Best AI for Endocrinology: Cited Answers on Diabetes and Hormones

Which AI tools answer endocrinology questions?

This guide compares the AI tools that endocrinologists, diabetes nurses, pharmacists, residents and students weigh for questions on diabetes, GLP-1 receptor agonists, the thyroid, the adrenal glands and insulin resistance, and each tool does a different job. AskMed is an AI medical reference that answers from journals, clinical guidelines and drug labels, including the American Diabetes Association's journals, with a link after every claim, for anyone in medicine in any country. OpenEvidence is a clinical AI search engine for verified healthcare professionals; its App Store listing says an NPI is required and names the American Diabetes Association among its content agreements. UpToDate Expert AI is the AI companion to the UpToDate clinical reference and needs an UpToDate subscription with Expert AI enabled. ChatGPT is OpenAI's general chatbot. AI scribes, glucose-monitor apps and software that reads ultrasound images are different categories and are left out. The table shows what each tool is, where its answers come from and who can use it.

OpenEvidence

OpenEvidence describes itself as a clinical decision support platform and medical search engine. Its US App Store listing names NEJM, JAMA and its 11 specialty journals, NCCN, Nature, the Cochrane Library and Wiley as official AI partners, and lists content agreements and collaborations with medical societies, the American Diabetes Association among them. The listing marks an NPI as required, says users must verify their healthcare professional status, and shows that the app contains advertising. OpenEvidence's Terms of Use say the service is not available in the European Union, the United Kingdom or the European Economic Area, so diabetes and endocrine teams there need another tool. AskMed vs OpenEvidence compares the two in detail.

UpToDate Expert AI

UpToDate Expert AI is an AI-powered companion to the UpToDate clinical reference. Its App Store listing says answers are grounded in UpToDate content and can be traced to supporting sources, and that access needs an UpToDate subscription with Expert AI enabled, through an enterprise account or a personal subscription. It suits clinicians whose hospital already uses UpToDate. AskMed vs UpToDate Expert AI compares the two side by side.

ChatGPT

ChatGPT is OpenAI's general AI chatbot, on the web, iPhone and Android. It helps to talk a feedback loop through, such as the hypothalamic-pituitary-adrenal axis, but it is not limited to medical sources, so check any clinical claim against the guideline first. OpenAI also runs ChatGPT for Clinicians, which its help center opens to clinicians in the United States who verify with an NPI and a license check. Is ChatGPT accurate for medical questions? covers what studies found.

The guidelines themselves

Many endocrine clinicians keep the guideline open next to their AI tool. The American Diabetes Association publishes its Standards of Care in Diabetes each year as a supplement to Diabetes Care and says the online version is kept current through a "living" Standards process. The American Thyroid Association lists its guidelines on thyroid.org, and the Endocrine Society and the European Society of Endocrinology publish theirs in their own journals. These are for reading the guideline itself, which is also where an AI answer should lead you.

How does AskMed answer an endocrinology question?

AskMed answers an endocrinology question in four steps that you can check. First, AskMed rewrites the question as a precise medical search query, and for management questions it adds "latest guideline update" so the newest guidance is found. Second, AskMed searches a fixed list of medical sites, never the open web or forums; for diabetes and hormones, that list includes the American Diabetes Association's journals, where the Standards of Care appear in Diabetes Care, plus NICE, KDIGO, the European Society of Cardiology, NEJM, The Lancet, JAMA, PubMed, PubMed Central and the Cochrane Library, and FDA and DailyMed drug labels. Third, AskMed ranks what it found: guidelines and health agencies first, then systematic reviews and meta-analyses, then reviews and references, then large trials. Fourth, an AI language model writes the answer with a link after every claim, gives the year of each guideline recommendation, and states HbA1c in both % and mmol/mol and glucose in both mg/dL and mmol/L. If no usable source turns up, AskMed says so.

Endocrine society guidelines reach AskMed in different ways. The 2015 American Thyroid Association guideline on thyroid nodules is in full on PubMed Central, while the 2023 European Society of Endocrinology guideline on adrenal incidentalomas and the 2025 Endocrine Society guideline on primary aldosteronism appear on PubMed. For evidence questions such as "Do GLP-1 receptor agonists reduce kidney outcomes in type 2 diabetes?", Research Mode reads the most relevant source in full before it writes the answer. How AskMed answers lists every source.

Can AI answer ADA Standards of Care questions on diabetes and GLP-1 drugs?

AI can answer ADA Standards of Care questions when it searches the Standards themselves and names the edition, and AskMed searches diabetesjournals.org, where the American Diabetes Association publishes them in Diabetes Care. The current edition is the Standards of Care in Diabetes—2026, released on 8 December 2025 as Supplement 1 of Diabetes Care volume 49, and the ADA says the online version is kept current through its "living" Standards process. Section 2 covers diagnosis, Section 9 pharmacologic approaches to glycemic treatment, and Section 10 cardiovascular disease and risk management, which calls SGLT inhibitors and GLP-1 receptor agonists with demonstrated cardiovascular and kidney benefit a fundamental element of risk reduction in type 2 diabetes. The ADA's list of 2026 updates includes obesity treatment in type 1 diabetes, glucose-lowering therapy in chronic kidney disease including dialysis, and revised blood pressure goals. AskMed explains each drug class, how it is chosen and what to monitor, but never gives insulin or GLP-1 receptor agonist doses or titration schedules.

• Diagnosis. Give the results and ask whether they meet the criteria in Section 2. The thresholds are an HbA1c of 6.5% (48 mmol/mol), a fasting plasma glucose of 126 mg/dL (7.0 mmol/L) or a 2-hour glucose of 200 mg/dL (11.1 mmol/L) on an oral glucose tolerance test, and AskMed gives both units.
• Choosing a drug class. Ask "Which glucose-lowering classes does the ADA favor in type 2 diabetes with heart failure or chronic kidney disease, and on what evidence?" The sources to open are Section 9 and Section 10.
• GLP-1 receptor agonists. Mechanism, gastrointestinal side effects, what the US labels on FDA and DailyMed list as contraindications and warnings, and how the class compares with SGLT2 inhibitors, all without doses.
• Pharmacists. Interactions between named diabetes drugs and the rest of a medication list, with the label linked; amounts and schedules come from the label and the prescriber.

Can AI help with a thyroid nodule work-up?

AI can help with a thyroid nodule work-up by setting the ultrasound pattern, nodule size and TSH result against the guideline that applies, with a link to each rule, but it cannot look at the ultrasound image itself. The American Thyroid Association's guideline page still lists the 2015 ATA guideline, published in Thyroid in 2016, for adult thyroid nodules; the newer 2025 ATA guideline covers differentiated thyroid cancer only, from diagnosis onward. The 2015 guideline recommends measuring serum TSH first and doing a radionuclide scan only if TSH is subnormal, then fine-needle aspiration by sonographic pattern and size: from 1 cm for high and intermediate suspicion, from 1.5 cm for low suspicion, and as an option from 2 cm for very low suspicion, never for purely cystic nodules. Many radiology reports use the American College of Radiology's TI-RADS from 2017 instead, and cytology comes back in the six categories of the 2023 Bethesda System.

• Paste the report, not the picture. Give the ultrasound report's wording and ask which ATA pattern and TI-RADS level it fits and whether FNA is advised. The 2015 ATA guideline and the ACR TI-RADS white paper are the sources to open.
• Cytology. Ask what a category means and what comes next: the 2023 Bethesda System names six categories, from nondiagnostic to malignant, each with an implied risk of malignancy, and splits atypia of undetermined significance into two subgroups.
• Thyroid cancer. The 2025 ATA guideline covers staging and risk assessment, initial treatment decisions, response assessment and monitoring for differentiated thyroid cancer.

Can AI help work up an adrenal incidentaloma?

AI can help work up an adrenal incidentaloma by laying out the two questions every guideline asks, whether the mass could be malignant and whether it makes too much hormone, with the source for each step. The current European guideline is the 2023 European Society of Endocrinology guideline, written with the European Network for the Study of Adrenal Tumors and published in the European Journal of Endocrinology. It says every adrenal mass needs dedicated adrenal imaging, and that a homogeneous lesion of 10 Hounsfield units or less on unenhanced CT is benign and needs no further imaging, whatever its size. Every patient also needs plasma or urinary metanephrines and an overnight dexamethasone suppression test read against a cortisol cut-off of 50 nmol/L (1.8 µg/dL); a higher value in someone without overt Cushing's syndrome is what the guideline calls mild autonomous cortisol secretion. AskMed gives such lab cut-offs in both units, while the test's drug amount comes from the local protocol.

The ESE/ENSAT guideline adds that surgery is not usually indicated for an asymptomatic, nonfunctioning unilateral mass with obviously benign imaging. The Endocrine Society's 2025 primary aldosteronism guideline rates its screening suggestion as conditional and also suggests a dexamethasone suppression test in everyone with primary aldosteronism and an adrenal adenoma. Ask AskMed how the two guidelines fit together for a given patient, and open both.

How can AI help measure insulin resistance?

AI can help measure insulin resistance by explaining which method fits the question and what its number means, since no fasting index has a universal cut-off. The reference method is the hyperinsulinemic euglycemic clamp, described by DeFronzo and colleagues in 1979: insulin is held high and glucose held at its basal level, so the glucose infusion rate measures how sensitive the tissues are to insulin. The clamp is too laborious for clinics, so fasting indexes stand in for it. HOMA-IR, from Matthews and colleagues in 1985, multiplies fasting insulin by fasting glucose and divides by 22.5, with glucose in mmol/L. QUICKI, from Katz and colleagues in 2000, is 1 divided by the sum of the logarithms of fasting insulin and glucose, and correlated with clamp results at r = 0.78. The TyG index, tested in 748 adults in 2008, uses fasting triglycerides and glucose instead of insulin, for laboratories without an insulin assay.

• Work out HOMA-IR. The HOMA-IR calculator on getaskmed.com takes glucose in mg/dL or mmol/L and insulin in µU/mL or pmol/L, shows HOMA-%B too, and links the 1985 paper.
• Read the number. HOMA-IR normal range explains why published cut-offs run from 1.85 to 3.80 in adult studies, depending on population, assay and method.
• Know the limits. Wallace, Levy and Matthews, from the same Oxford group, wrote in 2004 on the use and abuse of HOMA modelling, including why one fasting sample is imprecise in one person. Ask AskMed which index a study used before you compare its numbers with yours.

What should endocrinology fellows, diabetes nurses, pharmacists and students look for in an AI?

Endocrinology fellows, diabetes nurses and educators, pharmacists, residents and students need an AI tool whose answers can be checked in seconds and that names the guideline edition it quotes. Four checks matter. First, citations: each claim should link to the guideline, trial or drug label behind it, not to a reading list at the end. Second, the edition: the ADA updates its Standards of Care every year and online in between, while the ATA's nodule guideline dates from 2015, so an answer should give the year and say what changed. Third, units: HbA1c and glucose are reported differently between countries, and AskMed checks its own draft so that both units are given. Fourth, access: OpenEvidence's App Store listing says an NPI is required and its Terms exclude the UK, EU and EEA, and UpToDate Expert AI needs an UpToDate subscription with Expert AI enabled. AskMed needs no NPI or license, works in any country and carries no pharma advertising, on iPhone, iPad, Android and the web.

• Fellows and residents. Research Mode for journal club and for "what changed in this year's Standards of Care?" questions on rounds.
• Diabetes nurses and educators. Recognizing hypoglycemia, what to monitor on insulin or a GLP-1 receptor agonist and how drug classes differ, with sources and without doses.
• Students. Paste an exam-style question with its options: the first sentence names the correct option and the answer explains why each other option is wrong.
• Pharmacists and researchers. Drug label warnings, interactions between named drugs and where trial evidence agrees or conflicts. Best AI for pharmacists covers drug information tools.

What are the limits of AI in endocrinology?

The main limit of AI in endocrinology is that an answer can only be as good as its sources and the question it was given, and endocrine guidance changes at different speeds, yearly for the ADA Standards of Care but less often for thyroid and adrenal guidelines. AskMed searches a fixed list of medical sites, so an Endocrine Society or European Society of Endocrinology guideline may reach AskMed only as a PubMed abstract, and every AskMed answer is written by an AI language model that can misread a source; no person reviews it before you see it. Open the citation behind any sentence you act on and check it covers the same patients: evidence in type 2 diabetes does not answer a question about type 1. AskMed reads text, not ultrasound or CT images, and gives no doses; take those from the current label. AskMed is for education and reference by healthcare professionals and students, not for emergencies or for patients' self-diagnosis.

• Check the source. Every AskMed claim has its own link, so you check one sentence against one page.
• Check the edition. A current guideline outweighs an older review, and AskMed gives the year of each recommendation.
• Report what looks wrong. Tap the thumbs down under an AskMed answer and pick a reason, such as Inaccurate or Sources don't match.

Best AI for cardiology covers tools for the heart, where diabetes and cardiovascular risk meet, and best AI for nurses covers tools for nursing questions.