English

For therapists, the value of a new AI model is not its place in a benchmark ranking. It is whether it can make everyday work easier and leave more time for clients. Gemini 4 Argon is designed to handle complex tasks and large amounts of information. This could make it useful for tasks that involve several steps, though its value in therapy has not yet been established by clinical research. One practical use could be organizing lengthy, de-identified notes. A therapist might ask Argon to group observations under themes such as communication, daily activities, or participation. The therapist could then check the summary against the original notes and use it as a starting point for a report. It may also help adapt information for different readers. A therapist could ask it to explain a clinical term in plain English for a family, or make a draft report easier to understand. This could be especially helpful when families find professional language difficult. The therapist should review the wording to ensure it is accurate, respectful, and suitable for that family. When planning sessions, a therapist might use Argon to generate several activity ideas for a specific goal, such as helping a child practise dressing or join a group game. The therapist can then select and adapt ideas based on the client’s abilities, interests, culture, and everyday environment. The useful contribution is a wider range of possibilities, not a ready-made treatment plan. Argon may also help with work that takes time but does not require the therapist’s full clinical attention, such as creating a report outline, preparing questions for a team meeting, or organizing research notes. Its ability to manage multi-step tasks may help connect parts of this work. However, therapists should test these uses in their own setting rather than assume the model will save time or improve care. What can AI not do? It cannot build the trust that grows through a therapeutic relationship, notice every change in a client’s expression or behaviour, or understand the full meaning of a family’s experiences. It can suggest what to consider, but it cannot take responsibility for deciding what matters most to a particular person. Nor should Argon be treated as uniquely capable simply because it is new or performs well on company-reported tests. Those results do not show that it is more accurate than other AI tools for clinical tasks. The model may misunderstand information or produce a confident but incorrect answer, so its work needs human review. Privacy and accountability must guide its use. Therapists should not enter identifiable client information unless their organization has approved the tool and its privacy protections. They should be open with clients when AI has helped prepare information, and remain responsible for decisions made using its suggestions. Gemini 4 Argon may be most useful as a capable assistant for organizing, simplifying, and brainstorming, not as a therapist. Research should now examine whether it genuinely reduces paperwork, improves communication, or helps clinicians prepare for care. If it does, its success will be measured not by how much it can do alone, but by how well it supports therapists while keeping human judgment and relationships at the centre.

A child with developmental coordination disorder (DCD) may find handwriting, dressing, playing sports, or keeping up with classroom activities difficult. These challenges can affect confidence and participation, not just movement. Occupational therapy focuses on helping children take part in the activities that matter to them, while working with families and schools to make practice useful in everyday life. One 2025 randomized trial studied an AI-supported occupational therapy program for 42 children aged 8–12 who were at risk of DCD. Children in the intervention group attended therapy twice a week for eight weeks. Their handwriting improved in speed, alignment, size, spacing, shape, and legibility. These findings are encouraging, although the small study does not tell us whether the improvements lasted or transferred to everyday classroom writing (Demirci et al., 2025). A second study asks whether parents can lead home-based neuromotor practice with support from AI movement analysis. This is a trial protocol, not a report of results. It describes a planned study of 62 children with DCD, comparing conventional physiotherapy with conventional physiotherapy plus a 12-week, parent-led program. Video analysis is intended to estimate aspects of movement, such as coordination, symmetry, and speed. We must wait for the trial results before knowing whether the approach helps (Hullumani & Qureshi, 2026). Together, the studies point to two possible uses of AI: supporting a specific therapy program, such as handwriting practice, and helping families observe movement during home activities. In simple terms, AI may measure patterns in a video and help track change over time. But a measurement is not the same as understanding a child: camera position, lighting, clothing, and the child’s movement style can affect what the system detects. Results need careful checking. For occupational therapists, AI could help organize progress information, suggest activities for discussion, and show whether a child’s movement changes with practice. In the future, it may support work on balance, hand skills, planning, attention, or daily routines. These are possibilities to test not benefits we can assume. Therapists must check that AI-generated suggestions fit the child’s goals, abilities, culture, and home or school environment. Could parents, guided by AI, replace a therapist? AI may help parents practise activities between appointments, but it should not replace professional assessment or individualized care. Parents know their child well and can make practice part of daily life; therapists bring clinical training to interpret difficulties, adapt tasks, and respond when progress stalls or a new concern appears. A parent-led program works best as a partnership, with clear instructions and access to therapist guidance. The therapist’s role may therefore shift, but it does not disappear. Occupational therapists can set meaningful goals with children and families, guide safe practice, interpret AI-generated information, and adjust plans as needed. They also connect movement gains to everyday participation, such as playing, managing schoolwork, or dressing independently. Therapists must explain AI’s limits, protect children’s data, check for bias, and ensure that clinical decisions remain their responsibility. The next step: Future research should test whether these programs improve everyday participation, whether benefits last, and whether families can use them comfortably. AI may make practice and monitoring more accessible, especially where therapy services are limited. Its value will depend on keeping the human parts of care at the centre: listening to children, supporting families, and helping each child become more confident and independent. References Demirci, O., Guney Yilmaz, G., & Köse, B. (2025). Artificial intelligence-supported occupational therapy program on handwriting skills in children at risk for developmental coordination disorder: Randomized controlled trial. Research in Developmental Disabilities, 161, 105009. https://doi.org/10.1016/j.ridd.2025.105009 Hullumani, S., & Qureshi, I. (2026). Development, validation and effectiveness of a parents led neuromotor training module with AI-based movement analysis for improving motor proficiency and sports participation in children with developmental coordination disorder: A protocol for an assessor-blinded randomized superiority controlled trial. Frontiers in Human Neuroscience, 20, 1931622. https://doi.org/10.3389/fnhum.2026.1931622

Imagine preparing for a team meeting while also seeing clients, writing notes, and planning therapy. A tool that gathers approved information and prepares a draft summary could help with this workload. For therapists, the key question is how Dots might help without taking over clinical decisions. OpenAI introduced Dots in September 2026 as persistent agents powered by GPT-6 Astra, with cloud computers and access to connected tools. The company describes them as able to handle recurring work and follow up through ChatGPT, text message, email and Slack. The table below outlines four functions that may be relevant to therapists, along with practical examples of how they could be used: Function What it can do Practical example 🔍 1. Research & organize information Search connected sources, review documents, summarize findings, compare information, and identify relevant updates. Ask it to organize research papers about executive functions in children, summarize key findings, and prepare a list of sources for you to verify. 📋 2. Manage ongoing projects Follow a project over time, track progress, identify unfinished tasks, adapt plans when priorities change, and remind you about next steps. Ask it to manage your next month of occupational therapy Instagram content, track which posts are drafted, and suggest what needs to be completed next. ✍️ 3. Create content & documents Draft, revise, and update documents, presentations, reports, newsletters, and social media content. It can carry your feedback across related materials. Give it an interview transcript about AI in therapy and ask it to prepare a newsletter, several social media drafts, and a summary for therapists. ⚙️ 4. Automate workflows Use connected apps and its own computer to carry out multistep tasks, work with files, and follow through on approved actions. Ask it to review relevant emails, organize project documents, prepare a draft response, and ask for your approval before sending it. However, clinical information needs context. A change in a client’s performance may relate to pain, fatigue, the environment, or personal goals. A Dot might organize the information but miss why it matters. Therapists must check its work and decide what the information means. It is safest to begin with tasks that are simple, low-risk, and easy to check. For example, a team could ask a Dot to draft a meeting agenda or summarize de-identified attendance data. It should not independently assess a client, recommend treatment, or send clinical advice. Privacy also needs careful attention. Before using a Dot, therapists should check workplace rules, what information it can access, and how that information is handled. Patient details should only be used in systems approved by the organization, with clear permission and safeguards. OpenAI says Dots may retain context from conversations and connected apps while the Dot exists, so teams should understand data handling before use. Researchers may use Dots to flag missing information or prepare draft project updates. But a Dot should not change study data or replace the planned research methods. Researchers need to check its work and keep a clear record of how results were produced. AI can make mistakes, leave out important details, or reflect bias in the information it uses. Because its answers may sound convincing, users should not assume they are correct. Therapists and researchers remain responsible for checking the output, protecting information, and making decisions. Dots may be useful when they take on routine work while people stay in control. Teams can start with one small task and check whether the tool is accurate, saves time, and fits their workflow. The goal is not to automate care, but to reduce some of the work around care so therapists can focus on clients.

OpenAI released GPT-6 Sol and GPT-6 Luna about 90 minutes after Anthropic announced Claude Opus 5.5. OpenAI says the new models cost less than its previous models. For therapists, the key question is not just the price, it is whether these tools can help with everyday work while keeping care safe and personal. AI can help with writing and organizing tasks. For example, it can draft a simple explanation of anxiety, suggest questions for a group discussion, or prepare a first version of a between-session exercise. These are general descriptions, not guarantees. The best choice depends on the task and on whether the result is genuinely useful in your practice. To see how they might fit into our work, we explored what each tool is suited for: GPT-6 Sol: for more complex work Think of Sol as the stronger professional option for tasks that need careful reasoning or several steps. It may help with summarizing research, opening and analyzing very long documents, reviewing regulations and recommendations, and turning information into practical materials or plans. A therapist might use it to bring information together, plan a detailed workshop, or create a first draft of a training resource. OpenAI describes Sol as offering performance close to Astra at a lower cost, but its responses still need to be checked for accuracy and relevance to your clinical needs. GPT-6 Luna: for quick, everyday tasks Think of Luna as the fast and efficient option. It may be useful for focused tasks such as rewriting a paragraph in simpler language, translating text, drafting an email, summarizing notes, or generating ideas. It can also help with repetitive tasks and quick questions. For example, a therapist could use Luna to find everyday information quickly, draft emails, translate text, or create therapeutic materials, such as making a handout easier to read or suggesting several ways to explain a coping skill. The therapist can then choose and edit the version that works best. Sol and Luna at a glance  Sol  Luna Best for Complex or detailed tasks Everyday and repeated tasks Reasoning Stronger for multi-step work Good for focused requests Speed Fast Very fast Examples Reviewing long documents, planning, research summaries Rewriting, translation, emails, summaries, ideas Cost Higher than Luna Lower A practical way to begin is to choose one small, low-risk task. You could ask the tool to simplify a handout or suggest different ways to explain a coping skill. Check that the result is accurate, clear, respectful, and suitable for the people you work with. Protect client privacy: don’t enter names or identifying details unless your organization has approved the tool and its privacy protections. Use fictional or fully de-identified examples when testing AI. Then review the output for accuracy, respectful language, and cultural fit. AI may miss important details or make assumptions, so revise or discard anything that doesn’t fit the client or your professional judgment. Responsible AI use depends not just on the tool, but on our intention and how we use it. We need to choose the model that best fits the task, whether we’re summarizing research, simplifying a handout, or drafting an email. Learning to make that choice is a skill we can develop, helping us get more useful responses while keeping professional judgment at the center. GPT-6 Sol and Luna may offer therapists a quicker way to prepare useful materials and organize ideas. Their real value will depend on how carefully we use them. AI can support the work, but the therapist remains responsible for what is shared and how care is provided.

When a patient sends a message to a therapist or clinician, the reply needs to be accurate, clear, and caring. Artificial intelligence (AI) can help by preparing a draft response. But a draft is only a starting point: it may not understand the patient’s full situation, and a clinician must check it before sending. A 2025 study examined this approach in a large hospital in the Netherlands. Over 16 weeks, 100 physicians from 14 specialties responded to 919 patient messages. For each message, they could start with an AI-written draft or write a reply from a blank page. The study looked at how often clinicians chose drafts, how much they changed them, and how long replies took (Bootsma-Robroeks et al., 2025). Clinicians chose an AI draft for 58% of the messages. Among the replies that began with a draft, 43% kept at least some of its wording. This shows that clinicians were willing to try the tool, but they did not always use its suggestions in the same way. Some kept more of the draft; others changed most of it. The study found no clear time saving. Replies that used some of the AI draft took about the same time as replies written from a blank page: around two and a half minutes. Reviewing and correcting a draft may take time, so generating text does not necessarily make the whole task faster. For therapists, an AI draft may be useful for routine messages, such as confirming an appointment or explaining a general practice procedure. It may also help organize a response when a therapist is busy. This could leave more time for direct care—but the study did not show that these drafts saved time, so this benefit should not be assumed. Therapy messages often involve personal feelings, relationships, and sensitive experiences. An AI-generated reply may sound warm but still miss important meaning, context, or changes in a client’s emotional state. It could also suggest wording that feels impersonal or does not fit the therapist’s approach. A draft should never replace a therapist’s assessment, especially when a client may be at risk or needs urgent support. Therapists can use drafts as a starting point while keeping their own voice and clinical judgment. Before sending, they should check that the reply fits the client, responds to the actual concern, and does not promise more than the therapist can provide. If a message needs careful discussion, the therapist may decide that a session or phone call is more appropriate than a written reply. There are also important responsibilities. Patients’ private information must be protected, and health services should explain how these tools are used and check them for errors or unfair differences in performance. Clinicians remain responsible for the messages they send, even when AI has helped write them. This study provides an early look at AI-assisted patient communication in everyday care. The tool was used, but it did not yet show clear time savings. Future studies should examine whether it improves the clarity and quality of replies, how patients feel about receiving them, and whether it works safely for different people and types of messages. Reference Bootsma-Robroeks, C. M. H. H. T., Workum, J. D., Schuit, S. C. E., Hoekman, A., Mehri, T., Doornberg, J. N., van der Laan, T. P., & Schoonbeek, R. C. (2025). AI-generated draft replies to patient messages: Exploring effects of implementation. Frontiers in Digital Health, 7, Article 1588143. https://doi.org/10.3389/fdgth.2025.1588143

Imagine describing an activity you wish existed and getting a simple interactive prototype to explore. Claude Artifacts are stand-alone creations, such as interactive apps, visualizations, or tools, that Claude generates from a conversation. You can preview them, ask Claude to revise them, and, where available, share them with others. For therapists, this could mean exploring new ways to support learning, communication, movement, and everyday skills, not just drafting another worksheet. A therapist might prototype a visual activity where someone sorts objects, follows a sequence, or makes choices in a familiar situation. The activity could use pictures, sounds, movement prompts, or simple animations, then be adapted for pace, instructions, or level of challenge. An interactive routine might help someone practise getting ready, preparing for a transition, or completing a task in order. Another prototype could invite them to follow a movement pattern, choose a picture to continue a story, match an action to a sound, or practise asking for help. These are ideas to explore, not ready-made treatments or substitutes for professional assessment. The wider opportunity is that therapists can experiment with ideas that might otherwise be difficult or costly to build. A team could create a basic visual schedule, a turn-taking game, or an animation to demonstrate a movement or concept. The therapist brings the purpose and clinical expertise; AI can help explore how the idea might take shape. How to create an Artifact A working prototype is not automatically a suitable therapy tool. AI-generated activities should not be treated as validated assessments or interventions. The therapist remains responsible for deciding whether a tool fits the person, their goals, and the clinical context. Privacy matters throughout the process: do not include names, session notes, identifying details, or sensitive client information in prompts or prototypes. Follow workplace policies and professional guidance, and check sharing settings before publishing. Claude can help build and revise an idea, but it cannot observe a client, make clinical judgments, or take responsibility for care. Artifacts can turn a therapist’s idea into something tangible to explore and adapt. Start with one activity you wish existed, describe it without client information, and see what you can prototype, then use your professional judgment to decide whether it is useful, and for whom.

Artificial intelligence (AI) is becoming part of everyday school life. More than half of U.S. teenagers say they use AI chatbots to help with schoolwork. For therapists, educators, and families, the important question is not only whether AI helps students get better grades, but also how it may affect the way they learn, think, and solve problems. A 2025 study of almost 1,000 high school students found that students using ChatGPT improved their mathematics performance by 48%. However, when ChatGPT was removed, these students performed 17% worse than students who had never used it (Bastani et al., 2025). This suggests that students may complete tasks successfully with AI without fully learning how to do the work independently. One reason may be cognitive offloading. This means using an outside tool to do part of the thinking for us. People have always used tools such as calculators, calendars, and search engines. These tools can be helpful, but problems may occur when AI does the main thinking instead of supporting the student’s own reasoning. Young people learn through practice, mistakes, and effort. When students struggle with a difficult problem, they develop skills such as attention, memory, planning, and persistence. If AI immediately provides the answer, students may avoid this important learning process. They may finish the assignment more quickly but remember less afterward. This does not mean that all struggle is helpful. Students need support when work is too difficult or when they feel overwhelmed. However, some level of effort is necessary for learning. Therapists and educators should therefore help students use AI as a guide, not as a replacement for thinking. The same study showed that the way AI is designed makes a difference. Students benefited more when the system gave hints and explanations instead of simply providing the final answer. This type of support can encourage students to think, try again, and understand their mistakes. AI should gradually become less necessary as the student becomes more confident. These issues are especially important for young people with anxiety, attention difficulties, learning disabilities, or low confidence. AI may reduce stress in the short term, but constant dependence on it may make students believe they cannot succeed without technological help. Therapists should therefore ask not only, “Did the student complete the task?” but also, “What did the student understand and what can they do independently?” AI can still be useful in education and therapy. It may explain difficult ideas in simpler language, provide examples, support organisation, or offer feedback. However, professionals must consider privacy, accuracy, bias, and the possibility that AI may give confident but incorrect information. Human judgment and responsibility must remain central. The goal should not be to stop young people from using AI. Instead, they need to learn how to use it wisely. Future research should examine the long-term effects of AI on learning, confidence, and independent thinking. Used carefully, AI can support development; used without limits, it may reduce the effort and reflection that real learning requires. References Bastani, H., Bastani, O., Sungu, A., Ge, H., Kabakcı, Ö., & Mariman, R. (2025). Generative AI without guardrails can harm learning: Evidence from high school mathematics. Proceedings of the National Academy of Sciences, 122(26), e2422633122. https://doi.org/10.1073/pnas.2422633122

This week, several major AI leaders raised the same concern. Dario Amodei of Anthropic warned that AI is developing faster than expected. Sam Altman of OpenAI and Elon Musk of xAI supported his message. Microsoft CEO Satya Nadella also said that superintelligence, AI that could become better than humans at almost everything, should not be developed without human control. King Charles is also bringing leaders from Nvidia, Google DeepMind, OpenAI, and Anthropic together to discuss AI safety. As therapists, we find this development concerning because the warning is coming from the people building these systems. Superintelligence would not be limited to playing chess or writing computer code. It could potentially make decisions, create information, and influence human behaviour at a level beyond human understanding. A system this powerful could bring important benefits, but it could also cause serious harm if it is poorly controlled. Popular films such as 2001: A Space Odyssey and The Terminator often present AI as dangerous. These stories are exaggerated, but they raise a real question: what happens when technology becomes more powerful than the people using it? The main concern is not necessarily that AI will become a “killer robot.” The deeper concern is that people may lose control over systems they do not fully understand. For therapists, this question connects directly to human behaviour and mental health. AI already influences how people learn, communicate, work, and make decisions. Some individuals may trust AI more than professionals, while others may become emotionally dependent on AI systems. These changes could affect relationships, self-confidence, attention, decision-making, and social development. AI can be helpful in therapy and rehabilitation. It may support documentation, provide educational material, assist with communication, or help professionals organise information. However, it can also give incorrect answers, misunderstand emotions, or reflect bias in its training data. A response that sounds caring and professional may still be clinically unsafe. A therapist understands the patient’s history, emotions, culture, relationships, and nonverbal communication. AI does not have human experience or professional responsibility. It cannot fully understand the meaning behind a patient’s silence, fear, resistance, or personal story. The phrase “slow down” does not necessarily mean stopping all AI development. It may mean testing systems more carefully, studying their risks, protecting private data, and creating clear limits before allowing AI to make important decisions. In clinical practice, using AI is similar to introducing a new treatment: professionals need evidence, safety monitoring, and careful judgment before using it widely. During this period of rapid development, we, as healthcare professionals, also have an important responsibility. We need to guide how AI is integrated into our practice, understand its limitations, and consider its ethical consequences. This does not mean refusing to use AI; it means using it wisely, carefully, and with enough awareness to protect patients and maintain professional standards. The important question is whether these public warnings will lead to real changes. Will AI companies change their plans, delay risky systems, and create stronger safety rules? Or will they speak responsibly in public while continuing to compete privately? Words are valuable, but genuine responsibility must be shown through transparent decisions and measurable limits. As therapists, researchers, and members of society, we are all facing the same question: How far should AI go? AI may improve healthcare and human life, but progress should not be measured only by what technology can do. It should also be measured by whether it protects human dignity, safety, relationships, and professional accountability.

Healthcare professionals often work with a huge amount of information: clinical notes, assessment results, medical reports, and patient histories. Reading, organizing, and reviewing all of this can take valuable time. This is where MedGemma may become useful. MedGemma is a medical AI model developed by Google. It is based on the Gemma 3 model and adapted to work with healthcare-related information, including medical text and images. In simple terms, it is designed to help professionals understand and manage medical information more efficiently. MedGemma is not a doctor, therapist, or independent clinical decision-maker. It does not replace professional experience or human understanding. Instead, it can support everyday tasks such as summarizing records, explaining medical terms, organizing information, and preparing educational materials. One interesting possibility is its role in clinical reasoning. In a complex case, MedGemma may help organize the available information, identify missing details, suggest questions, or highlight areas that need further investigation. This can help professionals look at a case from different angles, but its suggestions must always be reviewed carefully. Compared with general AI tools, MedGemma has been specifically adapted for medical applications. This may make it more suitable for healthcare-related tasks. However, being a medical model does not mean that it is always correct. Its performance depends on the quality of the information provided and the way the tool is developed and used. For therapists, a MedGemma-based tool could summarize a long patient history before a session or help organize information from different reports. It might also support psychoeducation, case discussions, supervision, or research. By reducing some administrative work, it could give therapists more time to focus on listening and connecting with their patients. In some applications, it may also help interpret medical images. Still, the model can misunderstand information, miss important details, or provide an answer that sounds convincing but is incorrect. Because MedGemma is an open-weight model, it can be downloaded and adapted for different projects. Depending on the version and available equipment, it may be used locally on a computer or organizational server, or through cloud-based systems. The model itself may be freely available, but users may still need to pay for hardware, cloud services, software development, security, and maintenance. Using the model locally may help organizations keep sensitive information within their own systems. However, local use does not automatically guarantee privacy or safety. Healthcare teams still need to consider patient consent, data protection, cybersecurity, access controls, and secure record keeping. MedGemma also has clear limitations. It may struggle with rare conditions, incomplete information, poor-quality images, or cases that are very different from the data used during training. For this reason, professionals should always compare its outputs with the patient’s history, current presentation, clinical guidelines, and their own expertise. Overall, MedGemma should be seen as a developing support tool. Used thoughtfully, it may help therapists and clinicians manage information, prepare for sessions, and support research and education. Its greatest value will come from combining its technical abilities with human judgment, empathy, and responsibility.

In everyday clinical work, patient care is recorded in many places, including progress notes, imaging reports, medication records, and discharge summaries. Because this information is often incomplete or written in different ways, it can be difficult to know whether care followed recommended clinical guidelines. Medical process conformance checking is a method that compares what happened to what should normally happen according to professional guidance. A recent study by Leonardi and colleagues, titled “Orchestrating Large Language Models to Support Medical Process Conformance Checking,” examined how large language models, such as ChatGPT, could support this process. Large language models are computer systems that can understand and produce human language. In this study, they were used to read clinical documents and guidelines, identify important information, and compare patient care with recommended practice (Leonardi et al., 2026). The researchers used several language models rather than asking one system to complete the entire task. Each model had a specific role. One helped identify clinical events, another examined guideline recommendations, and another converted these recommendations into computer-readable rules. This step-by-step approach can make the process easier to check and correct. The study focused on stroke care. The system examined discharge letters and created a simple timeline of each patient’s care. It then compared these timelines with selected recommendations for stroke treatment. The results suggested that most patient records followed the expected care pathway. This shows that language models may help hospitals review large numbers of clinical records more efficiently. However, the results must be interpreted carefully. A medical record may not include everything that happened during treatment. A missing action in the documentation does not always mean that the action was not performed. In addition, patients may require different treatments because of their age, medical history, risks, or personal circumstances. For therapists and other healthcare professionals, this technology should be viewed as a support tool, not as a replacement for clinical judgment. It may help identify repeated delays, missing documentation, or differences between departments. These findings can then guide team discussions, quality improvement, and better communication between professionals. One important advantage of this approach is that it can make the reasoning process more visible. Users may be able to see the original clinical text, the guideline recommendation, and the rule used to evaluate care. This is safer than relying on a single unexplained answer from an artificial intelligence system. There are also important ethical concerns. Patient information must be protected, and healthcare organizations must carefully control how records are used. Language models can misunderstand information or produce biased results, especially when documentation is incomplete. Clinicians and institutions must remain responsible for decisions influenced by these systems. In conclusion, large language models may help connect clinical documentation with medical guidelines and support safer, more organized healthcare. Their role should remain supportive and transparent. Future research should examine how accurate these systems are in different hospitals and specialties, and whether they can genuinely improve clinical practice without reducing care to simple numbers. References Leonardi, G., Montani, S., Striani, M., Canessa, A., & Ferrandi, D. (2026). Orchestrating large language models to support medical process conformance checking. Neuroscience Informatics, 6(4), Article 100294. https://doi.org/10.1016/j.neuri.2026.100294