In January 2026 the RCGP published guidance on using generative AI in appraisal preparation, and its position can be stated in two sentences. AI may help you prepare, organise and improve your reflective entries. It should not write your reflections for you, and any AI use should be declared to your appraiser. Around those two sentences sits a set of distinctions worth understanding precisely, because they separate genuinely useful workflows from ones that quietly defeat the purpose of the exercise, and because vague anxiety about the rules is pushing some GPs to avoid legitimate help while others drift past the line without noticing.
The College's underlying logic
The RCGP's position on AI and reflection, running through both its appraisal guidance and its parallel guidance on generative AI in GP training, rests on one principle: AI can complement, not replace, the development of genuine reflective skill. Reflection is not paperwork about learning; it is the learning mechanism itself, the step where an experience becomes a change in how you practise. Outsource the writing and you have outsourced the thinking, which leaves the portfolio full and the doctor unchanged. Hence the College's consistent framing: prompts, structure and refinement are welcome; substitution is not. Overreliance, in the training guidance's phrase, risks undermining the very purpose of reflective practice.
What is acceptable
The permitted territory is broader than many GPs assume. Using AI to generate reflective prompts and questions, what did this case challenge in my usual approach, which curriculum areas does it touch, is explicitly the kind of scaffolding the guidance envisages. Using it to organise your own rough notes into a clearer structure, to tighten grammar and expression, or to summarise your own written material back to you, all fine: the substance remains yours. Using it to identify missing elements, you have described the event and the learning but not the change in practice, is arguably where AI helps most, functioning as a checklist with judgement. And using AI tools for the learning itself, guideline lookup, assessed questions, tutoring dialogue, is simply CPD, with the reflection then written by you about it.
What is not acceptable
Three practices fall clearly on the wrong side. Fabricated reflection: asking AI to write a reflective account of learning or events, or to invent the insight, produces text that is not evidence of anything except access to a chatbot. Generic generated entries: templated reflections with the topic swapped in fail the basic test that a reflection could only have been written by you; appraisers read hundreds of entries and pattern-match faster than authors expect. Undisclosed substitution: even genuinely AI-drafted text that you lightly edited becomes a problem when presented as unaided work, because the declaration principle exists precisely so the appraisal conversation stays honest. If you would be uncomfortable telling your appraiser how an entry was produced, that discomfort is the guidance talking.
Patient confidentiality is the hard boundary
Everything above concerns legitimacy; this concerns law and professional duty. No patient-identifiable information should be entered into a general-purpose AI tool when preparing reflections: no names, dates of birth, or combinations of details that could identify a person, and real caution with rare conditions and small communities where even sparse details narrow quickly. Reflect on anonymised substance, the clinical reasoning, the system issue, your response, rather than the identifying particulars, which appraisal never needed anyway. This boundary applies regardless of how good the tool is and regardless of what it promises about data handling.
How iatroX scaffolds without writing
This is the design line iatroX deliberately holds. The platform generates the raw material of reflection: the question you missed, the Tutor dialogue that surfaced why, the guideline answer with its citation, the re-test showing the correction held. My CPD assembles that record and prompts your reflection with the specifics of your own session. What it does not do is write the reflective sentences for you, because a reflection generated by the platform would be exactly the generic entry the College warns against, and because the four honest sentences you write about a diagnosed misconception are worth more at appraisal than four polished paragraphs written by anything else.
A transparent AI-use statement you can adapt
Declaring AI use need not be elaborate. A single adaptable statement, included in your portfolio or raised at the meeting, satisfies the principle:
"In preparing this appraisal I used AI tools as follows: [tool] to support clinical learning and identify knowledge gaps; [tool] to suggest reflective prompts and check my entries for completeness and clarity. All reflective content, judgements and descriptions of my practice are my own. No patient-identifiable information was entered into any AI tool."
Edit the bracketed sections to match reality, delete what does not apply, and the declaration is done.
Frequently asked questions
Does using AI make my reflection worth less at appraisal?
Used within the guidance, no; scaffolded reflection that remains yours is simply reflection. What devalues an entry is substituted thinking, with or without AI.
Do I need to declare using a grammar checker?
Proportionality applies: the guidance is aimed at generative tools shaping content. A blanket one-line declaration like the statement above covers the territory without itemising every spellcheck.
What about AI scribes and clinical documentation tools?
Different question, different governance: clinical AI tools run under your organisation's policies and data agreements. The appraisal guidance concerns your reflective writing; keep the two frameworks distinct.