Welcome to the forefront of conversational AI as we explore the fascinating world of AI chatbots in our dedicated blog series. Discover the latest advancements, applications, and strategies that propel the evolution of chatbot technology. From enhancing customer interactions to streamlining business processes, these articles delve into the innovative ways artificial intelligence is shaping the landscape of automated conversational agents. Whether you’re a business owner, developer, or simply intrigued by the future of interactive technology, join us on this journey to unravel the transformative power and endless possibilities of AI chatbots.
Most people instinctively give doctors more detail than they give apps. A headache becomes a fuller story, where it hurts, how long it has lasted, whether nausea or light sensitivity came with it.
In contrast, digital tools usually get the compressed version. Researchers now say that habit extends to medical chatbots as well.
A new study measured exactly how much information people leave out when they think AI is reading their symptoms — and whether those omissions change how useful the report becomes.
A team led by Moritz Reis, a research associate at the Institute of Psychology at the University of Würzburg (JMU) recruited 500 adults in the United Kingdom for a simple test.
Each person wrote two symptom reports, one for an unusual headache and one for flu-like illness.
Half were told a doctor would read their account. The other half were told an AI chatbot would.
The wording on the page changed, but instructions stayed identical. Reports written for the human doctor averaged 256 characters. Chatbot reports averaged 229 characters, about a sentence shorter.
To check whether shorter also meant worse, the team ran every report through a scoring system.
The question was how useful the description was for deciding who needed urgent care.
A higher score meant a doctor could read your sentences and confidently give advice.
Reports sent to the chatbot scored 8% lower on average. Four licensed physicians reviewed the data, two neurologists and two pulmonologists.
They observed a random subset without knowing whether a report had been written for a doctor or a chatbot. Their judgments lined up with the AI scoring.
What gets cut is the kind of context a doctor builds a full picture from. For instance, how long a headache lasted, or what a cough sounded like at 3 a.m.
None of this is hard to write. However, people simply wrote less of it when they thought a machine was reading.
Researchers traced the quality drop directly to length. Fewer characters meant a less useful report for self-triage – the early-stage filter that decides who needs a doctor right now.
AI tools are usually tested on standardized scenarios, not the messy paragraphs people actually type.
That often hides this problem. A chatbot can ace a benchmark and still misroute a real patient if the patient only gives it half the story.
That quality gap held even among participants who had relevant symptoms at the time, and not just those imagining them.
A separate paper on the accuracy of online symptom checkers reported similar caveats. Lab-grade accuracy doesn’t survive contact with everyday user input.
So why are people stingier with a chatbot? The team describes a phenomenon called uniqueness neglect – the belief that AI sees you as a category, not a person.
If a tool only matches patterns, the thinking goes, why bother spelling out the strange specifics?
“Many people assume that AI cannot grasp the individual nuances of their personal situation and instead merely matches standardized patterns,” explained Professor Wilfried Kunde.
Privacy worries could be part of it too. So might general skepticism about whether an algorithm can actually diagnose anything.
An earlier study by the same group found that people rate identical medical advice as less trustworthy and less worth following the moment they’re told an AI wrote it.
The fix isn’t a smarter model, but a smarter interview. The team argues that medical chatbots should actively prompt users for the details a doctor would ask about.
Details such as duration, severity, and what makes it better or worse – instead of waiting for the user to guess what counts.
Showing concrete examples of strong descriptions could help strengthen medical advice.
So could explaining what the system does with the information. People may type more when they understand the tool’s logic, not less.
“If we don’t trust a machine to understand our uniqueness, we may unconsciously withhold the information it would need to provide precise assistance,” said Reis.
Participants wrote about conditions they were asked to imagine, not situations where they were actually sick and urgently needed care.
The researchers note that real-world reports, where the emotional stakes are higher, could differ in ways this experiment couldn’t capture.
Whether the gap holds in actual clinical encounters remains to be tested and will require further study.
Until now, no one had measured what patients leave out before an AI ever sees their question. Evaluations of medical chatbots focused almost entirely on the model’s side of the conversation.
This study flipped that. It put a number on the human side, showing 8% lower-quality reports, driven by 27 fewer characters, in healthy adults talking to a machine.
Eight percent per person sounds modest. Multiplied across the millions of queries hitting symptom checkers and consumer chatbots, the missing detail confirms triage decisions made on incomplete information.
Developers now have a specific problem to design around. Patients have a reason to type out more, not less, even when no human is reading.
The study is published in Nature Health.
—–
Like what you read? Subscribe to our newsletter for engaging articles, exclusive content, and the latest updates.
Check us out on EarthSnap, a free app brought to you by Eric Ralls and Earth.com.
—–
