An LLM powered day in the GI clinic and endoscopy suite

by Dr Yuri Gorelik (Frontline Gastroenterology Global Taskforce 2025-6)

Since the introduction of ChatGPT 3.5 in November 2022, artificial intelligence (AI) and specifically large language models (LLMs) have been set to revolutionize our daily and work life. Multiple LLM based GI utilities have been developed and researched. In this blog post I will try to provide a mini review in the form of a description of a GI practitioner workday powered by such LLM based utilities and products. Many of the specific LLM applications and use-cases I mention below are discussed in Frontline Gastroenterology’s narrative review of generative AI in colorectal practice (Frontline Gastroenterology, 2025).

Morning Clinic

Upon arrival at the clinic all previous follow-ups and clinic referrals for the clinic patients on your list are already summarized by one of the multiple published tools for such tasks. Soon after, the patients start to come in, all history taking and patient communication is scribed by an LLM such as the Google health medical speech to text tools (Google cloud). This text is also transformed into a structured text with clinical terms and specific disease-centered classifications. In IBD, multiple sources of medical data (notes, endoscopy, imaging) contain abundant unstructured information and such tools showed high accuracy in the identification and structuring of these texts. The need for natural language processing of IBD patient data was previously highlighted in Frontline Gastroenterology’s review on AI in IBD (Frontline Gastroenterology, 2022). Finally, each clinic comes down to patient recommendations and these are derived with the assistance of multiple LLM based decision support tools that provide recommendations based on the summary of the structured visit in the context of the relevant guidelines. Such tools can be extremely useful across multiple fields of GI, often where guidelines are inconsistent or complex and where clinicians must integrate multiple data sources with the current patient’s clinical data. Even common clinic complaints such as functional dyspepsia require integrating multiple symptoms and clinical data for optimal work-up and treatment, as described in a recent Frontline Gastroenterology guidance (Frontline Gastroenterology, 2025). Care is extended as the patients leaves with a link to chatbots that can answer questions specific to his condition.

Afternoon in the Endoscopy Suite

On to the endoscopy suite. Your list allows open access endoscopy received from multiple referral sites. These referrals should be evaluated to decide on appropriate settings, medication management, and type of preparation. At your location the referrals are managed by an LLM based tool which already analyzed the referrals and identified the requested procedures, the indications, pre procedural medication management (anti-coagulants, GLP-1 receptor agonists, etc.) and type of bowel preparation. Patients who come in for a procedure are well informed since a chatbot already addressed multiple concerns and questions they had regarding their upcoming colonoscopy. Just like at the clinic, during colonoscopy, you narrate findings, while an AI tool transcribes the report with high accuracy, and as clinic notes these reports will include LLM generated, guidelines-based recommendations. The patient will get an automatically generated concise letter explaining the findings and forward recommendations. Of course, the entire endoscopy is AI powered with various pathology detection and diagnosis tools, but this is a subject for a different post.

Prior to heading home, you can deidentify your clinic notes and endoscopy reports from the day to a research AI agent, which is provided by all large models to look up or summarize any patient specific evidence that was not covered in guidelines or large reviews and evaluate some additional possibilities for patients. Guideline preparation in gastroenterology was already shown to benefit from augmentation with LLM-powered systematic reviews (using deep research AI tools such as those provided by all the major LLMs).

If you are in a hurry, you can convert the findings to an audio podcast, using tools such as NoteBookLLM, to listen to on your way home.

In summary, AI and specifically LLMs are set to revolutionize and potentially improve every aspect of our daily work as gastroenterologists, and the possibilities keep expanding. Worth mentioning that LLMs can also enable us to perform research by providing statistical and epidemiological advice and implementation and empower us to create our own tools with models that can automatically convert your prompts to apps (see vibe coding).

Now briefly go over the daily description above and consider “is a human gastroenterologist really needed here and where?”, and if you are thinking “clearly in the endoscopy part” then robotics also has some surprises coming.