Truace
Subscribe
48/100

Emerging evidence

Cite
More
Export CSVExport JSONEmailLinkedInX

Simple prompt change made GPT-5 breast imaging answers easier to read, study finds

Study title: Prompt-dependent readability and accuracy of Generative Pre-trained Transformer-5 thinking responses to breast imaging questions derived from European Society of Breast Imaging patient recommendations

Evidence type
Benchmark or lab evaluation
Sample
n = 96
Sources
1 sourcePeer-reviewed
Published
2 min read
Truvace record
TRV-2026-2868cite or check it

Study title: Prompt-dependent readability and accuracy of Generative Pre-trained Transformer-5 thinking responses to breast imaging questions derived from European Society of Breast Imaging patient recommendations

Researchers asked GPT-5 Thinking 48 questions drawn from European Society of Breast Imaging patient materials, 15 FAQs and 33 special information note questions, under a standard prompt and a patient-focused prompt, producing 96 responses. The patient-focused prompt gave significantly lower reading levels on every readability index (P < 0.001).

Read the study abstract

Purpose: Although professional societies provide standardized resources, the reliability of online health content remains inconsistent, particularly as large language models (LLMs) become a growing source of patient information. Therefore, the primary objective of this study was to determine whether patient-focused prompt optimization improves the readability of Generative Pre-trained Transformer-5 (GPT-5) Thinking responses compared with standard prompting while evaluating potential changes in expert-rated fact…

The gain

In a study of 48 European Society of Breast Imaging questions, a patient-focused prompt made GPT-5 Thinking answers significantly easier to read on all four readability indices (P < 0.001), while radiologist-rated accuracy stayed at 4.52 versus 4.54 out of 5.

In brief

Answer length did not explain the change: word count (P = 0.12) and sentence count (P = 0.63) did not differ significantly between prompts. Two radiologists scored accuracy on a 5-point scale, giving mean scores of 4.54 for the standard prompt and 4.52 for the patient-focused one (P = 0.75).

The authors call it a simple, low-cost prompt modification for patient education in breast imaging. Even with the change, the reading grade level stayed above widely recommended patient-education thresholds, and it is not yet known whether easier text improves patient comprehension.

Main points

  1. The study covered 48 questions: 15 FAQs and 33 questions derived from special information notes, all from European Society of Breast Imaging educational materials.
  2. Each question was answered under two prompts, giving 96 GPT-5 Thinking responses, all generated prospectively in January 2026.
  3. Readability was measured with four indices, averaged into a composite reading grade level.
  4. Patient-focused prompting lowered reading level across all indices (P < 0.001).
  5. Word count (P = 0.12) and sentence count (P = 0.63) did not differ significantly between the two prompts.
  6. Two radiologists rated accuracy on a 5-point Likert scale: 4.54 ± 0.63 versus 4.52 ± 0.59 (P = 0.75).
  7. Inter-rater agreement was almost perfect, with κ = 0.82 for FAQs and 0.81 for special information note questions.

The full story

The study tested whether a patient-focused prompt could make GPT-5 Thinking responses easier to read than a standard prompt, using European Society of Breast Imaging materials as the source of questions. All outputs were generated prospectively in January 2026.

Readability was scored with four indices combined into an averaged composite reading grade level. Two radiologists scored accuracy on a 5-point Likert scale, and their agreement was almost perfect (κ = 0.82 for FAQs, 0.81 for special information note questions).

The patient-focused prompt significantly lowered reading level on all indices (P < 0.001) without significant changes in word or sentence count. Mean accuracy was 4.54 ± 0.63 under the standard prompt and 4.52 ± 0.59 under the patient-focused prompt (P = 0.75).

The authors conclude the approach produces more accessible breast imaging content but note the resulting grade level remained above recommended patient-education thresholds. They say further work is needed to confirm that it improves patient comprehension.

Sources

1 on file
  1. Diagnostic and Interventional Radiologydoi:10.4274/dir.2026.264309
    Peer-reviewed2026-11-12

Talk it over

Three questions to start on, two more if you have timeThe patient-focused prompt lowered reading level on all four readability indices (P < 0.001), yet grade level stayed above recommended patient-education thresholds. How would you weigh that improvement against the gap that remains?

Start with a question

Simple prompt change made GPT-5 breast imaging answers easier to read, study findshttps://truvace.com/item/prompt-dependent-readability-and-accuracy-of-generative-pre-trained-transformer-

  1. The patient-focused prompt lowered reading level on all four readability indices (P < 0.001), yet grade level stayed above recommended patient-education thresholds. How would you weigh that improvement against the gap that remains?

    Answer in the thread

  2. Two radiologists rated accuracy at 4.54 versus 4.52 out of 5 on 48 questions from European Society of Breast Imaging materials. What can that check tell us about whether easier answers stay correct for real patients?

    Answer in the thread

  3. Which result would convince you that easier text actually helps patients understand breast imaging, given the study did not test comprehension?

    Answer in the thread
If you have more time

  • The 48 questions came from European Society of Breast Imaging patient materials. Which patients, languages or kinds of questions might be missing, and how could that change the readability results?

    Answer in the thread

  • If a breast imaging clinic wanted to use the patient-focused prompt, who should review the answers before patients see them, and what would they check?

    Answer in the thread

What readers say

More from Health