Using AI without losing the patient voice

Advocates Day at the World Bladder Cancer Patient Forum 2026 in Hong Kong made room for a subject every organisation in the room is already wrestling with. In a workshop titled “AI tools for patient advocates: a practical introduction” Theodoros Yfantis, WBCPC Project Coordinator, set out where artificial intelligence can genuinely help an advocacy team, and where it should be kept well away.

He began by taking the mystery out of it. A large language model is built from enormous amounts of text and tuned to recognise patterns in language, so its output sounds human because it mimics human writing, not because it understands anything. It does not know what bladder cancer feels like, what a patient needs, or whether its own answer is true. Three principles followed: amplify, do not replace, because expertise, judgement and the patient voice stay with the team; treat it like an employee, which means giving it a clear brief, checking its work and publishing nothing unreviewed; and remember that it is the smartest baby in the room, with vast knowledge and no context whatsoever. The rule that comes out of all three is simple. The quality of the output is the quality of the prompt.

The workflows he showed are the ones that eat a small team’s week. Turning meeting notes into a board summary, a press release into a LinkedIn post, a patient story into a video script, or a Health Technology Assessment (HTA) report into a one-page brief. Making complex text simple, from a scientific paper into plain language, or medical terminology into something patient-friendly, and pulling action items out of a long transcript. Adjusting tone, warmer for a patient newsletter, more formal for a policymaker briefing. And proofreading, including checking British English and date formats and flagging terminology used inconsistently.

He was just as clear about the limits. Never trust it blindly, because it sounds confident even when it is wrong, and because generic “AI voice” has to be caught and replaced with our own: warm, patient-centred, British English. Never trust the numbers, because models invent statistics and attach citations that do not exist. Verified figures go in from sources such as GLOBOCAN and our member organisations, and if a source does not check out, the number does not go in. Never share personal or sensitive information, because anything typed in may be stored or used for training, which rules out patient names, medical records and any uncleared member data. And do not let it do the thinking for you.

The session’s practical tips, for advocates taking this back to their own organisations:

• Write a brief, not a nudge. State the audience, the purpose, the format, the length and the tone every time, and supply the facts yourself rather than hoping the model has them.
• Never let it produce a number. Paste in verified figures from GLOBOCAN or your own organisation, and if a citation does not check out, the statistic does not go in the draft.
• Anonymise before you paste. Treat every prompt as if a stranger could read it, which means no patient names, no medical records and no uncleared member data.
• Edit out the “AI voice.” A first draft usually reads generic, so replace it with your organisation’s own voice, warm, patient-centred and in British English.
• Start with the admin, not the advocacy. Summaries, translations, transcripts and formatting are where the hours come back; strategy, judgement and the patient voice stay with the team.
• Work in small pieces and check the final version. Iterating slide by slide or paragraph by paragraph is far easier to review, and nothing should go out unread.

Theodoros also walked through the practical side, from choosing a faster everyday model over a more powerful one for complex work, to using AI inside Word, Excel and PowerPoint for formulas, data clean-up, speaker notes and alt text.

The closing slide put it plainly: garbage in, garbage out. A model does not think, can be confidently wrong, knows nothing about your audience or your organisation, and is not a confidant. Used with a clear brief and a careful review, it hands a small team back some hours. What it cannot hand back is the thing our advocacy rests on, the lived experience of patients, survivors and carers, and that stays with the people in the room.

Follow the rest of our Forum coverage here and at #BladderCancerForum26 on social media.