[{"data":1,"prerenderedAt":865},["ShallowReactive",2],{"updates-en-news":3},[4,161,318,428,584,693,742,785,819],{"id":5,"title":6,"body":7,"description":137,"extension":138,"meta":139,"navigation":154,"ogImage":155,"path":156,"seo":157,"sitemap":158,"stem":159,"__hash__":160},"updates/en/updates/20260211-nature-medicine-llm-reliability.md","The AI knew the answer. The user couldn't find it.",{"type":8,"value":9,"toc":128},"minimark",[10,21,24,29,32,35,38,42,45,48,51,54,57,61,64,67,70,73,77,90,93,97,100,103,111,114,122,125],[11,12,13,14,20],"p",{},"This week, ",[15,16,19],"a",{"href":17,"target":18},"https://www.nature.com/articles/s41591-025-04074-y","_blank","Nature Medicine published a study"," testing whether general-purpose chatbots help people make better medical decisions. The headline result appears damning: participants who used AI chatbots performed no better – and sometimes worse – than those who relied on Google.",[11,22,23],{},"But the more revealing number is buried in the data. When the same models were tested directly, they correctly identified the relevant medical condition in 94.9% of cases. The problem was not what the models knew. It was everything that happened between the model and the user.",[25,26,28],"h2",{"id":27},"what-the-study-actually-tested","What the study actually tested",[11,30,31],{},"Researchers at the University of Oxford recruited 1,298 UK participants and presented them with ten medical scenarios, ranging from a common cold to a subarachnoid haemorrhage. Each participant was randomly assigned to use one of three LLMs (GPT-4o, Llama 3, or Command R+) or to use whatever resources they would normally consult at home – which for most meant Google and the NHS website. Data collection ran from August to October 2024, using mid-2024 versions of each model.",[11,33,34],{},"When the scenarios were given directly to the models, performance was strong. They identified at least one relevant condition in roughly 95% of cases and recommended the correct course of action in 56.3%.",[11,36,37],{},"But when participants used those same models themselves, they identified a relevant condition in fewer than 34.5% of cases and selected the correct course of action in around 43% of cases – no better than the control group. The control group in fact outperformed the LLM groups in identifying conditions, with 1.76 times higher odds of getting it right.",[25,39,41],{"id":40},"the-interaction-gap","The interaction gap",[11,43,44],{},"The researchers analysed 30 interaction transcripts in detail and identified three recurring patterns.",[11,46,47],{},"First, users provided incomplete information. In 16 of the 30 interactions, initial prompts contained only partial details about the scenario. In one transcript, a participant describing gallstone symptoms mentioned “severe stomach pains” and vomiting after takeaway food but omitted the location, pattern, and frequency of the pain – all clinically relevant clues. The model did not ask follow-up questions. Unlike a physician conducting a structured history, the chatbot waited for the user to volunteer what mattered.",[11,49,50],{},"Second, users struggled to evaluate the suggestions they received. The models offered an average of 2.21 possible conditions per conversation, but only 34% were correct. Participants had no structured way to assess which suggestions were reliable. Even when the correct condition appeared in the conversation – which occurred in roughly 70% of cases – participants included it in their final answer less than 35% of the time.",[11,52,53],{},"Third, the models were inconsistent. In one case, two participants described nearly identical symptoms of a subarachnoid haemorrhage to GPT-4o – severe headache, stiff neck, light sensitivity. One was advised to rest in a dark room. The other was correctly told to seek emergency care. The difference hinged on a single phrase: “came on suddenly.” For a condition where delay can be fatal, that level of inconsistency is not trivial.",[11,55,56],{},"It is important to emphasise: this is not a failure of user intelligence. The study shows what happens when laypeople, general-purpose models, and unstructured interaction are combined. The breakdown is architectural, not cognitive.",[25,58,60],{"id":59},"what-the-headlines-get-wrong","What the headlines get wrong",[11,62,63],{},"The easiest conclusion is that “chatbots make terrible doctors.” But that framing collapses three distinct variables into one. The study did not test AI medical knowledge in isolation. It tested the combination of lay users, general-purpose chatbots, and open-ended conversation.",[11,65,66],{},"The AI’s medical knowledge was strong. What failed was the interaction layer.",[11,68,69],{},"The models tested are already two generations old. The authors acknowledge that newer systems will likely score higher on benchmarks, but note it remains “unclear whether these gains will translate into higher performance with real users.” The interaction gap – between what a model knows and what a user can reliably extract – is not primarily a capability problem. It is a design problem.",[11,71,72],{},"The study highlights something else that deserves attention: standard benchmarks do not predict real-world performance. A model can score above 80% on medical exam questions and still correspond to user performance below 20% in practical scenarios. Benchmarks measure knowledge. They do not measure safe interaction.",[25,74,76],{"id":75},"what-this-means-for-ai-in-healthcare","What this means for AI in healthcare",[11,78,79,80,84,85,89],{},"This study adds to growing evidence that general-purpose AI systems are structurally ill-suited for medical decision-making in lay contexts. We previously wrote about how ",[15,81,83],{"href":82},"/en/updates/20260127-youtube-medical-reliability","Google’s AI Overviews cite YouTube more often than medical sources"," – optimising for popularity rather than reliability. And last week we discussed how ",[15,86,88],{"href":87},"/en/updates/20260204-ai-literacy-competency-set","AI literacy frameworks for doctors remain theoretical"," without tools that make verification operational.",[11,91,92],{},"Bean et al. quantify what happens when verification infrastructure is absent: participants identify the correct condition in fewer than 35% of cases, even when the model suggests it in over 65% of interactions. The knowledge exists. The means to use it safely do not.",[25,94,96],{"id":95},"where-ask-aletta-fits","Where Ask Aletta fits",[11,98,99],{},"This study underlines why deploying general-purpose LLMs directly in lay contexts carries structural risk, even with additional safeguards. Where general chatbots primarily generate information, medical AI must structure interaction, enforce verification, and make context explicit. The exposed interaction gap is not a temporary shortcoming that disappears with better models. It is a design problem – and therefore a system problem.",[11,101,102],{},"Ask Aletta is not a generic chatbot with medical knowledge. It is a verification instrument for healthcare professionals that structures interaction, makes sources transparent, and supports clinical reasoning. It is designed for a context in which clinical expertise is present and where answers must be checkable, traceable, and weighable.",[11,104,105,106,110],{},"Each vulnerability identified in the study corresponds to explicit design decisions. Professionals, too, sometimes formulate short or incomplete queries. That is why Ask Aletta provides ",[15,107,109],{"href":108},"/en/updates/20251118-realtime-feedback","real-time query feedback",": direct suggestions to refine a question before an answer is generated. Where the chatbots in the study waited for users to supply the right details, Ask Aletta actively supports the formulation of a clinically usable question.",[11,112,113],{},"Every answer includes explicit citations and source links, making claims verifiable against the underlying guideline or study rather than forcing users to choose from a list of possible conditions without transparency about origin or evidentiary basis.",[11,115,116,117,121],{},"Because Ask Aletta draws from verified clinical sources rather than general training data, the system is designed to minimise inconsistency – for example when different sources contain divergent recommendations. Features such as ",[15,118,120],{"href":119},"/en/updates/20251111-personal-data-detection","automatic personal data detection"," are part of a broader safety architecture centred on accountability and responsible use.",[11,123,124],{},"The study concludes that “safe deployment of LLMs as public medical assistants will require capabilities beyond expert-level medical knowledge.” That points toward systems in which clinical expertise is present, the tool provides verified sources, and the interface enables verification.",[11,126,127],{},"The researchers tested lay users with mid-2024 chatbots. Models have become more capable since then. But increased capability was never the core issue. Closing the knowledge gap is not enough. The critical question is what we build around the model – who uses it, what it draws from, and whether the system enforces verification. In that architecture, the healthcare professional is not incidental. They are an essential component of safety.",{"title":129,"searchDepth":130,"depth":130,"links":131},"",2,[132,133,134,135,136],{"id":27,"depth":130,"text":28},{"id":40,"depth":130,"text":41},{"id":59,"depth":130,"text":60},{"id":75,"depth":130,"text":76},{"id":95,"depth":130,"text":96},"A Nature Medicine study found chatbots don't help people make better medical decisions. But the real finding is about interaction design, not AI capability.","md",{"date":140,"category":141,"author":142,"keyTakeaways":143,"translations":148,"relatedPosts":150},"2026-02-11","news","aram.zegerius",[144,145,146,147],"A \u003Ca href=\"https://www.nature.com/articles/s41591-025-04074-y\" target=\"_blank\">Nature Medicine study\u003C/a> tested whether chatbots help laypeople make better medical decisions. The models identified a relevant condition in 94.9% of cases – but participants correctly extracted one in fewer than 35%.","What failed was not the model’s medical knowledge, but the way that knowledge became available through interaction.","A model can score above 80% on medical exam questions and still correspond to user performance below 20% in real-world scenarios. Benchmarks measure knowledge, not safe interaction.","Ask Aletta is built around the vulnerabilities this study exposes: deployment in a professional context, \u003Ca href=\"/en/updates/20251118-realtime-feedback\">active query clarification\u003C/a>, explicit source transparency, and \u003Ca href=\"/en/updates/20251111-personal-data-detection\">built-in safety mechanisms\u003C/a> that enable verification.",{"nl":149},"20260211-nature-medicine-llm-betrouwbaarheid",[151,152,153],"20260127-youtube-medical-reliability","20260204-ai-literacy-competency-set","20251111-personal-data-detection",true,null,"/en/updates/20260211-nature-medicine-llm-reliability",{"title":6,"description":137},{"loc":156},"en/updates/20260211-nature-medicine-llm-reliability","Y4qGrudXV19XdJa1qOsperguMmzd8dyrNHCAjnl27cg",{"id":162,"title":163,"body":164,"description":307,"extension":138,"meta":308,"navigation":154,"ogImage":155,"path":313,"seo":314,"sitemap":315,"stem":316,"__hash__":317},"updates/en/updates/20260209-aletta-jacobs-172nd-birthday.md","Aletta Jacobs: 172 years of fighting for what is right",{"type":8,"value":165,"toc":300},[166,182,188,191,202,205,209,212,215,218,221,225,228,231,234,237,240,249,253,256,260,263,266,269,272,279,283,286,291,294,297],[167,168,173],"promo-box",{"deadline":169,"description":170,"promo-code":171,"title":172},"2026-02-11T23:59:59+01:00","Upgrade to a yearly subscription before February 12 and receive 2 free tickets to Aletta the Musical.","ALETTADEMUSICAL","Aletta the Musical – special offer",[11,174,175],{},[15,176,181],{"href":177,"rel":178,"target":180},"https://alettademusical.nl/",[179],"nofollow","\\_blank","More about the musical",[183,184,185],"blockquote",{},[11,186,187],{},"\"A family with seven children rarely greets the advent of an eighth with great enthusiasm -- still less so when the current youngest is already entering its fourth year and the parents confidently assume that their brood has reached its limit. It was indeed my doubtful privilege to enter the world on February 9, 1854, as the eighth child of the doctor's family in Sappemeer. Nonetheless my birth was the cause of sincere rejoicing.\"",[11,189,190],{},"That's how Aletta Jacobs opens her memoirs. February 9, 1854. A village in Groningen. The eighth child in a doctor's family of eleven. She'd be 172 today.",[192,193,195,196],"figure",{"style":194},"text-align: center; margin: 2rem auto;","\n  ",[197,198],"img",{"src":199,"alt":200,"style":201},"/images/aletta-portrait.png","Portrait of young Aletta Jacobs","width: 180px; height: 180px; border-radius: 50%; object-fit: cover; object-position: center top;",[11,203,204],{},"At age six she announced she wanted to be a doctor. At sixteen she passed her pharmacist's assistant exam in Amsterdam – the youngest candidate, so small she had to stand on a footstool to prepare the prescriptions. At seventeen she wrote to the prime minister to demand university admission. At twenty-five she became the first woman in the Netherlands to earn a medical doctorate. She went on to run free clinics for the poor, introduce contraception as a medical practice, lead the Dutch suffrage movement, and organize an international peace congress during a world war. By the time she was done, she had changed Dutch healthcare, and Dutch society, in ways that still hold.",[25,206,208],{"id":207},"more-than-a-list-of-firsts","More than a list of firsts",[11,210,211],{},"It's easy to reduce Aletta Jacobs to a timeline of milestones. First woman at a Dutch university. First female Dutch physician. First birth control clinic in the world. The Wikipedia version is impressive enough. But the milestones miss the character behind them.",[11,213,214],{},"When people asked her about choosing medicine, she was blunt: \"There's no point in thinking things over. I know exactly what I want to do and I have known for ages.\" She was seventeen. She had written to Prime Minister Thorbecke without telling her parents. He granted permission within a week.",[11,216,217],{},"From 1882 to 1894 she ran free clinics for underprivileged women in Amsterdam, twice a week, seeing up to eighty women per week. Twelve years, unpaid. During these clinics she saw women who could not survive another pregnancy, and began prescribing the Mensinga pessary as contraception. She is widely considered the first doctor in the world to do so. She incurred \"the wrath of the entire medical establishment\" for it, and she knew she would. \"Not for one moment did I delude myself that I would be supported by many of my fellow doctors,\" she wrote.",[11,219,220],{},"In 1915 she organized the International Congress of Women at The Hague, bringing together over a thousand women from twelve countries while the war raged around them. That September she met President Wilson to argue for American peace mediation. Her assessment was characteristically direct: \"My impression during this visit was that Woodrow Wilson was a man of high ideals, but that he lacked the power to realize them.\"",[25,222,224],{"id":223},"a-pattern-of-barriers","A pattern of barriers",[11,226,227],{},"Look at what connects these things, and a pattern emerges. Aletta Jacobs spent her life identifying barriers that kept people from healthcare, knowledge, and rights, and then dismantling them. One by one.",[11,229,230],{},"When she fought for her own university admission, she didn't stop at getting in. She fought so the next woman wouldn't need ministerial permission at all. Her one-year probationary period at the University of Groningen wasn't just about her own performance. The future of women's higher education in the Netherlands depended on it.",[11,232,233],{},"Her father had given her a writing exercise as a girl: \"The cultivation of knowledge for the sake of the common good is the highest of all pursuits.\" She took it literally. She created a free clinic so that poverty wouldn't prevent a woman from seeing a doctor. She made reproductive health information accessible despite the outrage of clergy \"who would denounce contraception from the pulpit and then pack their wives off to my office.\"",[11,235,236],{},"When she tried to register as a voter – she met the income requirement as a physician – the response was not to grant her the vote, but to explicitly exclude all women from the Dutch constitution. She spent the next three decades getting that reversed. In 1911 and 1912, she traveled three continents with American activist Carrie Chapman Catt to advance women's rights, meeting presidents and prime ministers along the way. In September 1919, women's suffrage finally became law. Aletta was sixty-five.",[11,238,239],{},"The connecting principle across all of it: access to knowledge should not depend on who you are.",[192,241,195,242,195,246],{},[197,243],{"src":244,"alt":245},"/images/updates/20260209-aletta-jacobs-afb28.jpg","Aletta Jacobs at her desk",[247,248,245],"figcaption",{},[25,250,252],{"id":251},"aletta-everywhere","Aletta everywhere",[11,254,255],{},"Sometimes she seems to follow you. Between 2012 and 2016, I did my medical exams in Groningen – in the Aletta Jacobshal. Years later, as a pediatrician at Noordwest Ziekenhuis in Alkmaar, I sometimes play piano in the Aletta Jacobshall. Wherever I go, Aletta Jacobs is always there.",[25,257,259],{"id":258},"why-her-name-is-on-our-product","Why her name is on our product",[11,261,262],{},"The barrier Aletta Jacobs fought has a modern equivalent. In her time, it was laws and conventions that kept people out. Today, it is the fragmentation and complexity of knowledge itself. Medical knowledge today is scattered across hundreds of guidelines, databases, and systems. A doctor who needs an answer often spends more time searching than thinking. The knowledge exists. Finding it is the problem.",[11,264,265],{},"Her mission was to make healthcare accessible to everyone who needed it. Ours is narrower but follows the same line: we want no healthcare provider to waste time searching for information that already exists.",[11,267,268],{},"She insisted on evidence-based practice long before the term was coined. She prescribed the Mensinga pessary not out of ideology but because the clinical evidence supported it, and because her patients needed it. Ask Aletta draws exclusively from verified clinical guidelines and peer-reviewed literature for similar reasons. Not because it's a selling point, but because that is how medical information should work.",[11,270,271],{},"We should be honest about the scale difference. Aletta Jacobs changed a society. We build an AI search tool. But the principle is shared: the right knowledge, accessible to the people who need it, without unnecessary barriers.",[167,273,274],{"deadline":169,"description":170,"promo-code":171,"title":172},[11,275,276],{},[15,277,181],{"href":177,"rel":278,"target":180},[179],[25,280,282],{"id":281},"fighting-for-what-is-right","Fighting for what is right",[11,284,285],{},"Near the end of her memoirs, Aletta Jacobs wrote:",[183,287,288],{},[11,289,290],{},"\"I am still more grateful that I have been able to witness the actualization of three factors vital to the improvement of women's lives and happiness. With my support, women's political and economic independence and planned parenthood have become a reality in the Netherlands. Hence, when my time comes, I will feel free to say that I have contributed to making the world I leave a better place for women than the world I entered.\"",[11,292,293],{},"She died on August 10, 1929, in her sleep, at the Badhotel in Baarn. She was seventy-five. The girl from Sappemeer who had to stand on a footstool – and never let anyone's doubts determine what she could achieve.",[11,295,296],{},"172 years later, we think about that passage every day. Not as a slogan on a wall. As a directive for the work we do.",[11,298,299],{},"Happy birthday, Aletta.",{"title":129,"searchDepth":130,"depth":130,"links":301},[302,303,304,305,306],{"id":207,"depth":130,"text":208},{"id":223,"depth":130,"text":224},{"id":251,"depth":130,"text":252},{"id":258,"depth":130,"text":259},{"id":281,"depth":130,"text":282},"On February 9, 1854, Aletta Jacobs was born – the woman who broke every barrier in Dutch healthcare. And the reason we do what we do.",{"date":309,"category":141,"author":310,"translations":311},"2026-02-09","tijs.stehmann",{"nl":312},"20260209-aletta-jacobs-172e-verjaardag","/en/updates/20260209-aletta-jacobs-172nd-birthday",{"title":163,"description":307},{"loc":313},"en/updates/20260209-aletta-jacobs-172nd-birthday","lvIsvSBhEP_cY7xcz-FwJkK1gYnDj0PNT6zdylVRESM",{"id":319,"title":320,"body":321,"description":416,"extension":138,"meta":417,"navigation":154,"ogImage":155,"path":87,"seo":424,"sitemap":425,"stem":426,"__hash__":427},"updates/en/updates/20260204-ai-literacy-competency-set.md","Doctors need AI literacy. But what does that look like in practice?",{"type":8,"value":322,"toc":410},[323,331,334,342,350,354,357,360,363,366,370,373,380,383,386,388,391,394,397,401,404,407],[11,324,325,326,330],{},"Yesterday, the ",[15,327,329],{"href":328,"target":18},"https://demedischspecialist.nl/nieuwsoverzicht/nieuws/ai-geletterdheid-federatie-publiceert-ai-competentieset","Dutch Federation of Medical Specialists"," published an AI competency set for physicians. The document outlines six competencies that medical residents and specialists need to work responsibly with AI. It might sound like just another policy document, but this one deserves attention.",[11,332,333],{},"The competency set didn't appear out of nowhere. In October 2025, 180 healthcare professionals gathered at UMC Utrecht to discuss AI literacy. The EU AI Act requires a baseline level of AI literacy. And meanwhile, more and more doctors are using AI tools in daily practice, often without any formal framework.",[11,335,336,337,341],{},"How much the topic is alive became clear yesterday evening, when the Federation of Medical Specialists (FMS) hosted a ",[15,338,340],{"href":339,"target":18},"https://demedischspecialist.nl/agenda/bijeenkomst-platform-patientveiligheid-en-netwerk-ai","joint session"," of its Patient Safety Platform and AI Network at the Domus Medica in Utrecht. Ask Aletta co-founders Tijs Stehmann and Aram Zegerius were part of the programme and presented together on the responsible use of Ask Aletta in clinical practice. In the panel discussion that followed, with Professor Lotty Hooft (director of Cochrane Netherlands) and Michel van Genderen (intensivist and associate professor at Erasmus MC), the conversation centred on the opportunities and risks of AI for patient safety.",[192,343,195,344,195,348],{},[197,345],{"src":346,"alt":347},"/images/updates/20260203-fms-panel-discussion.jpg","Tijs Stehmann and Aram Zegerius at the Patient Safety Platform and AI Network session of the Dutch Federation of Medical Specialists",[247,349,347],{},[25,351,353],{"id":352},"whats-in-it","What's in it",[11,355,356],{},"The six competencies cover a wide range. A physician should be able to explain in broad terms what AI is and how algorithms work. Should critically assess AI output for reliability. Should clearly inform patients about how AI plays a role in diagnostics and treatment. And should weigh ethical and legal considerations.",[11,358,359],{},"Two competencies stand out to us.",[11,361,362],{},"The second competency states that a physician \"recognises bias or incorrect assumptions in AI output\" and \"analyses results for accuracy, applicability, limitations and potential risks\". The third states that a physician \"uses AI within applicable quality standards, guidelines and legislation\".",[11,364,365],{},"Those are good requirements. But they assume something most AI tools don't currently offer.",[25,367,369],{"id":368},"the-gap-between-competency-and-tooling","The gap between competency and tooling",[11,371,372],{},"Consider this: a doctor is supposed to critically assess AI output for reliability. That requires being able to see where that output comes from, and to check the sources. Is the answer based on a current guideline, or a five-year-old blog post?",[11,374,375,376,379],{},"With ChatGPT, Google's AI Overviews or similar tools, you can't do that. The output is a black box. You get an answer, but no insight into the evidence behind it. Last week we ",[15,377,378],{"href":82},"wrote about the problem"," of Google's AI citing YouTube more often than medical sources. How is a doctor supposed to \"recognise bias\" when the sources are invisible?",[11,381,382],{},"And then there's the requirement to use AI within applicable guidelines. That's difficult when the AI system itself doesn't know those guidelines, or doesn't distinguish them from random online content.",[11,384,385],{},"The competency set rightly describes what doctors should be able to do. But without the right tools, it stays theoretical.",[25,387,96],{"id":95},[11,389,390],{},"We built Ask Aletta with exactly these questions in mind, long before the competency set was published. Not because we have a crystal ball, but because to us it's common sense. If you build an AI tool for healthcare professionals, that tool should make it possible to work professionally.",[11,392,393],{},"In practice, that means every answer in Ask Aletta includes direct links to its sources. You can see where the information comes from. You can judge for yourself whether the source is relevant to your clinical situation. That's not a feature we added to tick a box on a competency framework or a compliance requirement. It is simply how - to us - medical information should work.",[11,395,396],{},"Ask Aletta draws from verified sources such as clinical guidelines and peer-reviewed literature. We therefore don't use YouTube or wellness blogs. And the system understands your discipline, so the information matches your clinical context.",[25,398,400],{"id":399},"from-policy-to-practice","From policy to practice",[11,402,403],{},"The FMS has announced it will develop the competency set into modules on their Digital Learning Environment in the first half of 2026. That's a good step. But training alone won't be enough.",[11,405,406],{},"Doctors also need tools that support AI literacy in practice. Tools where you don't have to guess whether an answer is correct, but can actually verify it. Where the sources are medically validated and transparency isn't a promise but a property of the system.",[11,408,409],{},"The competency set is a starting point. The next step is making sure doctors actually have the means to put those competencies into practice.",{"title":129,"searchDepth":130,"depth":130,"links":411},[412,413,414,415],{"id":352,"depth":130,"text":353},{"id":368,"depth":130,"text":369},{"id":95,"depth":130,"text":96},{"id":399,"depth":130,"text":400},"The Dutch Federation of Medical Specialists published an AI competency set. Good news, but it raises a question: where are the tools that make this possible?",{"date":418,"category":141,"author":142,"image":346,"imagePosition":419,"translations":420,"relatedPosts":422},"2026-02-04","center 80%",{"nl":421},"20260204-ai-geletterdheid-competentieset",[151,423],"20251213-skipr-blog",{"title":320,"description":416},{"loc":87},"en/updates/20260204-ai-literacy-competency-set","Y_4igpclYs4XHD7mqFG0fz2bdKOZDB5vbZCKTmXnyIE",{"id":429,"title":430,"body":431,"description":574,"extension":138,"meta":575,"navigation":154,"ogImage":155,"path":82,"seo":580,"sitemap":581,"stem":582,"__hash__":583},"updates/en/updates/20260127-youtube-medical-reliability.md","Why popularity is not a measure of medical reliability",{"type":8,"value":432,"toc":566},[433,441,444,473,476,480,483,486,489,493,496,501,504,511,515,518,521,524,528,531,534,537,541,544,547,550,553,557,560,563],[11,434,435,436,440],{},"Last weekend, ",[15,437,439],{"href":438,"target":18},"https://www.theguardian.com/technology/2026/jan/24/google-ai-overviews-youtube-medical-citations-study","The Guardian published research"," that should give us pause. An analysis of over 50,000 health-related search queries in Germany reveals that Google's AI Overviews – the AI-generated summaries at the top of search results – cite YouTube more frequently than any medical website.",[11,442,443],{},"The numbers are sobering:",[445,446,447,455,461,467],"ul",{},[448,449,450,454],"li",{},[451,452,453],"strong",{},"YouTube",": 4.43% of all citations (20,621 out of 465,823)",[448,456,457,460],{},[451,458,459],{},"NDR.de"," (German public broadcaster): 3.04%",[448,462,463,466],{},[451,464,465],{},"MSD Manuals",": 2.08%",[448,468,469,472],{},[451,470,471],{},"Netdoktor.de",": 1.61%",[11,474,475],{},"No hospital network or academic institution comes close to YouTube's dominance. Academic journals represent just 0.48% of all citations. Government health authorities combined account for less than 1%.",[25,477,479],{"id":478},"the-problem-is-not-youtube-itself","The problem is not YouTube itself",[11,481,482],{},"Let's be fair: valuable medical content exists on YouTube. University hospitals publish surgical videos. Cardiologists explain ECG interpretation. Dermatologists demonstrate diagnostic techniques.",[11,484,485],{},"Google's own defence points to this: of the 25 most-cited YouTube videos in the study, 96% came from medical channels.",[11,487,488],{},"But here's the catch: those 25 videos represent less than 1% of all YouTube links that AI Overviews cite for health questions. What about the other 99%? The researchers are cautious: \"With the rest of the videos, the situation could be very different.\"",[25,490,492],{"id":491},"popularity-versus-reliability","Popularity versus reliability",[11,494,495],{},"Hannah van Kolfschooten, a researcher at the University of Basel specialising in AI and health law, identifies the core issue:",[183,497,498],{},[11,499,500],{},"\"The findings show that these risks are embedded in the way AI Overviews are designed. In particular, the heavy reliance on YouTube rather than on public health authorities or medical institutions suggests that visibility and popularity, rather than medical reliability, is the central driver for health knowledge.\"",[11,502,503],{},"That's the problem. The algorithms determining what AI systems cite are not optimised for medical accuracy. They optimise for engagement, views, and click behaviour.",[11,505,506,510],{},[15,507,509],{"href":508,"target":18},"https://seranking.com/blog/health-ai-overviews-youtube-vs-medical-sites/","The researchers"," estimate that only 34% of cited sources have formal safeguards for medical reliability. The remaining 66% do not. A wellness influencer with a million views can appear algorithmically \"more valuable\" than a clinical guideline. Broscience and lifestyle gurus get the same weight as peer-reviewed research.",[25,512,514],{"id":513},"the-loop-closes","The loop closes",[11,516,517],{},"Online discussions about this research point to a disturbing phenomenon: AI systems are increasingly citing AI-generated content. Videos with AI voiceovers and AI-generated imagery appear as sources for AI summaries.",[11,519,520],{},"This creates a closed circuit where information can no longer be traced back to primary sources like clinical trials or consensus guidelines. The origin of medical claims becomes increasingly opaque.",[11,522,523],{},"For consumers, this is worrying. For healthcare professionals, it's simply unworkable.",[25,525,527],{"id":526},"what-healthcare-professionals-actually-need","What healthcare professionals actually need",[11,529,530],{},"A GP who needs to quickly verify whether a medication is safe during pregnancy doesn't need a YouTube compilation. A nurse looking for the latest sepsis guidelines gains nothing from content that's popular but potentially outdated.",[11,532,533],{},"Healthcare professionals need information that's traceable to specific sources, based on current guidelines, and tailored to clinical practice.",[11,535,536],{},"General-purpose AI search engines can't provide this. They're built to answer the average question from the average user – not to meet the requirements of professional medical decision-making.",[25,538,540],{"id":539},"how-ask-aletta-does-things-differently","How Ask Aletta does things differently",[11,542,543],{},"At Ask Aletta, we take a different approach: the source determines the value, not popularity.",[11,545,546],{},"Our medical AI assistant draws exclusively from verified sources: clinical guidelines and peer-reviewed medical literature. We don't use YouTube, wellness blogs, or other content of unclear origin.",[11,548,549],{},"Every answer includes direct links to sources. Not because we use transparency as a marketing term, but because healthcare professionals have the right to verify what an AI system claims.",[11,551,552],{},"Ask Aletta also understands your discipline. A GP receives different information than a paediatrician or an emergency physician. Not because the medical facts differ, but because the relevant guidelines and clinical context are different.",[25,554,556],{"id":555},"the-future-of-medical-information","The future of medical information",[11,558,559],{},"The SE Ranking study is a snapshot from December 2025, conducted in Germany. The researchers acknowledge limitations: results may vary by region, by time period, by question phrasing.",[11,561,562],{},"But the underlying dynamic is universal. As long as AI systems optimise for popularity and engagement, they will continue to cite suboptimal sources for medical queries.",[11,564,565],{},"The solution doesn't lie in improving general-purpose AI search engines. The solution lies in specialised systems designed from the ground up for medical reliability.",{"title":129,"searchDepth":130,"depth":130,"links":567},[568,569,570,571,572,573],{"id":478,"depth":130,"text":479},{"id":491,"depth":130,"text":492},{"id":513,"depth":130,"text":514},{"id":526,"depth":130,"text":527},{"id":539,"depth":130,"text":540},{"id":555,"depth":130,"text":556},"Google's AI cites YouTube more than medical sources. What does this mean for healthcare professionals?",{"date":576,"category":141,"author":142,"translations":577,"relatedPosts":579},"2026-01-27",{"nl":578},"20260127-youtube-medische-betrouwbaarheid",[423],{"title":430,"description":574},{"loc":82},"en/updates/20260127-youtube-medical-reliability","bpvjtgHGm402EVGpVayB_1J9ZVj2Ofw_VYH1hnNNabI",{"id":585,"title":586,"body":587,"description":682,"extension":138,"meta":683,"navigation":154,"ogImage":155,"path":688,"seo":689,"sitemap":690,"stem":691,"__hash__":692},"updates/en/updates/20251230-data-leaks-and-ai.md","Data leaks through AI: why healthcare must be extra vigilant",{"type":8,"value":588,"toc":676},[589,598,602,605,608,612,615,618,621,625,628,631,635,638,646,649,673],[11,590,591,592,597],{},"Last weekend, Dutch financial newspaper Het Financieele Dagblad published ",[15,593,596],{"href":594,"rel":595},"https://fd.nl/bedrijfsleven/1582289/tientallen-meldingen-over-datalekken-door-ai-gebruik-privacywaakhond-waarschuwt-voor-risico-s",[179],"an article"," about a trend that doesn't surprise us, but does concern us: the number of data leaks caused by AI use in the workplace is growing. The Dutch Data Protection Authority received dozens of reports in 2024 and 2025, and the number keeps rising.",[25,599,601],{"id":600},"what-went-wrong-in-eindhoven","What went wrong in Eindhoven?",[11,603,604],{},"The FD describes an incident at the municipality of Eindhoven. After a sample period of just 30 days, it turned out that employees had shared CVs, youth care documents, and internal reports with free AI chatbots like ChatGPT and Claude.",[11,606,607],{},"How big the leak actually is, the municipality doesn't know. The data was only retained for 30 days. After that, the scope can no longer be determined.",[25,609,611],{"id":610},"why-this-directly-affects-healthcare","Why this directly affects healthcare",[11,613,614],{},"The article explicitly mentions youth care documents. Those are by definition documents containing sensitive patient information, protected by medical confidentiality.",[11,616,617],{},"But the problem is broader. In daily practice, healthcare professionals copy and paste dozens of times a day. From EHR to letter, from lab result to consult, from note to search query. It's efficient and inevitable.",[11,619,620],{},"But in that hectic flow, it can happen: you accidentally include personal data in a question to an AI tool. With free tools like ChatGPT, that information disappears into a black box. You don't know where it goes, whether it's used for training, or whether it ever comes back in answers to other users.",[25,622,624],{"id":623},"shadow-ai-the-invisible-risk","Shadow AI: the invisible risk",[11,626,627],{},"The FD introduces the term \"shadow AI\": employees who, on their own initiative, use free AI tools, out of sight of the organisation. Even when an organisation offers paid, secure alternatives, people often reach for what they know.",[11,629,630],{},"The solution isn't to ban AI. That doesn't work, and isn't desirable either. The solution is to provide secure alternatives that make the right behaviour easy.",[25,632,634],{"id":633},"how-ask-aletta-does-this-differently","How Ask Aletta does this differently",[11,636,637],{},"We have taken this risk seriously from the start. That's why we built automatic PII detection into Ask Aletta.",[11,639,640,641,645],{},"Every question is automatically checked for personal data, ",[642,643,644],"em",{},"before"," it is processed. We do this with our own detection model, running entirely on our own infrastructure in the Netherlands. No external cloud providers. No third parties.",[11,647,648],{},"If we detect something? You get a notification immediately as a user. With one click, you anonymise the sensitive data and your search query can still be carried out safely.",[650,651,195,654,195,664],"div",{"className":652},[653],"image-row",[192,655,656,657,656,661,195],{},"\n    ",[197,658],{"src":659,"alt":660},"/images/updates/pii-detection-nl.png","PII detection warning in Ask Aletta",[247,662,663],{},"Detection of personal data",[192,665,656,666,656,670,195],{},[197,667],{"src":668,"alt":669},"/images/updates/pii-detection-masked-nl.png","Anonymised question in Ask Aletta",[247,671,672],{},"Anonymised with one click",[11,674,675],{},"In the rush of daily practice, a small mistake is quickly made. We help you prevent it.",{"title":129,"searchDepth":130,"depth":130,"links":677},[678,679,680,681],{"id":600,"depth":130,"text":601},{"id":610,"depth":130,"text":611},{"id":623,"depth":130,"text":624},{"id":633,"depth":130,"text":634},"The Dutch Data Protection Authority is sounding the alarm. We explain what this means for healthcare professionals.",{"date":684,"category":141,"author":142,"translations":685,"relatedPosts":687},"2025-12-30",{"nl":686},"20251230-datalekken-en-ai",[153],"/en/updates/20251230-data-leaks-and-ai",{"title":586,"description":682},{"loc":688},"en/updates/20251230-data-leaks-and-ai","WM0biZbCBxVgYzA7LFRUb8zp5t79ip8Jqsj-CyvWQoE",{"id":694,"title":695,"body":696,"description":731,"extension":138,"meta":732,"navigation":154,"ogImage":155,"path":737,"seo":738,"sitemap":739,"stem":740,"__hash__":741},"updates/en/updates/20251216-shoko-partnership.md","Ask Aletta and SHoKo GP emergency post collaborate",{"type":8,"value":697,"toc":728},[698,701,710,721,725],[11,699,700],{},"Do you work as a GP, practice nurse, or other healthcare provider in Brabant's Kempen region? Then there's a good chance you can use Ask Aletta.",[11,702,703,704,709],{},"Ask Aletta and ",[15,705,708],{"href":706,"rel":707},"https://shoko.nl/",[179],"SHoKo"," GP emergency post (Samenwerkende Huisarts Organisatie Kempen en Omstreken) are working together, so that more than 50 healthcare professionals at this emergency post can safely and quickly consult both national guidelines and their own protocols.",[183,711,712,715],{},[11,713,714],{},"\"Ask Aletta searches more guidelines and protocols than I would normally ever open and gives me a quick, clear answer.\"",[11,716,717,720],{},[451,718,719],{},"Ylva Onderwater",", GP and medical director of SHoKo",[25,722,724],{"id":723},"flexible-access","Flexible access",[11,726,727],{},"Together we have set up a solution where not only permanent staff have access, but also healthcare providers who work there occasionally can use Ask Aletta.",{"title":129,"searchDepth":130,"depth":130,"links":729},[730],{"id":723,"depth":130,"text":724},"More than 50 healthcare professionals at the emergency post in Brabant's Kempen region can now use Ask Aletta.",{"date":733,"category":141,"author":734,"translations":735},"2025-12-16","luuk.gruijs",{"nl":736},"20251216-shoko-samenwerking","/en/updates/20251216-shoko-partnership",{"title":695,"description":731},{"loc":737},"en/updates/20251216-shoko-partnership","mzxzyXLmciSDnUa4LX_KP7xjDINuGexHnhNpobGRWds",{"id":743,"title":744,"body":745,"description":777,"extension":138,"meta":778,"navigation":154,"ogImage":155,"path":780,"seo":781,"sitemap":782,"stem":783,"__hash__":784},"updates/en/updates/20251213-skipr-blog.md","Ask Aletta as an example of responsible AI in healthcare",{"type":8,"value":746,"toc":774},[747,752,761,764,767,771],[183,748,749],{},[11,750,751],{},"\"A good example of how it should be done is the Dutch AI program Ask Aletta\"",[11,753,754,755,760],{},"Writes em. professor Robert Kreis in his blog ",[15,756,759],{"href":757,"rel":758},"https://www.skipr.nl/blog/ai-in-de-zorg-laat-manipulerende-algoritmes-maar-weg/",[179],"on Skipr",".",[11,762,763],{},"In a time when AI tools are often powered by opaque algorithms and unverifiable data, he advocates for a different approach: reliable sources, transparency, and no manipulation.",[11,765,766],{},"He mentions Ask Aletta as an example of a medical search engine that - unlike ChatGPT or Microsoft Copilot for example - is entirely based on verified literature, with source citations for every answer.",[25,768,770],{"id":769},"thats-exactly-what-were-building-for","That's exactly what we're building for",[11,772,773],{},"Transparency and reliability are central to everything we do. Every answer Ask Aletta gives is based on sources you know and trust, with direct references so you can always verify where the information comes from.",{"title":129,"searchDepth":130,"depth":130,"links":775},[776],{"id":769,"depth":130,"text":770},"Em. professor Robert Kreis mentions Ask Aletta as a good example of how AI in healthcare should be done in his blog on Skipr.",{"date":779,"category":141,"author":310},"2025-12-13","/en/updates/20251213-skipr-blog",{"title":744,"description":777},{"loc":780},"en/updates/20251213-skipr-blog","hqyt9lldjIGQLG3-PtMCgLvWP-lKqVZcAS5KYoRju2g",{"id":786,"title":787,"body":788,"description":811,"extension":138,"meta":812,"navigation":154,"ogImage":155,"path":814,"seo":815,"sitemap":816,"stem":817,"__hash__":818},"updates/en/updates/20251124-dekinderkliniek.md","New partnership with DeKinderkliniek",{"type":8,"value":789,"toc":808},[790,798,801,805],[11,791,792,797],{},[15,793,796],{"href":794,"rel":795},"https://dekinderkliniek.nl/",[179],"DeKinderkliniek",", a pediatric hospital in Almere, was born from the conviction that pediatrics can be done differently. That vision is reflected in everything they do: innovative, efficient, and with the child and family at the center.",[11,799,800],{},"With medical and psychological care under one roof, short lines and quick decision-making, they make care more personal and accessible. Exactly the type of organization where our AI search engine adds direct value.",[25,802,804],{"id":803},"the-partnership","The partnership",[11,806,807],{},"From now on we take an extra step together: all healthcare professionals at DeKinderkliniek have access to Ask Aletta, which supports them within seconds with reliable medical information.",{"title":129,"searchDepth":130,"depth":130,"links":809},[810],{"id":803,"depth":130,"text":804},"Pediatric hospital DeKinderkliniek in Almere chooses Ask Aletta to support their healthcare professionals with fast, reliable medical information.",{"date":813,"category":141,"author":734},"2025-11-24","/en/updates/20251124-dekinderkliniek",{"title":787,"description":811},{"loc":814},"en/updates/20251124-dekinderkliniek","HQrIhEICWbrMtQn4D7EaPJ5LrISV4FFHnLq6_8My1YY",{"id":820,"title":821,"body":822,"description":857,"extension":138,"meta":858,"navigation":154,"ogImage":155,"path":860,"seo":861,"sitemap":862,"stem":863,"__hash__":864},"updates/en/updates/20251017-flexdokters.md","All Flexdokters GP practices get access to Ask Aletta",{"type":8,"value":823,"toc":855},[824,827,830,833,841,852],[11,825,826],{},"Every day we see GPs doing their utmost to deliver good care - often under high pressure, with limited time and endless streams of information.",[11,828,829],{},"That's exactly why Ask Aletta exists: to help them quickly access reliable medical knowledge, so more time remains for what really matters: the patient.",[11,831,832],{},"Today we take a great step in that direction: all GP practices of Flexdokters get access to Ask Aletta's AI search engine.",[11,834,835,840],{},[15,836,839],{"href":837,"rel":838},"https://flexdokters.nl/",[179],"Flexdokters"," is a cooperative of independent practice owners who, like us, believe in smart, modern support for healthcare professionals.",[183,842,843,846],{},[11,844,845],{},"\"Ask Aletta's AI search engine helps our doctors quickly find reliable medical information, creating more time for the patient. This partnership fits perfectly with our mission to make practice ownership more attractive.\"",[11,847,848,851],{},[451,849,850],{},"Herm Jan Mateboer",", COO Flexdokters",[11,853,854],{},"Together we work on healthcare that is better and more enjoyable - without losing the human touch.",{"title":129,"searchDepth":130,"depth":130,"links":856},[],"Flexdokters, a cooperative of independent practice owners, chooses Ask Aletta to support their doctors with fast, reliable medical information.",{"date":859,"category":141,"author":734},"2025-10-17","/en/updates/20251017-flexdokters",{"title":821,"description":857},{"loc":860},"en/updates/20251017-flexdokters","QDjmJqmwt-KklO_cmcZJ33s4V9Iyoc4SzXWEkJC6Vhw",1778680910643]