What AI can tell you about your blood test

The limbo between having blood drawn and receiving the results can be stressful for patients. Then the jargon-filled blood test report arrives before a doctor can review it and translate the findings. They may never follow up if they see no cause for alarm. 

Enter the artificial intelligence-powered chatbot. Since major large language models like ChatGPT, Claude, and Gemini debuted a few years ago, an increasing number of patients have shared their lab results — or uploaded them to — the chatbot of their choice, seeking guidance. 

Companies, including the wellness and lifestyle brands Whoop and Levels, see an opportunity in this information gap and have made a compelling pitch: Concierge-level attention to a consumer’s lab work, courtesy of AI. 

Typically, their AI product is an explanatory report, written in accessible language, that provides a personalized plan with next steps, like dietary changes, lifestyle modifications, and consultation with a doctor. The service, which is typically available with a subscription, can cost a few hundred dollars or more per year.  

Dr. John Whyte, CEO of the American Medical Association, understands the appeal, especially when patients find their results confusing. 

“Physicians are [not always] the best communicators,” Whyte says. “I wish we were, and [that we] had more time.” 

Still, he says there’s no rigorous research or evidence demonstrating that AI can effectively and accurately interpret blood results and make personalized lifestyle recommendations to improve or optimize one’s health. In other words, the companies offering AI interpretation of blood tests don’t yet know if their product is better than simply consulting a chatbot for free, or more accurate than a doctor’s opinion.  

“I think you have to be skeptical about some of the claims,” Whyte says. 

________________________________________________________________________________________________________

A new Mashable series, AI + Health, will examine how artificial intelligence is changing the medical and health landscape. We’ll explore how to keep your health data safe, prompting chatbots effectively for health questions, and learn how two women are using AI to detect a dangerous form of heart disease, and much more.

________________________________________________________________________________________________________

AI for blood test results still has a long way to go.

The AI models, including Gemini and ChatGPT, used by companies that do blood interpretation work aren’t validated or benchmarked for this purpose, according to their makers, Google and OpenAI.

When Google recently partnered with Quest Diagnostics to bring an AI tool to customers of the nationwide lab work company, it focused on explaining medical terminology, identifying trends in their personal data, and suggesting questions to ask their doctor. It does not give medical guidance or lifestyle advice.

A spokesperson for the company said that while Google frequently publishes research on Gemini’s medical benchmarks, the Quest partnership is designed to “solve real-world literacy and data-navigation challenges for patients.”

OpenAI’s HealthBench, which tests how well its models perform on realistic health scenarios, includes examples of understanding blood results, according to a company spokesperson. Still, OpenAI doesn’t have a standalone benchmark for blood testing. 

Jonathan Kron, co-founder and CEO of the company BloodGPT, readily acknowledges that there are no widely accepted benchmarks for comprehensively interpreting blood tests at scale. 

During its early testing in 2024, BloodGPT found that uploading full lab reports directly to general-purpose chatbots like Claude and ChatGPT consistently yielded errors. Biomarkers were missed entirely or confused with each other. Sometimes the chatbot hallucinated recommendations. 

These experiences prompted BloodGPT to build what it calls a “structured pipeline” with multiple checks for validating its findings. While BloodGPT can swap out large language models based on their evolving strengths, the company currently uses enterprise Gemini, OpenAI, and Anthropic models for different tasks, including classification, reasoning, interpretation, and consistency checking. It also consults with specialists on specific biomarkers, like the reproductive hormones estrogen and testosterone, to check the accuracy and insight of its AI. 

Though BloodGPT considers its primary business selling software to other companies, including clinical laboratories, healthcare systems, and health diagnostic companies like LabCorp, it offers AI interpretation to individual consumers. Their consumer plans range from $9.99 to $17.99 per month.

The company says its algorithms are based on established clinical guidelines and validated medical reference data, and are tested iteratively by clinicians. BloodGPT hasn’t published peer-reviewed research proving the success of its methods — yet. 

Kron says the company will embark on a massive research project using 100,000 de-identified patient records through a partnership with an Israeli health system. The goal is to benchmark BloodGPT’s accuracy in multiple ways against the results contained in the patients’ medical records. 

“We’re not choosing the easy way here,” Kron says. 

Why model accuracy is important

Dr. Girish N. Nadkarni, an internist and nephrologist at New York’s Mt. Sinai Hospital, argues that companies selling an AI blood interpretation service need to demonstrate their success by comparing it retrospectively against de-identified patient data and by enrolling people in a prospective study that compares their AI findings to an expert’s.

“I don’t think that anything has to have 100 percent accuracy to succeed, because humans are not 100 percent accurate,” says Nadkarni, an AI health researcher and director of Mt. Sinai’s Hasso Plattner Institute for Digital Health. “But the conversation becomes really hard because I don’t know what the accuracy of this model is…how does it work, and where does it fail?” 

Nadkarni says current AI blood work products might be “OK” for a majority of patients, but problems may emerge at the extremes, such as missed diagnoses or false positives that cause anxiety and potentially harmful additional testing. 

Levels CEO Josh Clemente is an advocate of frequent blood testing. If cost weren’t a concern, Clemente would recommend lab work far more often than the typical annual physical, for pro-active health monitoring.

Levels, which focuses on metabolic health, markets its subscription plans as a way to “live healthier, longer.” Each tier offers access to the company’s app and proprietary dashboards, along with glucose monitoring. But only two tiers, available for $499 and $1,499 annually, include lab work, clinician-reviewed reports, and AI health insights. 

Clemente currently favors Claude and Gemini for the Levels AI product, which is also trained on medical articles and biomarker research, plus Levels blog posts and podcasts featuring metabolic experts on topics like diet, weight loss, and hormones. 

The AI product derives its insight from these layers of information and guardrails, and a doctor reviews every lab work report that a consumer receives. An app-based AI chatbot also draws from Levels expert content to recommend lifestyle changes that could improve biomarkers like cholesterol and glucose. 

“We’re using it again as a clinician support tool, which is, in my opinion, the right way to use these tools today,” Clemente says. Levels is not currently conducting independent research on the accuracy of its AI product. 

Nadkarni supports human oversight, but cautions consumers against believing this is a failsafe. Instead, physicians can unwittingly fall victim to automation bias, or the tendency to rubber-stamp AI outputs. 

Can AI truly personalize blood test results?

Whoop, the wearable made famous by athletes like soccer star Cristiano Ronaldo and quarterback Patrick Mahomes, just began offering blood testing last fall. One annual test is $199, but members can test six times a year for $899. The company partners with Quest to perform the lab work and, like Levels, has a physician review the results. 

Alexi Coffey, vice president of product at Whoop, says the company’s AI product is powered by OpenAI and individual member data. She declined to comment on whether the company is pursuing clinical research on the accuracy of its AI. 

“We never want to over claim or over suggest relationships between things,” Coffey says, “but we do want to provide value to our members by helping them understand things that may be connected.” 

Coffey says that Whoop’s ability to integrate physiological data — including heart rate, sleep quality, and exercise patterns — into blood work reports gives customers unique health insights. 

If, for example, a customer ran a marathon the day before their blood was drawn, the AI might take the physical stress into account when interpreting inflammation biomarkers. (Whoop actually advises its members to avoid strenuous physical activity prior to a blood draw because it may influence the results.) 

As promising as this sounds, Whyte, of the AMA, says there’s no evidence demonstrating that such AI applications are “truly personalized.” 

“I think we really have to question what the data is,” he says, noting that companies would need a massive dataset to make individual recommendations. If these companies have that information, they haven’t published peer-reviewed research based on it yet.

Blood tests are crucial diagnostic tools, Whyte says, but they’re also the focus of social media hype, which often positions them as a silver bullet for fixing health challenges like low energy or sleeplessness. He warns consumers that some companies measure elements and minerals that “are not that helpful for anything.” Hormone level testing can be similarly futile, depending on how it’s done.

Until scientific research establishes the accuracy and meaningfulness of AI blood test interpretation, Whyte recommends patients use AI to help plainly explain their results, rather than relying on the technology to generate personalized insights for them. 

“People think a lab test is black and white and it’s the final ruler on whether you have disease or not,” Whyte says. “And that’s not always true.”

________________________________________________________________________________________________________

The information contained in this article is for educational and informational purposes only and is not intended as health or medical advice. Always consult a physician or other qualified health provider regarding any questions you may have about a medical condition or health objectives.

Disclosure: Ziff Davis, Mashable’s parent company, in April 2025 filed a lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.

​Mashable