WCAG 3.1.5 — Reading Level

What this requires

When text requires reading ability more advanced than the lower secondary education level after removal of proper names and titles, supplemental content or a version that does not require reading ability more advanced than the lower secondary education level is available. "Lower secondary" maps roughly to grades 7-9 in the US, or ages 11-14.

How AI coding tools fail this

When asked to write a privacy policy, a help article, or a healthcare explanation, AI tools default to a register that is grammatically complex and lexically dense. Sentences are long, voice is passive, and the vocabulary is professional-document standard. The content passes legal review and fails reading-level review.

The second pattern: technical docs that explain a concept once at expert level, with no "what this means" plain-language gloss. Users who aren't already expert in the domain bounce off.

The third: a single content surface for a mixed audience. The same page tries to serve compliance officers and end users; both walk away frustrated.

Failing examples

A privacy policy written at college-graduate reading level:

<p>
  Notwithstanding the foregoing, we may, in our sole discretion,
  disclose your personal data to third-party service providers in
  furtherance of the legitimate business interests enumerated in
  Section 4 of this policy.
</p>

Passing examples

Write a plain-language version alongside the legal text:

<>
  <section aria-labelledby="plain-heading">
    <h2 id="plain-heading">In plain language</h2>
    <p>
      We share your personal information with companies that help us
      run our service — for example, the company that hosts our
      website, or the company that sends our email. We list these
      companies in section 4.
    </p>
  </section>
  <section aria-labelledby="legal-heading">
    <h2 id="legal-heading">Legal text</h2>
    <p>
      Notwithstanding the foregoing, we may, in our sole discretion,
      disclose your personal data to third-party service providers
      in furtherance of the legitimate business interests enumerated
      in Section 4 of this policy.
    </p>
  </section>
</>

Edge cases

Reading-level scoring (Flesch-Kincaid, SMOG, automated readability index) approximates the criterion. None of them perfectly predict comprehension; they're heuristics.
Domain-specific content can keep technical language as long as a plain-language equivalent is offered.
Legal and regulatory text is often required to be precise — the remedy is a parallel plain-language version, not rewriting the legal text.
Translation doesn't satisfy 3.1.5. Reading level is per language, not per phrase.
"Add a simple version" can be a single paragraph at the top of a technical page. It doesn't need to be a separate page.

How Jeikin handles this

The scanner runs a readability metric across the page text and surfaces pages above the lower-secondary threshold. The dashboard records the score and asks the reviewer to confirm whether a plain- language version exists on or near the page. The metric is a leading indicator; the call is editorial.

References

W3C — Understanding 3.1.5 Reading Level (opens in new tab)