptions about mutual orientation, sequential organisation, and repairability: when one “participant” cannot negotiate meaning or accept responsibility for its moves, intersubjective work must be redistributed across human actors. Recent empirical and conceptual work on clinical decision support and human–AI interaction documents these shifts and argues for re- framing decision support as part of clinicians’ reasoning processes rather than as neutral background tools (van Baalen, Boon, Verhoef, 2021; Funer, Wiesing, 2024; Jiang et al., 2024). Scholarship on algorithmic accountability and epistemic status further explains why algorithmic contributions are both powerful and problematic in interactional settings (Diakopoulos, 2016; Ferrario et al., 2024), while literature on trust and automation clari- fies how humans come to treat non-human outputs as warranting deference or scepticism (Chiou, Lee, 2023).
The growing use of generative-AI and NLP for clinical documentation is reshaping the genre of illness narratives in institutional medicine. Recent studies demonstrate that large language models can convert free-text pathology reports into structured, coded data with high fidelity – effectively compressing rich narrative descriptions into algo- rithm-ready diagnostic vectors (Balasubramanian et al., 2025; Shahid et al., 2025). In this transformation, the patient’s experiential story becomes subordinate to structured fields optimised for computational processing: metaphorical language is attenuated, temporal and causal sequences are regularised, and affective or evaluative elements tend to be stripped away in favour of pattern-recognisable biomedical features. This reflects not only a change in documentation format but the emergence of a new institutional narrative genre shaped by computational constraints, extending earlier critiques of how medical records conventionalise and constrain patient storytelling (Mishler, 1984; Hydén, 1997; Semino, 2010).
Although current evidence stems primarily from pathology and other written-report domains, the implications for spoken clinical encounters are significant. As AI-mediated documentation systems are integrated into real-time workflow, there is a risk that the narra- tive material clinicians attend to will increasingly mirror what AI can capture – structured, symptom-centred, and diagnostically actionable – rather than the broader experiential and metaphorical dimensions emphasised in narrative medicine (Charon, 2006). This raises the possibility that algorithm-ready illness accounts may shape both how clinicians record stories and how patients learn to present them.
Despite these developments, empirical research on how AI transforms illness narratives during real clinical interactions remains limited. No published conversation-analytic stud- ies yet examine how generative-AI suggestions or summaries intervene in the sequential organisation of consultations, nor how algorithmic outputs influence the construction, negotiation or repair of patient stories in situ. Future work should therefore combine Conversation Analysis, linguistic ethnography, and clinical NLP evaluation to trace how spoken narratives are filtered, reformatted or truncated through AI-mediated pathways. Longitudinal studies could further examine whether patients adapt their storytelling practices to align with the expectations of digital triage systems or AI-supported docu- mentation tools. Addressing these gaps will be crucial for assessing whether emerging algorithm-ready genres enhance clinical reasoning or inadvertently diminish narrative richness and patient agency.
Electronic medical records (EMRs) have reshaped the genre of clinical documentation by shifting from narrative, free-text accounts toward structured, template-based formats designed for standardisation and data reuse (Rosenbloom et al., 2010; Rosenbloom et al., 2011). These systems privilege coded entries, checklists, and predefined fields, reducing the expressive and narrative dimensions of clinical writing traditionally used to capture temporal nuance, uncertainty, and patient-centred detail.
The introduction of artificial intelligence further accelerates this shift by creating hybrid documentation genres that combine clinician input with automated structuring, annota- tion and summarisation. AI-assisted tools increasingly extract clinical entities, normalise terminology, and generate draft notes or summaries (Perkins et al., 2024; Shemtob et al., 2025), thereby reshaping what counts as clinically salient information. As a result, the boundaries of authorship, narrative form, and evidential framing are no longer deter- mined solely by human clinicians but co-constructed with algorithmic systems. A recently developed Polish voice-input platform illustrates this shift in practice, combining speech recognition and clinical context analysis to generate structured EHR entries directly from clinician speech (Szplit et al., 2026).
Genre theory emphasises that institutional genres shape and constrain patterns of think- ing,communication, and professional accountability (Swales, 1990; Biber, Conrad,
2009). In the clinical domain, AI-mediated documentation introduces several significant implications.
First, medico-legal accountability becomes more complex as authorship is distributed across clinicians and machine-assisted drafting systems, raising questions about respon- sibility for inaccuracies or omissions (Cabitza, Rasoini, Gensini, 2017; Naik et al., 2022). Second, documentation conventions influence clinical reasoning itself: research demonstrates that the structure and linguistic organisation of clinical records shape how evidence is framed, sequenced, and interpreted within the diagnostic process (Drew,
Chatwin, Collins, 2001; Maynard, Heritage, 2005; Rosenbloom et al., 2011).
Finally, patient access to clinical records – now common in many healthcare systems – has been shown to improve engagement and understanding (Walker et al., 2019). However, it remains uncertain how AI-generated notes will affect transparency and comprehensi- bility, particularly if automated systems prioritise coded data over narrative clarity; this possibility requires dedicated empirical study.
Collectively, these developments highlight the need for applied linguistic, sociolin- guistic, and discourse-analytic approaches to examine how AI is generating new clinical subgenres and transforming the communicative ecology of medical documentation. Such research is essential to understand how hybrid human–machine authorship reshapes narrative structure, evidentiality, interactional accountability, and patient participation in clinical communication.
AI foregrounds sociolinguistic inequalities: dialect speakers, older patients, and individuals with lower health literacy risk misrecognition or misclassification. This connects linguis- tic theory to digital health equity, establishing new cross-disciplinary research domains.
AI output in clinical documentation should follow standards of plain language, such as those codified in ISO 24495-1:2023 – prioritising clarity, readability, and actionable structure for the intended audience. Evidence from health communication research shows that using plain language and simplifying medical text improves patient comprehension and reduces misinterpretation of clinical information (Stableford, Mettger, 2007; Warde et al., 2018). In healthcare settings, plain-language writing has been proposed as a core competency for clinicians to mitigate low health literacy among patients (Warde et al., 2018). Thus, when generating patient-facing communications or summaries, AI systems should avoid unnecessarily complex syntax, dense nominalisations or opaque medical jargon and, whenever appropriate, adopt patient-centred metaphors and straightforward phrasing consistent with plain-language guidelines.
Conversation Analysis shows that repair is a fundamental interactional practice by which participants manage problems of speaking, hearing, and understanding, thereby sustaining intersubjectivity (Schegloff, 2007; Albert, de Ruiter, 2018). In clinical encounters, repair enables clarity, alignment, and mutual orientation – all crucial for safe and effective interaction. In the context of AI-mediated systems, where algorithmic outputs may not respond to clarification or negotiation, clinicians must adopt compensatory strategies. Training for AI-mediated consultations should therefore include prompts that check patient understanding (e.g., “Jak Pan/Pani rozumie komunikat systemu?”), and strate- gies for reframing ambiguous or opaque AI-generated messages into patient-accessible language. Although no published study yet documents how AI-human repair sequences unfold in clinical consultation, the combination of CA insights on interactional repair and emerging human–AI alignment research suggests that clinician-led repair work will become even more important to maintain interactional symmetry and reduce epistemic confusion (Heritage, Maynard, 2006; Heritage, 2012).
Clinical documentation is a professional genre whose structure influences clinical reason- ing (Swales, 1990; Rosenbloom et al., 2011). AI should support – not overwrite – clinician narrative practices. Research shows that narrative and stance expressions carry diagnostic and interpersonal functions (Charon, 2006; Biber, Conrad, 2009). Systems must:
preserve patient quotations (Charon, 2006),
maintain temporal and experiential detail (Mishler, 1984),
mark stance and uncertainty explicitly (Perkins et al., 2024),
allow clinicians to override summaries (Perkins et al., 2024).
This ensures documentation retains its interpersonal and evidentiary functions. Ap- plied work on voice-generated documentation within intelligent clinical forms in Poland illustrates this genre-sensitive approach in practice (Czyżewski, 2023b).
Bias in NLP systems correlates with a lack of dialectal and sociolinguistic representation (Blodgett et al., 2020). Validation must include:
dialectal corpora — e.g. Korpus Spiski, documenting the speech of inhabitants of Pol- ish Spisz across age cohorts (Grochola-Szczepanek et al., 2019).
age-specific / generational speech variation – Korpus Spiski was created with infor- mants from “the youngest and middle generations”, as well as older speakers, making it suitable for generational / sociolinguistic variation studies (Grochola-Szczepanek et al., 2019).
sociophonetic / phonetic diversity – e.g. PRODIS, a large, publicly available Polish speech database designed for phoneme-level analysis, suitable for phonetic and pro- sodic variation studies (Malisz et al., 2024).
healthcare-specific Polish / domain-specific speech data – e.g. ADMEDVOICE, a Polish medical-speech dataset collected in realistic (possibly noisy) conditions and including domain-specific vocabulary (drugs, illnesses), which can support develop- ment and validation of medical-domain NLP/ASR systems (Czyżewski et al., 2025; Pondel-Sycz et al., 2025) and emerging adaptive AI platforms that combine speech recognition with diagnostic data structuring for Polish clinical decision support (Szplit et al., 2025).
Such comprehensive coverage – dialectal variation, generational variation, phonetic/ sociophonetic diversity, and domain-specific (medical) language – reduces systemic bias by ensuring representation of diverse sociolinguistic backgrounds and registers, particularly for vulnerable or underrepresented populations.
Research on clinician–patient communication demonstrates that the tone, wording, and structure of medical information significantly influence patient understanding, illness expe- rience, and adherence to treatment. Polish studies show that communication that is respect- ful, clear, non-authoritarian, and oriented toward the patient’s knowledge and emotional state supports a more positive experience of illness and improves trust in medical staff (Kaźmierczak, 2011). Likewise, analyses of health and medical communication in public and institutional contexts indicate that overly formal or authoritative language can alienate recipients and decrease the effectiveness of educational messages (Białek-Szwed, 2019). Work on interpersonal communication in healthcare organisations further underscores the need for clarity, empathy, and avoidance of unnecessary medical jargon to ensure that messages are understood and socially appropriate (Meller, Milik, 2014).
Large-scale empirical evidence confirms this: a meta-analysis of over three decades of research shows a strong association between high-quality physician communication and patient adherence (Haskard Zolnierek, DiMatteo, 2009). Similarly, collaboration-oriented communication – framing recommendations clearly and involving the patient in under- standing why an action matters – has been shown to significantly improve adherence to medical advice (Arbuthnott, Sharpe, 2009).
Therefore, AI-generated patient materials should:
avoid face-threatening or overly authoritative language, consistent with findings on the importance of respectful and non-hierarchical communication in healthcare (Kaźmierczak, 2011; Białek-Szwed, 2019);
provide explicit framing of recommendations – using clear, actionable structures such as “co należy zrobić”, “jak to zrobić”, and “dlaczego ma to znaczenie” – which supports comprehension and treatment uptake (Arbuthnott, Sharpe, 2009);
use metaphors, analogies, and everyday language explanations to make complex medical concepts cognitively accessible and reduce the communicative distance be- tween expert and patient (Meller, Milik, 2014);
avoid excessive medicalised morphology and specialist terminology, replacing it with plain language and patient-centred explanations to minimise misunderstanding and confusion (Białek-Szwed, 2019).
Embedding pragmatics-informed communication – respectful tone, clarity, explicit guidance, and linguistic accessibility – ensures that patient-targeted materials are social- ly appropriate, comprehensible across different levels of health literacy, and conducive to patient engagement and adherence to recommended care (Arbuthnott, Sharpe, 2009; Haskard Zolnierek, DiMatteo, 2009).
AI may enhance efficiency but risks reducing patient narrative richness (Hydén, 1997), weakening rapport (Blease et al., 2019), amplifying misunderstandings when outputs are vague or numeric (Peters et al., 2007), and shifting authority toward computational systems (Braun et al., 2021).
Potential benefits include standardisation and reduced documentation burden (Lee et al., 2024). Risks include:
misclassification and performance degradation due to biased or stigmatizing language and linguistic ambiguity (Liu, Wang, Gao, 2023),
clinician overreliance on system-generated interpretations and risk of reduced human oversight (Hossain et al., 2023; Barwise et al., 2024),
loss of soft clinical cues or nuanced patient-history that often emerge only through human storytelling, rapport, and in-person dialogue (Barwise et al., 2024).
Unequal linguistic recognition creates unequal care. Patients with dialects, low literacy, or atypical speech patterns risk being misunderstood or deprioritised by AI systems (Blodgett et al., 2020; Chinta et al., 2024). Ensuring equity requires linguistic calibration and continuous monitoring (Rajkomar et al., 2018).
AI influences not only interaction but workflow, policy, and medico-legal accountability. Institutions must establish standards for:
attribution of responsibility (Cabitza et al., 2017),
documentation accuracy (Rosenbloom et al., 2011),
transparency of decision pathways (Regulation (EU) 2024/1689).
AI is not merely a technical tool but a new linguistic participant in Polish primary care. Its presence reshapes the texture of clinical encounters: how stories are told, how evidence is constructed, and how decisions are communicated. To ensure equitable, patient-centred care, implementation must be grounded in linguistic theory – recognising that communi- cation quality is as important as technological accuracy.
Polish primary care stands at a crossroads. With linguistically informed design, AI can enhance clarity, accessibility, and interactional coherence. Without such grounding, it risks deepening disparities, eroding rapport, and diminishing the patient’s voice. A robust linguistic framework – supported by CA, narrative theory, genre analysis, and sociolin- guistics – offers a pathway toward AI that strengthens rather than weakens the human core of healthcare.
Albert S., de Ruiter J.P. (2018), Repair: The interface between interaction and cognition, “Topics in Cognitive Science”, vol. 10(2), pp. 279–313, https://doi.org/10.1111/tops.12339
Arbuthnott A., Sharpe D. (2009), The effect of physician–patient collaboration on patient adherence in non-psychiatric medicine, “Patient Education and Counseling”, vol. 77(1),
pp. 60–67, https://doi.org/10.1016/j.pec.2009.03.022
Baalen van S., Boon M., Verhoef P. (2021), From clinical decision support to clinical reasoning support systems, “Journal of Evaluation in Clinical Practice”, vol. 27(3), pp. 520–528, https:// doi.org/10.1111/jep.13541
Balasubramanian J.B., Adams D., Roxanis I., Berrington de Gonzalez A., Coulson P., Almeida J.S., García-Closas M. (2025), Leveraging large language models for structured information extraction from pathology reports, “Journal of Pathology Informatics”, vol. 19, article 100521, https://doi.org/10.1016/j.jpi.2025.100521
Barnes R.K. (2019), Conversation analysis of communication in medical care: Description and beyond, “Research on Language & Social Interaction”, vol. 52(3), pp. 300–315, https:// doi.org/10.1080/08351813.2019.1631056
Barwise A.K., Curtis S., Diedrich D.A., Pickering B.W. (2024), Using artificial intelligence to promote equitable care for inpatients with language barriers and complex medical needs: Clinical stakeholder perspectives, “Journal of the American Medical Informatics Association”, vol. 31(3), pp. 611–621, https://doi.org/10.1093/jamia/ocad224
Białek-Szwed O. (2019), Komunikowanie o zdrowiu i medycynie – rzeczywistość a przekazy medialne, “Teka Komisji Politologii i Stosunków Międzynarodowych”, vol. 14(1), pp. 61–71, https://doi.org/10.17951/teka.2019.14.1.61-71
Biber D. (1995), Dimensions of register variation: A cross-linguistic comparison, Cambridge: Cambridge University Press.
Biber D., Conrad S. (2009), Register, genre, and style, Cambridge: Cambridge University Press.
Blease C., Kaptchuk T.J., Bernstein M.H., Mandl K.D., Halamka J.D., DesRoches C.M. (2019), Artificial intelligence and the future of primary care: Exploratory qualitative study of UK general practitioners’ views, “Journal of Medical Internet Research”, vol. 21(3), arti- cle e12802, https://doi.org/10.2196/12802
Blodgett S.L., Barocas S., Daumé III H., Wallach H. (2020), Language (technology) is power: A critical survey of “bias” in NLP, [in:] D. Jurafsky, J. Chai, N. Schluter, J. Tetreault (eds.), Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, Association for Computational Linguistics, Online, pp. 5454–5476, https://doi.org/10.18653/ v1/2020.acl-main.485
Braun M., Hummel P., Beck S., Dabrock P. (2021), Primer on an ethics of AI-based decision support systems in the clinic, “Journal of Medical Ethics”, vol. 47(12), article e3, https://doi. org/10.1136/medethics-2019-105860
Cabitza F., Rasoini R., Gensini G.F. (2017), Unintended consequences of machine learning in medicine, “JAMA”, vol. 318(6), pp. 517–518, https://doi.org/10.1001/jama.2017.7797
Charon R. (2006), Narrative medicine: Honoring the stories of illness, Oxford: Oxford Uni- versity Press.
Chen Z., Liang N., Zhang H., Li H., Yang Y., Zong X., Chen Y., Wang Y., Shi N. (2023), Har- nessing the power of clinical decision support systems: Challenges and opportunities, “Open Heart”, vol. 10(2), article e002432, https://doi.org/10.1136/openhrt-2023-002432
Chinta S.V., Wang Z., Zhang X., Viet T.D., Kashif A., Smith M.A., Zhang W. (2024), AI-
-driven healthcare: A survey on ensuring fairness and mitigating bias, “arXiv”, https://doi. org/10.48550/arXiv.2407.19655
Chiou E.K., Lee J.D. (2023), Trusting automation: Designing for responsivity and resilience, “Human Factors”, vol. 65(1), pp. 137–165, https://doi.org/10.1177/00187208211009995
Czyżewski A. (2023a), Optimizing medical personnel speech recognition models using speech synthesis and reinforcement learning, “Journal of the Acoustical Society of America”, vol. 154(4_Suppl), pp. A202–A203, https://doi.org/10.1121/10.0023271
Czyżewski A. (2023b), Głosowe tworzenie dokumentacji medycznej w inteligentnych for- mularzach [Voice creation of medical documentation in intelligent forms], paper presented at Artificial Intelligence in Medicine Panel, Nauka dla Społeczeństwa National Congress, Warszawa, 3.06.2023.
Czyżewski A., Cygert S., Marciniuk K., Szczodrak M., Harasimiuk A., Odya P., Galanina M., Szczuko P., Kostek B., Graff B., Szplit D., Budzisz M., Narkiewicz K. (2025), A comprehensive Polish medical speech dataset for enhancing automatic medical dictation, “Scientific Data”, vol. 12, article 1436, https://doi.org/10.1038/s41597-025-05776-1
Diakopoulos N. (2016), Accountability in algorithmic decision making, “Communications of the ACM”, vol. 59(2), pp. 56–62, https://doi.org/10.1145/2844110
Drew P., Chatwin J., Collins S. (2001), Conversation analysis: A method for research into interactions between patients and health-care professionals, “Health Expectations”, vol. 4(1),
pp. 58–70, https://doi.org/10.1046/j.1369-6513.2001.00125.x
Dryl T. (2024), Poprawa komunikacji z pacjentem poprzez chatboty oparte na sztucznej inteligencji (AI): Spostrzeżenia empiryczne i zalecenia menedżerskie, “Problemy Jakości”, vol. 56(3), pp. 25–31, https://doi.org/10.15199/46.2024.3.4
Ferrario A., Facchini A., Termine A. (2024), Experts or authorities? The strange case of the presumed epistemic superiority of artificial intelligence systems, “Minds and Machines”, vol. 34(3), article 30, https://doi.org/10.1007/s11023-024-09681-1
Funer F., Wiesing U. (2024), Physician’s autonomy in the face of AI support: Walking the eth- ical tightrope, “Frontiers in Medicine”, vol. 11, article 1324963, https://doi.org/10.3389/ fmed.2024.1324963
Greatbatch D., Hanlon G., Goode J., O’Cathain A., Strangleman T., Luff D. (2005), Telephone triage, expert systems and clinical expertise, “Sociology of Health & Illness”, vol. 27(6),
pp. 802–830, https://doi.org/10.1111/j.1467-9566.2005.00475.x
Grochola-Szczepanek H. (2020), Warianty morfologiczne czasownika „być” w mowie miesz- kańców Spisza na podstawie danych korpusowych, “LingVaria”, vol. 15(1), pp. 147–162, https://doi.org/10.12797/LV.15.2020.29.10
Grochola-Szczepanek H., Górski R.L., von Waldenfels R., Woźniak M. (2019), Korpus języka mówionego mieszkańców Spisza, “LingVaria”, vol. 14(1), pp. 165–180, https://doi. org/10.12797/LV.14.2019.27.11
Haskard Zolnierek K.B., DiMatteo M.R. (2009), Physician communication and patient adhe- rence to treatment: A meta-analysis, “Medical Care”, vol. 47(8), pp. 826–834, https://doi. org/10.1097/MLR.0b013e31819a5acc
Heritage J. (2012), The epistemic engine: Sequence organization and territories of knowledge, “Research on Language & Social Interaction”, vol. 45(1), pp. 30–52, https://doi.org/10.108 0/08351813.2012.646685
Heritage J., Maynard D.W. (eds.) (2006), Communication in medical care: Interaction between primary care physicians and patients, Cambridge: Cambridge University Press.
Hossain E., Rana R., Higgins N., Soar J., Barua P.D., Pisani A.R., Turner K. (2023), Natural language processing in electronic health records in relation to healthcare decision-making: A systematic review, “Computers in Biology and Medicine”, vol. 155, article 106649, https:// doi.org/10.1016/j.compbiomed.2023.106649
Hydén L.-C. (1997), Illness and narrative, “Sociology of Health & Illness”, vol. 19(1),
pp. 48–69, https://doi.org/10.1111/j.1467-9566.1997.tb00015.x
ISO (2023), ISO 24495-1:2023. Plain language – Part 1: Governing principles and guidelines, Geneva: International Organization for Standardization.
Jefferson G. (2004), Glossary of transcript symbols with an introduction, [in:] G.H. Lerner (ed.), Conversation analysis: Studies from the first generation, Amsterdam–Philadelphia: John Benjamins Publishing Company, pp. 13–31, https://doi.org/10.1075/pbns.125.02jef
Jiang T., Sun Z., Fu S., Lv Y. (2024), Human-AI interaction research agenda: A user-cen- tered perspective, “Data and Information Management”, vol. 8(4), article 100078, https://doi. org/10.1016/j.dim.2024.100078
Kaźmierczak A. (2011), Wpływ komunikacji z personelem medycznym na doświadczanie choroby przez pacjentów, “Homo Communicativus”, vol. 1(6), pp. 133–140.
Kaźmierczyk P., Kupis M., Maj M. (2022), Biała Księga AI w praktyce klinicznej: Stosowanie sztucznej inteligencji przy udzielaniu świadczeń zdrowotnych, Warszawa: wZdrowiu.
Lee C., Britto S., Diwan K. (2024), Evaluating the impact of artificial intelligence (AI) on clinical documentation efficiency and accuracy across clinical settings: A scoping review, “Cureus”, vol. 16(11), article e73994, https://doi.org/10.7759/cureus.73994
Liu Y., Wang W., Gao G.G., Agarwal R. (2023), Echoes of biases: How stigmatizing language affects AI performance, “arXiv”, https://doi.org/10.48550/arXiv.2305.10201
Malisz Z., Foremski J., Kul M. (2024), PRODIS – a speech database and a phoneme-based language model for the study of predictability effects in Polish, [in:] N. Calzolari, M.-Y. Kan,
V. Hoste, A. Lenci, S Sakti, N. Xue (eds.), Proceedings of the 2024 Joint International Con- ference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024), Torino: ELRA and ICCL, pp. 13068–13073.
Maynard D.W., Heritage J. (2005), Conversation analysis, doctor–patient interaction and medical communication, “Medical Education”, vol. 39(4), pp. 428–435, https://doi.org/10.1111/ j.1365-2929.2005.02111.x
Meller D., Milik A. (2014), Komunikacja interpersonalna w zakładach opieki zdrowotnej, “Zarządzanie i Finanse”, vol. 12(2), pp. 133–144.
Mishler E.G. (1984), The discourse of medicine: Dialectics of medical interviews, Norwood, NJ: Ablex Publishing Corporation.
Naik N., Hameed B.M. Z., Shetty D.K., Swain D., Shah M., Paul R., Aggarwal K., Ibrahim S., Patil V., Smriti K., Shetty S., Rai B.P., Chlosta P., Somani B.K. (2022), Legal and ethical consideration in artificial intelligence in healthcare: Who takes responsibility?, “Frontiers in Surgery”, vol. 9, article 862322, https://doi.org/10.3389/fsurg.2022.862322
OECD (2021), Health at a glance 2021: OECD indicators, Paris: OECD Publishing, https:// doi.org/10.1787/ae3016b9-en
Perkins S.W., Muste J.C., Alam T., Singh R.P. (2024), Improving clinical documentation with artificial intelligence: A systematic review, “Perspectives in Health Information Management”, vol. 21(2), article 1d.
Peters E., Hibbard J., Slovic P., Dieckmann N. (2007), Numeracy skill and the communication, comprehension, and use of risk-benefit information, “Health Affairs”, vol. 26(3), pp. 741–748, https://doi.org/10.1377/hlthaff.26.3.741
Pondel-Sycz K., Bilski P., Bobiński P., Morzyński L., Lewandowski M., Kozłowski E., Szc- zepański G., Jasiński M., Makarewicz G., Pietrzak A.P., Buchowicz A., Mazurek P., Bilski A., Olejnik J., Olejnik I. (2025), A comparative study of deep End-to-End Automatic Speech Rec- ognition models for doctor-patient conversations in Polish in a real-life acoustic environment, “International Journal of Electronics and Telecommunications”, vol. 71(3), pp. 1–8, https:// doi.org/10.24425/ijet.2025.153609
Qin J., Nan Y., Li Z., Meng J. (2025), Effectiveness of communication competence in AI conversational agents for health: Systematic review and meta-analysis, “Journal of Medical Internet Research”, vol. 27, article e76296, https://doi.org/10.2196/76296
Rajkomar A., Hardt M., Howell M.D., Corrado G., Chin M.H. (2018), Ensuring fairness in machine learning to advance health equity, “Annals of Internal Medicine”, vol. 169(12),
pp. 866–872, https://doi.org/10.7326/M18-1990
Regulation (EU) 2024/1689 (2024) Artificial Intelligence Act.
Rosenbloom S.T., Stead W.W., Denny J.C., Giuse D., Lorenzi N.M., Brown S.H., Johnson
K.B. (2010), Generating clinical notes for electronic health record systems, “Applied Clinical Informatics”, vol. 1(3), pp. 232–243, https://doi.org/10.4338/ACI-2010-03-RA-0019
Rosenbloom S.T., Denny J.C., Xu H., Lorenzi N., Stead W.W., Johnson K.B. (2011), Data from clinical notes: A perspective on the tension between structure and flexible documentation, “Journal of the American Medical Informatics Association”, vol. 18(2), pp. 181–186, https:// doi.org/10.1136/jamia.2010.007237
Schegloff E.A. (2007), Sequence organization in interaction: A primer in conversation analysis I, Cambridge: Cambridge University Press.
Semino E. (2010), Descriptions of pain and metaphor, “Metaphor and Symbol”, vol. 25(4),
pp. 205–226, https://doi.org/10.1080/10926488.2010.510926
Shahid F., Hsu M.-H., Chang Y.-C., Jian W.-S. (2025), Using generative AI to extract struc- tured information from free text pathology reports, “Journal of Medical Systems”, vol. 49(1), article 36, https://doi.org/10.1007/s10916-025-02167-2
Shemtob L., Nouri A., Harvey-Sullivan A., Qiu C.S., Martin J., Martin M., Noden S., Rob T., Neves A.L., Majeed A., Clarke J., Beaney T. (2025), Comparing artificial intelligence- vs clinician-authored summaries of simulated primary care electronic health records, “JAMIA Open”, vol. 8(4), article ooaf082, https://doi.org/10.1093/jamiaopen/ooaf082
Stableford S., Mettger W. (2007), Plain language: A strategic response to the health literacy challenge, “Journal of Public Health Policy”, vol. 28(1), pp. 71–93, https://doi.org/10.1057/ palgrave.jphp.3200102
Swales J.M. (1990), Genre analysis: English in academic and research settings, Cambridge: Cambridge University Press.
Szetela P.P. (2025), Coordinated care in primary health care in Poland in 2021–2024: First experiences, “Medycyna Ogólna i Nauki o Zdrowiu”, vol. 31(3), pp. 166–176, https://doi. org/10.26444/monz/206089
Szplit D., Czyżewski A., Graff B., Szyndler A., Pałczyński P., Siemiński M., Glaner N., Stefaniak T., Bogdan J., Słomińska K., Dąbkowska A., Budzisz M., Zielonka M., Galani- na M., Kurowski A., Zaporowski S., Kotus J., Narkiewicz K. (2025), Adaptive AI platform for processing Polish medical language: Speech recognition, diagnostic data structuring, and enhanced therapeutic decision support, Research Square, preprint, https://doi.org/10.21203/ rs.3.rs-6975175/v1
Szplit D., Czyżewski A., Graff B., Szyndler A., Pałczyński P., Siemiński M., Glaner N., Ste- faniak T., Bogdan J., Słomińska K., Dąbkowska A., Budzisz M., Zielonka M., Cieślak D., Zaporowski S., Kotus J., Narkiewicz K. (2026), Developing an AI platform for voice-input medical records in Polish: A digital application for optimisation of physicians’ workload, “Scientific Reports”, vol. 16, article 21514, https://doi.org/10.1038/s41598-026-50490-6
Świącik R. (2024), Wpływ pandemii COVID-19 na rozwój sztucznej inteligencji w medycynie, “Inżynier i Fizyk Medyczny”, vol. 13(6), pp. 591–593.
Topol E.J. (2019), High-performance medicine: The convergence of human and artificial intel- ligence, “Nature Medicine”, vol. 25(1), pp. 44–56, https://doi.org/10.1038/s41591-018-0300-7
Walker J., Leveille S., Bell S., Chimowitz H., Dong Z., Elmore J.G., Fernandez L., Fossa A., Gerard M., Fitzgerald P., Harcourt K., Jackson S., Payne T.H., Perez J., Shucard H., Stametz R., DesRoches C., Delbanco T. (2019), OpenNotes after 7 years: Patient experiences with ongo- ing access to their clinicians’ outpatient visit notes, “Journal of Medical Internet Research”, vol. 21(5), article e13876, https://doi.org/10.2196/13876
Warde F., Papadakos J., Papadakos T., Rodin D., Salhia M., Giuliani M. (2018), Plain language communication as a priority competency for medical professionals in a globalized world, “Canadian Medical Education Journal”, vol. 9(2), pp. e52–e59, https://doi.org/10.36834/ cmej.36848
Wyderka B. (2014), O rozwoju polskich dialektów, “Poznańskie Studia Polonistyczne. Seria Językoznawcza”, vol. 21(2), pp. 103–113.