Skip to main content
All articles
Technology July 14, 2026 6 min read

AI Chatbots Just Took the Endodontic Board Exam. Here's How They Did — and What It Means for You

A July 2026 Journal of Endodontics study had GPT-4o and Gemini 2.5 Pro sit a simulated American Board of Endodontics oral exam. Both scored remarkably high. Here's what that does — and doesn't — mean for your dental care.

By Dr. Jason Kung — Specialist Endodontist · UCLA DDS · OHSU MS

Medically reviewed by Dr. Jason Kung, DDS, MS · Specialist Endodontist · UCLA DDS · OHSU MS ·

Every board-certified endodontist has faced the American Board of Endodontics oral examination — a live, case-based grilling in which examiners probe diagnosis, treatment planning, and the literature behind every decision. In the July 2026 Journal of Endodontics, researchers asked a timely question: how would today's AI chatbots do in that chair?

How the experiment worked

Two academic, board-certified endodontists built three complete oral-board cases — full medical history, dental history, clinical test results — each followed by 20 consecutive open-ended, board-style questions. Two publicly available AI models, GPT-4o and Gemini 2.5 Pro, took the exam. The same two examiners independently graded every answer on a 0–3 rubric for response validity, citation validity, and overall performance.

The results

  • Gemini 2.5 Pro: mean overall score of 2.83 / 3
  • GPT-4o: mean overall score of 2.73 / 3
  • Statistical modeling found no significant difference between the two models' odds of an "excellent" answer.

By any standard, that's a strong showing on specialist-level material. The authors' conclusion, though, was carefully scoped: the results "support the concept of using AI chatbots as aid in endodontic education" — a study tool for residents, a way to rehearse case reasoning. Not a clinician.

What an exam score doesn't measure

An oral-board simulation hands the AI a perfectly documented case: every test already performed, every finding already recorded. Real endodontics is the part that happens before that write-up exists — and after it:

  • Producing the findings. Cold testing, percussion, palpation, bite tests, reading subtle radiographic changes — a chatbot can't touch a tooth. And tooth pain is a notorious impostor: even AI-assisted imaging has real limits.
  • Judgment under ambiguity. Real patients rarely present textbook cases. Referred pain, cracked teeth, and healing anatomy don't come with an answer key.
  • The hands. Locating a calcified canal under a microscope or performing an apicoectomy is a physical skill no language model possesses.

Where AI genuinely helps — including in our practice

We're not AI skeptics. This website has an AI assistant that answers common questions about root canals, insurance, and appointments around the clock — because for education and access, AI is excellent, and this study reinforces that. Where we draw the line is diagnosis and treatment: those stay with a specialist who can actually examine you.

The bottom line

AI scoring 2.8/3 on a board-style exam is a genuine milestone for dental education — and zero reason to skip a proper evaluation. If a tooth is bothering you, ask our chat assistant anything you like, then book with a human specialist. Call (669) 234-2354 — we're open weekends for emergencies.

Have a question about your tooth?

Dr. Kung sees emergency cases the same day when possible. Most consultations are 30 minutes and include a microscope examination.

Related Articles