National ADA Symposium Takeaway #3: AI Can Assist with Document Remediation. Human Review Protects Independence.

I am still thinking about the sessions at the National ADA Symposium from July 2026. There was a lot of talk regarding AI and whether it is good for document remediation. Many points were shared, and I walked away with a deeper understanding of the relationship among document accessibility, effective communication, and risk.

We generally regard document accessibility as primarily a technical issue, yet the information in a document can affect a person’s safety, health, legal rights, benefits, housing, or ability to participate in a government program.

In those situations, accessibility is not simply a technical issue. It may become a life-or-death issue.

Can AI fix a document in the right way?

My answer is: not by itself.

AI can assist with document remediation, but it is not consistently accurate enough to complete the process without qualified human review.

AI may add tags, suggest alternative text, identify headings, or attempt to establish a reading order. However, its work may appear technically correct yet still be inaccurate, incomplete, or confusing for someone using assistive technology.

For example, AI may:

• Generate alternative text that is technically present but fails to communicate an image’s purpose.
• Place content in a reading order that does not reflect the intended meaning of a document.
• Wrongly identify headings, lists, decorative elements, or table headers.
• Create a table structure that looks visually correct but does not make sense to a screen reader.
• Miss instructions that depend on location, color, shape, or visual formatting.
• Fail to determine whether someone can understand and complete a form on their own when tagging it.

AI does not always understand the document’s context or the consequences of mispresenting information.

That is a significant limitation when the person relying on an accessible document for accurate information encounters incomplete or incomprehensible details.

The person using the accessibility feature relies on the document’s accuracy. They have no way of knowing that there is a problem with the information presented.

Can AI accurately verify its own work?

This is where I believe organizations need to be especially cautious.

Using one automated tool to remediate a document and another to check it may create a false sense of security. A checker can confirm that certain technical elements are present, but it cannot always determine whether those elements are correct, meaningful, or usable.

A document can get a good automated score. It may still have serious accessibility problems. The fact that an automated check has been passed does not mean that equal access has been provided.

Some documents carry a much greater risk.

Not every inaccessible document creates the same level of immediate harm. Public entities may have thousands of documents. This is why they need to establish priorities as they work toward accessibility.

Documents that are both high-risk and frequently used should receive immediate attention. This may include:

• Emergency alerts and evacuation instructions
• Shelter locations and disaster-response information
• Court notices, hearing instructions, and legal deadlines
• Healthcare information and medication instructions
• Benefits applications and eligibility decisions
• Voting information
• Complaint and appeal procedures
• Forms required to access public programs or services

Imagine receiving evacuation instructions that your screen reader cannot interpret accurately.  Missing a court date because information was presented in an incorrect reading order. What about being unable to complete a benefits application or to understand how to appeal a decision because the form was inaccessible? These are not minor inconveniences. They can affect a person’s safety, independence, legal rights, and ability to make informed decisions.

I think that if the consequences of an error are serious, letting AI make the final accessibility decision involves an unnecessary risk.

Qualified human review matters

A qualified accessibility professional can evaluate documents that require context, judgment, and an understanding of how people with disabilities use assistive technology.

How do you find a qualified accessibility professional?

Look for those who have completed accredited accessibility-testing programs, such as the Department of Homeland Security Section 508 Trusted Tester certification or the International Association of Accessibility Professionals (IAAP). These certifications indicate that the individual has received formal training in a standardized, repeatable method for assessing accessibility compliance.

Training employees to create more accessible source documents can also reduce errors and help control remediation costs. However, training, AI, and automated tools do not eliminate the need for qualified human review.

Takeaway #3

When a document affects someone’s safety, health, legal rights, benefits, or access to government services, “the computer said it passed” is not enough.

Although AI can assist with document remediation, it should not be relied upon to carry out or verify the whole process without a qualified human being involved.

Accessibility ensures that people with disabilities can access information, understand it, make decisions, and take action without relying on others. It also ensures that people can remain independent by managing their own healthcare, handling an emergency, understanding their legal obligations, applying for benefits, voting, and protecting their rights, all with the same level of privacy and control as everyone else.

A person has lost the ability to manage their own life when they must ask someone else to read a notice, fill in a form, or explain their rights. So, AI is not a suitable cost-saving method when it comes to someone’s life, safety, and independence.

Related: National ADA Symposium Takeaway #2: Effective Communication is a Title II Responsibility