The GMC has not yet published Portfolio Pathway-specific rules on generative AI, but the Faculty of Intensive Care Medicine and the Royal College of Paediatrics and Child Health have. Both draw the same line: AI tools can reasonably help with structure, grammar and organising a first draft, but the clinical detail, the judgement and the learning point in a reflection, audit or structured report must be genuinely yours. Never paste patient-identifiable information into a consumer AI tool. If your own college has not published guidance, follow the GMC's general framework, that you remain fully responsible for anything you submit, and apply the same structure-yes, substance-no principle FICM and RCPCH have already set out.
Why this became a Portfolio Pathway question so quickly
Two years ago, an article like this would have been a curiosity. It is not one now. GMC-commissioned research published in late 2024 found that 29% of doctors surveyed had used at least one AI system in their work in the previous twelve months, with 52% describing themselves as optimistic about AI's role in healthcare, and the doctors already using it tended to view it more favourably still. Portfolio Pathway candidates, who are by definition senior, time-poor, and writing under real pressure to produce structured reflections, audit summaries and CPD records on top of a full clinical job, are exactly the population most likely to reach for a tool that promises to save an hour on a Sunday evening.
The problem is that the Portfolio Pathway is not a generic writing exercise. It is an assessment of you: your four GMC domains, your Capabilities in Practice, and specifically, under domain four, your probity, meaning your honesty about what is actually your own work. Reflective writing, structured reports, audit and significant event analyses only count as evidence if the judgement inside them is genuinely the applicant's. That is precisely the kind of content a generative AI tool is good at producing convincingly, and precisely the kind of content where convincing is not the same as true.
Doctors in that same GMC research described real uncertainty about their own professional responsibilities when using AI, and said they had not received enough training on the risks, which included bias, lack of transparency, and the danger of over-reliance. That uncertainty is the honest starting point for this article. Nobody, including the GMC, has finished working this out. What follows is what has actually been published so far, not a guess at where the rules will eventually land.
What FICM has said about generative AI in ICM training and revalidation
The Faculty of Intensive Care Medicine has moved further and faster on this than any other UK medical body. FICM has published two separate pieces of material worth knowing about if you are working toward the Intensive Care Medicine Portfolio Pathway, or reading across for principles that apply more widely.
The first is interim guidance addressing generative AI tools by name, including ChatGPT, Gemini and Claude, and their use by intensivists specifically for training portfolio work, Portfolio Pathway submissions (still commonly referred to by ICM candidates as CESR, the route's former name) and revalidation. FICM frames this explicitly as interim: a holding position issued while the GMC continues its own broader review of generative AI's role in medical education and assessment, not a final settled rule. The guidance draws a distinction that recurs throughout every college position published so far, that AI assistance with structuring a document, checking language, or helping organise a first draft sits differently to AI assistance with the substance, the clinical reasoning or the personal reflection, which it does not sanction.
The second is FICM's broader position statement on medical AI, which is not portfolio-specific but sets the wider professional context. It sets out that for patients to benefit from medical AI, models need to demonstrate safety and efficacy through rigorous evaluation, and it is notably firm on large language models used for medical purposes in intensive care: FICM's position is that LLM use for clinical purposes in ICM should happen only as part of a medical device with formal regulatory clearance, or within ethically approved research. A generic consumer chatbot used to draft a reflection or a structured report sits outside both of those categories, which is exactly why FICM treats portfolio and revalidation use as a separate, narrower question from clinical decision-support use.
ICM has a long history of taking AI and technology governance seriously in the clinical setting, given how much monitoring, prediction and decision-support technology already sits in a modern intensive care unit. FICM's willingness to also publish training and revalidation-specific guidance early is consistent with that culture, not a one-off. Candidates in other specialties should not assume their own college is far behind simply because nothing has been published yet; check directly, since these documents can appear with little advance notice.
What RCPCH has said about AI in ePortfolio entries
The Royal College of Paediatrics and Child Health is the other body with published, specific material, in its case covering AI use in ePortfolio entries and assessments more generally, rather than the Portfolio Pathway alone. RCPCH's guidance, issued in August 2025, covers the responsible use of AI in ePortfolio entries and a related resource on AI in ePortfolio and assessments specifically. Although it is written primarily for paediatric trainees rather than Portfolio Pathway candidates, the principles are written broadly enough, and the underlying assessment logic is similar enough, that they are useful reading for any specialty weighing up the same question.
RCPCH's position mirrors FICM's core distinction closely: AI can help structure an entry, tidy language, or prompt reflection through general questions, but it cannot generate the substance of what should be a genuine account of the trainee's own experience and learning. The guidance is also explicit that entries must remain an honest, accurate record, which is the same probity principle domain four of the Portfolio Pathway is built on, described in a different framework but pointing at the same underlying expectation.
What is notable, reading FICM's and RCPCH's material side by side, is how much agreement there already is between two colleges that assess entirely different specialties, working independently of each other and without a shared GMC template to follow. That convergence is a reasonable signal of where the wider professional consensus is heading, even in specialties where nothing has been formally published yet.
Where the GMC currently stands
The GMC's own position is broader and, for Portfolio Pathway purposes, less specific than either FICM's or RCPCH's. Good Medical Practice 2024, which came into effect for doctors on 30 January 2024, references software, diagnostic tools and digital technologies, including those using AI, mainly in the context of a doctor's responsibility to report adverse incidents involving medical devices. The GMC's separate learning material on artificial intelligence and innovative technologies sets out a high-level, technology-neutral framework rather than a Portfolio Pathway-specific rulebook: doctors, physician associates and anaesthesia associates are expected to apply their existing professional judgement and the principles already in their professional standards to any AI tool they use, and they remain personally responsible for the decisions they take, whatever informed those decisions.
In parallel, the GMC has published its own qualitative research into doctors' experiences of using AI, based on in-depth interviews with doctors who had used AI tools in the previous year, as a follow-up to earlier survey work. The research is descriptive rather than regulatory, but its recommendations, more support on the ethical considerations around AI and more structured education and training, tell you plainly that the GMC itself regards the current guidance picture as unfinished business, not a settled state.
As of this article's publication, the GMC has not issued a rule specific to generative AI use in Portfolio Pathway evidence. The general principle, that you remain responsible for anything you submit, already applies and is not going to soften. Do not wait for a more detailed rule before applying it.
The college guidance landscape, at a glance
Most Portfolio Pathway candidates are not applying through ICM or paediatrics, so the honest picture for most readers is that their own Royal College or Faculty has not yet published anything specific. The table below sets out where things stand across the bodies most relevant to this site's audience, as of July 2026. Check the primary source directly before relying on this, since this is one of the fastest-moving areas of medical regulation right now and a college could publish new material at any time.
| Body | Specific AI guidance published? | Core position |
|---|---|---|
| GMC | General framework only, no Portfolio Pathway-specific rule | Doctors apply existing professional judgement and remain fully responsible for anything they submit, whatever tool assisted them |
| FICM (Intensive Care Medicine) | Yes, interim guidance plus a separate position statement on medical AI | AI may help with structure and language; it must not generate clinical substance or judgement; interim while the GMC review continues |
| RCPCH (Paediatrics) | Yes, two related guidance documents on AI in ePortfolio entries and assessments | Same structure-yes, substance-no distinction; entries must remain an honest, accurate record of the trainee's own experience |
| RCoA / Anaesthetics, RCP / medical specialties, RCR / Radiology, RCPath, other Faculties | Not yet published, at time of writing | Default to the GMC's general framework and the FICM / RCPCH principles until specific guidance appears |
This table will look dated quickly, in the best possible way. Colleges are actively working on this, and RCPCH's own guidance appeared with relatively little warning in August 2025. If your specialty is not yet listed with published guidance, check your college or Faculty's website directly before assuming the gap means anything goes.
What generative AI can reasonably help with, and what it cannot
Pulling FICM and RCPCH's positions together, and reading them against the GMC's general framework, produces a reasonably clear practical line, even without a single unified rulebook. The comparison below sets out where that line currently sits.
Reasonable to use AI for
Not sanctioned by any published guidance
The pattern is consistent enough across FICM, RCPCH and the GMC's general principles that it is a reasonable working rule for any specialty, including the many where no college has published anything yet. Use AI the way you would use a competent editor: valuable for making your own material clearer, never a substitute for having done and thought about the work yourself.
The three real risks for your Portfolio Pathway evidence
Three specific risks come up repeatedly once you look past the general principle, and each maps onto a different part of the Portfolio Pathway.
Probity risk: a reflection that is not genuinely yours
Domain four assesses honesty about your own work. If the substance of a reflective entry, not just its wording, was generated rather than lived and thought through, it fails the basic test of what reflective evidence is for, regardless of how well it reads. This is the single biggest risk, and the one every college guidance published so far addresses directly.
Accuracy risk: hallucinated detail in audit or CPD entries
Generative AI tools are well documented to produce fabricated references, invented statistics, and plausible-sounding but false detail, a pattern the wider medical and academic literature calls hallucination. An audit or CPD entry containing a fabricated figure or a misremembered guideline reference, generated confidently by a tool and not checked, is a factual accuracy problem on top of the probity one.
Confidentiality risk: patient detail in a consumer tool
This is not a new AI-specific rule; it follows directly from existing GMC confidentiality guidance. Most consumer generative AI tools are not covered by an NHS data processing agreement and are not approved for handling patient-identifiable information. Typing real case details, even partially anonymised ones, into a public AI tool while drafting a significant event analysis or structured report is a confidentiality breach in its own right, entirely separate from what any Portfolio Pathway assessor thinks of the resulting text.
Before you use an AI-assisted sentence in a portfolio entry, ask: could I defend this sentence, unprompted, in a conversation with an assessor, using my own memory of the case? If the honest answer is no, the entry has drifted from evidence about you into content about a case that never fully happened. Rewrite it from what you actually remember.
A practical checklist before you use AI on an entry
The checklist below turns FICM's and RCPCH's published principles, and the GMC's general confidentiality and probity framework, into something usable at your desk on the evening you are actually writing an entry.
Before you use AI on any portfolio entry
Run through this in order, every time, not just the first time
What to do if your own college has not published anything yet
Most readers of this article are in exactly this position, since only two bodies have published specific material so far. The sequence below sets out a sensible way to proceed in the absence of a specific rule, without waiting indefinitely for one to appear.
A sensible sequence when your college has been silent so far
Before you use AI in your portfolio: a one-page checklist
A printable version of the checklist above, so it is on your desk, not buried in a browser tab, the next time you sit down to write an entry.
Should you disclose that you used AI at all?
Neither FICM nor RCPCH's published guidance requires a formal AI-use disclosure statement on individual Portfolio Pathway entries, and the GMC has not introduced one either, as of this article's publication. That is a different position to academic publishing, where journals including the BMJ have taken an increasingly firm line on AI-generated content, requiring human authorship and full accountability, and warning that unverified AI-generated claims put the credibility of a piece of work at risk. Portfolio Pathway entries are not academic manuscripts, but the underlying caution is the same: a tool assisting with language is a different thing to a tool authoring the substance, and the second is where trouble starts.
The practical answer is that formal disclosure is not currently required, but honesty if asked is not optional. If a Royal College introduces a disclosure requirement, and given how quickly this area is moving that is plausible, comply immediately and update how you have been working. Until then, the safest position is to use AI only in the ways set out above, and to be able to say plainly, if a supervisor, referee or assessor ever asks, exactly what a tool did and did not contribute to a given entry.
The honest summary of this entire topic is that the profession has not finished deciding the rules. FICM and RCPCH have moved first; the GMC's own review is ongoing; most colleges have said nothing yet. Treat everything in this article as the current best reading of a fast-moving area, and check primary sources directly before relying on any of it for a live application.
Where this sits in the practical cluster
This article opens a new cluster of cross-cutting practical guides that sit alongside, rather than inside, any single tier of evidence or specialism content. The remaining pieces will cover the day-to-day mechanics of actually building and submitting a portfolio, from the systems you use to how you keep going when the day job gets in the way.
For the foundational context this article assumes, our guides to the Portfolio Pathway explained, the four GMC domains, and the Specialty Specific Guidance are the right starting points. If you are working specifically toward Intensive Care Medicine or Anaesthetics, our Intensive Care Medicine and Anaesthetics overviews cover FICM and RCoA's assessment structures in full. Once you are ready to compile and submit, our walkthroughs of the GMC Online application and what happens after submission take you through the remaining stages, and our guide to the structured CV format covers a document every applicant has to produce, AI-assisted or not.