A tool that reads a medication label through a camera and one that explains clinical information in conversation can both aim to improve access to healthcare information. What was built, whether people with visual impairment used it, and what an evaluation measured are separate questions. I designed a scoping review to keep those questions distinct while mapping the relevant literature.
I registered the protocol with OSF on 23 May 2026. Registration is complete, but the main review remains on hold. The project ledger dated 5 September still records database searching, screening and synthesis as not started. A fresh check on 7 September confirmed that the public registration remains available.
Defining the intersection of AI and healthcare information
The protocol covers AI tools intended to help people with visual impairment access information such as medication instructions, patient education materials and medical records. Different technologies, including text recognition, language models and conversational systems, can qualify if they meet that population and healthcare context. A study covering several disability groups would need to report the visual-impairment subgroup separately.
Basic screen readers without an AI component and general accessibility tools outside healthcare fall outside this review. That boundary determines which literature to collect; it is not a judgment about those tools’ value. I also separated development reports, user evaluation and deployment status in the planned record. A description of a working tool would not by itself establish a benefit to users.
The search and screening procedures are still plans
The intended search covers English-language scholarly literature from January 2015 to the actual search date. Alongside sources such as PubMed and Scopus, the protocol includes IEEE Xplore and the ACM Digital Library to reach engineering and computing research. With the Cochrane Library as a cross-check for relevant reviews, seven databases are specified. Product documentation, blogs and preprints are excluded from formal screening. This keeps the scope on published scholarly work but can miss tools that appear first as products, as well as research in other languages.
For a sole-author review, I planned a repeat check of my own screening decisions: re-screen a random 10% of records after one week and revisit the eligibility criteria if decisions differ. This is distinct from independent screening by two reviewers. The charting form—the fields used to record information from each study—would first be tried on the first five included studies and then refined. These procedures have not yet been carried out.
Recording what an evaluation actually assessed
The planned synthesis would describe each tool’s inputs and outputs, the people and methods involved in its evaluation, and whether it remained a prototype or had been deployed. Results measured on different scales would remain in their original form rather than being forced into a single effect estimate. There is no included-study count or finding about tool effectiveness to report at this stage.
The public plan is AI-Based Accessibility Tools for Healthcare Information Delivery to Persons with Visual Impairment: A Scoping Review. The decision I fixed in the protocol was to distinguish reports of tool development from evaluations of whether people understood the information, and record what each study actually established.