AI writing tools as accommodations: assistive technology, detectors and humanizers

At some institutions, yes, under the heading of assistive technology. CSUN's Universal Design Center quotes the Assistive Technology Industry Association's definition, software "used to increase, maintain, or improve the functional capabilities of persons with disabilities", and states that "AI and now generative AI expand the possibilities of assistive technologies" (CSU Northridge, Universal Design Center, Artificial Intelligence, Accessibility, and Accommodations, read 9 November 2026). The page lists examples by task: automated captions, image description, executive function support, and "AI tools can support people with cognitive or learning disabilities (or language differences) with reading and writing", naming Claude, Goblin Tools and Grammarly as examples and adding that "the tools listed are examples, not endorsements".

The same page gives faculty an instruction. "If you limit the use of AI in your course, but a student requests to use it as assistive technology, you or the student (or both) should consult with DRES and/or NCOD", the campus's disability resources and deaf and hard of hearing services. The decision sits with the disability office, not with the instructor and not with a detector.

UNESCO's Guidance for generative AI in education and research, published 7 September 2023 as the organisation's first global guidance on the subject, proposes a human-centred approach to regulating the tools, including data privacy protection and an age limit for independent conversations with generative AI platforms (UNESCO, Guidance for generative AI in education and research, read 9 November 2026). It is the policy frame above the institution. It does not decide an individual accommodation, and nothing in it names a humanizer.

Because detectors score regularity, and assistive tools produce it. A detector estimates how predictable each word is given the words before it, a quantity called perplexity. Liang and colleagues ran seven detectors over 91 human-written TOEFL essays in 2023 and found an average false positive rate of 61.22 percent, with the essays every detector flagged showing significantly lower perplexity than the rest; the same detectors were near-perfect on 88 US eighth-grade essays (Liang et al. 2023, Patterns, read 9 November 2026). The paper studied second-language writers, and its mechanism applies to anyone whose vocabulary is narrowed or regularised by circumstance or by a tool.

Grammar checkers, word prediction, speech-to-text and sentence-level rewrite suggestions each push a text toward the most probable phrasing, because that is what the suggestion is. A student with dyslexia who accepts a checker's suggestions across an essay ends up with text that is more regular than their first draft, which is the point of the accommodation and also what a perplexity-based detector responds to. The non-native English page describes the same bind for second-language writers.

Weber-Wulff and colleagues' 2023 test of 14 detectors adds that none reached 80 percent accuracy and that a student accused "only on this basis would have no possibility for a defence" because a detector returns a label without evidence (Weber-Wulff et al. 2023, International Journal for Educational Integrity, read 9 November 2026). A student with an accommodation letter has evidence the paper says is missing.

No, on the vendors' own descriptions. Under the definition CSUN quotes, assistive technology improves the functional capabilities of a person with a disability. A humanizer's stated job on most vendor pages checked on 8 and 9 November 2026 is to change how text reads to an AI detector; a few, such as Grammarly's, describe it as making AI text sound natural while keeping the message, and Grammarly's page says the tool "is not intended to bypass AI detectors", read 9 November 2026. Neither description is a functional capability of the student. No accessibility office page read for this guide names a humanizer as an accommodation, and this site ranks none for the purpose.

The practical difference is evidence. An approved assistive tool appears in an accommodation letter, and a reviewer can read the letter. A humanizer pass appears nowhere, removes the match between the student's draft and the submission, and carries the meaning risks the rubric on this site counts: a changed qualifier, a swapped term, a dropped clause. A student who was flagged after using an approved tool is in a strong position; a student who was flagged after an undisclosed humanizer pass has given up the strong position for a weaker one.

Process evidence and the accommodation record. GPTZero, a detector sold to schools, tells educators to ask students to show their process through edit history or in-person assessment and to look at patterns across several assignments rather than act on one scan (GPTZero, for educators page, read 9 November 2026). A student whose tool is in a letter and whose drafts are saved can meet that request in full. The steps are on the flagged-for-AI guide, and the mechanism of false flags is on the false positives guide.

The published method on this site runs six human-written control texts through five named detectors in every run and reports each detector's false positive rate on the same genres as the AI inputs, so a student or a disability office can see what a human text scores on the detector their institution uses. No detector score should be the only evidence in a misconduct case, and an accommodation letter is exactly the kind of evidence a detector cannot weigh. The rest of this series is in the guides hub.