Can Generative AI Check the Readability of Consent Materials?

Can ChatGPT Tell You Whether Participants Will Understand Your Consent Form?

You have prepared an informed consent form for a study involving university instructors. The document explains the research purpose, interview procedures, confidentiality safeguards, data retention, and participants' rights.

Before submitting it for ethics review, you ask ChatGPT to evaluate its readability.

The system reports that the document is suitable for an eighth-grade reading level. It identifies several lengthy sentences and recommends replacing technical terms with simpler expressions.

The feedback appears useful, but you wonder how the model calculated the reading level. Did it apply an established readability formula, or did it simply estimate difficulty from the wording?

More importantly, even if the document is easy to read, does that mean participants will understand concepts such as voluntary participation, pseudonymization, withdrawal limitations, or external AI processing?

Can generative AI meaningfully evaluate informed consent readability, and what evidence is needed before researchers can claim that their materials are understandable?

AI Can Review Readability, but It Cannot Establish Participant Comprehension

Using AI to Evaluate Consent Readability Without Overstating What It Measures

What Is Readability?

Readability concerns how easily a written text can be read, often assessed through features such as sentence length, vocabulary, word length, and grammatical complexity.

For example, a consent form containing lengthy sentences and unfamiliar terminology may require more reading effort than one using shorter sentences and clear explanations.

However, readability is not a single universally defined property. Different formulas and evaluation approaches emphasize different textual characteristics.

A document's readability may also depend on the language, intended population, and subject matter.

Researchers should therefore specify what they mean by readability rather than assume that one numerical score captures every aspect of textual accessibility.

Readability Is Not the Same as Comprehension

One of the most important distinctions concerns the difference between textual difficulty and actual understanding.

For example, the sentence "Your data will be pseudonymized" is short, but a participant unfamiliar with the term may not understand what will happen to their information.

A longer explanation stating that names will be replaced with codes and the identifying information stored separately may be more understandable despite containing more words.

Researchers should not assume that a low reading-grade score establishes meaningful informed consent.

What Can Generative AI Identify?

Generative AI may assist with identifying language features that could make consent materials difficult to read.

These observations may support revision, but researchers should verify that suggested changes preserve the intended meaning.

What Are Established Readability Formulas?

Readability formulas estimate textual difficulty using specified linguistic features.

They provide structured measures, but their results should be interpreted according to the formula's assumptions and limitations.

These formulas were developed for particular linguistic contexts and should not be applied uncritically to every language or participant population.

They also do not directly measure whether participants understand research procedures or consent conditions.

Can ChatGPT Calculate a Flesch-Kincaid Score?

A general-purpose language model may describe readability formulas and estimate the difficulty of a passage.

However, when asked to provide a numerical score without a verified calculation, the model may generate an inaccurate value.

For example, it may state that a document has a Flesch-Kincaid Grade Level of 8.2 without correctly counting sentences, words, or syllables.

Researchers should distinguish a qualitative model judgment from a score computed using the actual formula.

If a numerical readability score is required, use a suitable readability calculation tool or verified implementation and document the method used.

AI can help interpret the resulting score, but it should not be assumed to have performed the calculation merely because it provides a plausible number.

How Does Flesch Reading Ease Work?

Flesch Reading Ease estimates English text difficulty using average sentence length and average syllables per word.

The formula illustrates why shorter sentences and words with fewer syllables may improve the score.

However, replacing an important technical term with an inaccurate shorter word may increase readability while reducing the quality of informed consent.

Researchers should therefore avoid optimizing the score at the expense of substantive meaning.

How Does Flesch-Kincaid Grade Level Work?

The Flesch-Kincaid Grade Level formula uses similar features but expresses the result as an estimated United States school-grade level.

This does not mean every eighth-grade student will understand the document or that adults with different educational backgrounds will interpret it identically.

The score should be treated as an estimate of textual difficulty rather than a direct measure of comprehension.

What Reading Grade Level Should a Consent Form Target?

Some institutional guidance recommends plain-language consent materials and may suggest particular reading-grade targets.

However, there is no universally appropriate grade level for every research study or participant population.

For example, a study involving university instructors may include technical educational terminology that is familiar to participants, while a study involving younger students may require different explanations.

Researchers should follow applicable institutional requirements and consider the literacy, language proficiency, and familiarity of the intended population.

Even where a target grade level is specified, the document should still be evaluated for conceptual clarity and accuracy.

A grade-level target should guide revision rather than become the sole criterion for consent quality.

Why Shorter Sentences Are Not Always Better

Shorter sentences may reduce reading difficulty, but excessive simplification can fragment important relationships among ideas.

For example, a confidentiality statement may include a condition explaining when certain information could be disclosed.

Separating the condition from the main statement without preserving their relationship may create a misleading impression.

Researchers should ensure that sentence revisions preserve the logical connections among procedures, exceptions, and participant rights.

The goal is to reduce unnecessary complexity while retaining the meaning required for informed participation.

Can Readability Scores Detect Misleading Consent Statements?

No. Readability formulas do not establish factual accuracy.

For example, the sentence "Your data will always be safe" is short and easy to read.

However, it may be an unjustified guarantee if the study cannot establish absolute security.

Similarly, "You may withdraw your data at any time" may be readable but inaccurate when withdrawal is subject to specified limitations.

Researchers should separately evaluate whether the document accurately describes the approved protocol.

The issue of simplifying informed consent language with AI therefore requires both readability review and substantive verification.

What About Consent Materials Written in Filipino or Other Languages?

Many established readability formulas were developed for English and rely on language-specific features such as syllable patterns and sentence structure.

Applying an English readability formula directly to Filipino or another language may produce results that are not appropriately interpretable.

Researchers should use measures and evaluation approaches suitable for the language and intended population.

Generative AI may identify potentially difficult expressions in multilingual materials, but its judgments should be verified by qualified language users and relevant subject-matter reviewers.

Researchers should also examine whether translation or simplification changes the meaning of consent conditions.

A fluent translation is not necessarily an accurate or comprehensible explanation of research rights and procedures.

Can AI Evaluate Conceptual Difficulty?

Generative AI may help identify concepts that require additional explanation, even when the surrounding sentences are grammatically simple.

For example, it may flag the terms pseudonymization, secondary data use, or automated profiling as concepts that prospective participants may not understand.

However, AI cannot establish actual comprehension merely by predicting which terms appear difficult.

The model's assessment reflects patterns in language and the information supplied in the prompt, not direct evidence of how the intended participants interpret the document.

Researchers should treat these observations as hypotheses for further review rather than findings about participant understanding.

How Should You Prompt AI to Review Consent Readability?

A useful prompt asks for specific language problems and distinguishes qualitative feedback from numerical readability calculations.

The researcher should review the suggestions and verify that revisions do not change the approved information.

Should Researchers Combine AI With Readability Software?

Combining qualitative AI feedback with an established readability calculation may provide a more informative review than relying on either approach alone.

For example, readability software may identify a high estimated reading-grade level, while AI may highlight particular sentences or unexplained terms contributing to difficulty.

The researcher can then revise the document and recalculate the score using the same method.

However, a lower score does not necessarily mean the revision is substantively better.

Researchers should verify the revised content and consider whether the explanation remains suitable for the intended population.

Readability scores are most useful as indicators for review rather than definitive judgments of consent quality.

How Can Researchers Evaluate Actual Comprehension?

Comprehension may be assessed through approaches that ask participants or representatives of the intended population to explain what they understand from the materials.

For example, a researcher might ask someone to describe in their own words whether participation is voluntary, what information will be collected, and when withdrawal is possible.

Such approaches may reveal misunderstandings that readability formulas cannot detect.

However, comprehension assessment should be designed appropriately for the research context and conducted under applicable institutional procedures.

Responses from a small group should not automatically be generalized to all prospective participants.

The objective is to obtain evidence about understanding rather than assume that grammatical simplicity guarantees it.

Can AI Simulate Participant Comprehension?

Researchers may ask AI to respond as though it were a prospective participant reading the consent form.

Such simulations may generate useful questions or identify ambiguities for further examination.

However, simulated responses do not constitute empirical evidence of actual participant comprehension.

For example, a model instructed to behave like a university instructor may produce a plausible explanation of the consent form, but this does not establish how real instructors will understand it.

Researchers should distinguish simulated feedback from evidence collected from the intended population.

AI can help prepare comprehension questions, but it should not replace appropriate participant-based evaluation when that evidence is needed.

What If the AI and Readability Formula Disagree?

Different evaluation methods may emphasize different aspects of a document.

For example, a readability formula may assign a relatively low grade level to a paragraph containing short sentences. AI may nevertheless identify unfamiliar terminology or unclear conceptual relationships.

These observations are not necessarily contradictory.

The formula evaluates specified textual features, while the AI provides a qualitative interpretation of possible difficulties.

Researchers should examine the underlying evidence rather than choose whichever result appears more favorable.

Where actual participant comprehension is important, appropriately designed feedback or comprehension assessment may provide additional information.

Does a Favorable Readability Score Mean Ethics Approval Is More Likely?

Clear consent materials may support the quality of an ethics application, but readability alone does not determine ethical acceptability.

Reviewers may also consider whether the document accurately explains procedures, risks, benefits, confidentiality, participant rights, and applicable safeguards.

A consent form with a favorable readability score may still omit material information or contain inaccurate statements.

Researchers should therefore avoid presenting a numerical score as proof that the document satisfies all informed consent requirements.

Institutional review remains necessary where required, regardless of the readability assessment.

Using AI and Readability Measures to Improve a Consent Statement

The revised sentence is easier to read, but the researcher must still ensure that participants understand the relevant confidentiality procedure.

The example illustrates why readability improvement should be evaluated alongside accuracy and comprehension rather than through a numerical score alone.

Common Misconceptions About AI and Consent Readability

Use AI to Identify Difficult Language, Then Verify Readability and Meaning

When reviewing consent materials, ask AI to identify specific wording problems rather than rely on an unsupported overall reading-level judgment.

Where numerical readability scores are useful or required, calculate them through an appropriate verified method. Then examine whether the document remains accurate and understandable for the intended population.

Document the evaluation method and review any substantive changes against the approved protocol.

Readability assessment is useful when it helps researchers improve communication. It becomes misleading when a numerical score is treated as evidence that participants understand everything they are being asked to agree to.

Before Accepting an AI Readability Assessment of Consent Materials

Frequently Asked Questions About AI Consent Readability Assessment

A Consent Form Can Be Easy to Read Without Being Easy to Understand

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.