Readability of National Surgical Consent Forms: Artificial Intelligence Generation Versus Simplification

Consent forms often exceed the reading ability of populations they serve. Large language models may improve accessibility of patient-facing information, but existing research has predominantly assessed de novo generation, with limited benefit. AI-assisted simplification of existing, clinically validated material represents a distinct workflow that remains underexplored. This study compared the readability of all 64 consent forms published by NHS England's Getting it right first time (GIRFT) with matched de novo AI-generated and AI-simplified versions across six surgical specialties.

For each procedure, two AI counterparts were produced using Claude Opus 4.8 with standardised prompts: a de novo AI-generated form and an AI-simplified version derived from the original (192 documents across three paired arms). Readability was assessed using five validated metrics, Flesch reading ease (FRE) and word count. Primary outcomes were average reading age (RA) and grade level (GL). Achievement of the nationally recommended RA (≤ 11 years) was also assessed.

Original GIRFT forms required a mean RA of 16.39 ± 0.85 years. AI-generated forms were harder to read (17.65 ± 0.73 years, p < 0.001), whereas AI-simplified forms were easier (13.27 ± 0.67 years, p < 0.001). No form achieved a RA ≤ 11 years. FRE ≥ 50 was achieved by 100% of AI-simplified, 45.3% of originals and 15.6% of AI-generated forms.

De novo AI-generation reduced readability relative to nationally standardised consent forms, whereas AI-simplification improved readability. Neither approach achieved the recommended reading age. AI may therefore have a role in improving the accessibility of existing consent forms, but outputs require objective readability assessment and clinical validation before implementation.