The customer pays for institutional prose
In a study with 192 adults, researchers reduced one clinical-trial consent document from Flesch–Kincaid grade 12.3 to 8.2. Comprehension improved with an effect size of 0.68; 52.6 percent of participants answered more questions correctly, 32.6 percent were unchanged, and 15.1 percent did worse. Plain language created real value, not uniform value.
Plain language is sometimes opposed as a loss of precision. The real choice is rarely precision versus simplicity. It is whether the answer preserves the distinctions that change a decision while removing syntax, jargon, repetition, and structure that do not. An understandable explanation can be more precise in use because people can correctly identify what applies to them.
The evidence supports simplification, with an important warning
Research on simplified informed consent found that plain language and simpler syntax can reduce cognitive reading burden across adults with differing reading skills and working-memory capacity. W3C guidance similarly recommends familiar words, short sentences, clear blocks, and supportive visual layout. These are universal-design choices: they help people with specific needs and often make the experience better for everyone.
The minority who did worse is the warning. Readability is a population-level proxy, not proof that one person understood the conditions, risks, and next action. Simplification is a transformation under constraints, not free summarization: the answer must become easier to use without silently changing what the institution is obliged to communicate.
Readability is not understanding
Shorter words and sentences can improve a readability formula while leaving the decision model opaque. A customer may understand every sentence but not know which condition applies, what happens next, or which exception overrides the default. Conversely, a necessary technical term can be retained and explained if it anchors an important distinction.
Define comprehension operationally. Can the reader identify the recommendation, reasons, risks, alternatives, and next step? Can they apply the rule to a new case? Can they detect when the answer no longer applies? These tasks turn “clear” from a stylistic preference into a measurable performance property.
Build a protected-content map
Before asking an LLM to simplify, classify the source. Mark facts that may be paraphrased freely, terms that require an approved definition, warnings that must remain prominent, conditional rules that must preserve logic, and passages requiring human escalation. This creates a content protocol against which the transformed answer can be checked.
Then design layers. A short action summary serves the immediate task. Expandable detail supports scrutiny. Definitions sit beside unfamiliar terms. A visual sequence shows the process. The complete legal or clinical text remains accessible. Progressive disclosure is not concealment when the hierarchy is honest and critical information is never pushed behind curiosity-dependent interaction.
Test understanding, not the reading score
Choose a consequential letter, consent explanation, policy answer, or eligibility decision. Compare the current text with a plain-language version and a layered version that adds structure and visual cues. Use expert review to confirm semantic and protocol completeness before exposing participants. Measure decision comprehension, application to a near-transfer case, time, support-seeking, and false reassurance.
Add one “dangerous simplification” probe: a case where a missing exception would change the right action. If a version improves average comprehension but fails this probe, it is not ready. The commercial value comes from reducing burden without transferring hidden risk to the customer or the institution. In regulated businesses, that combination is the product.