A website can look polished on a designer’s screen and still be difficult for a customer to use. A heading may be styled to look important without being marked as a heading. A form may work with a mouse but not a keyboard. A colour choice may look attractive while making text hard to read. An image may be meaningful to the business but silent to someone using a screen reader.

AI can help a small business find possible accessibility issues faster. It can review page text, suggest clearer labels, describe patterns in a page structure and help turn technical findings into a work list. It cannot experience a website as every person does, prove that a site is accessible or replace testing with disabled people.

This guide explains how to use an AI accessibility checker NZ businesses can understand and challenge. The emphasis is on practical improvement, respectful testing, and avoiding the dangerous idea that a score is the same as inclusion.

What an AI accessibility check can do

An AI-assisted review may inspect headings, links, image descriptions, form labels, language, colour contrast or repeated layout patterns. It can help identify missing information and draft plain-language alternatives. It may also summarise technical issues so a business owner can decide which ones need a developer, content editor or designer.

These suggestions are useful because accessibility work often fails through small, repeated omissions. A page may contain many images without meaningful alternative text, or several forms with labels that do not explain what is needed. A review can make those patterns visible and help a small team prioritise.

AI does not know the full context automatically. A decorative image should not receive a long description. A chart may need a text summary as well as alternative text. A warning colour may have meaning that is lost without words. The tool can suggest a change, but a person must decide whether it is accurate and useful.

Begin with the customer journey

Do not start by scanning every page and chasing a perfect score. Choose one important journey: finding opening information, requesting a quote, booking a service, reading an article, completing a purchase or contacting support. Test the journey on a real page from start to finish. This makes accessibility a business experience rather than a pile of isolated technical warnings.

Write down the task, the page, the expected outcome and the barriers found. Ask whether the customer can understand what to do, identify an error, move through the page, complete the form and receive confirmation. Include mobile and desktop views, different zoom levels and keyboard-only navigation where relevant.

An accessibility review should consider people with visual, hearing, mobility, speech, cognitive and learning disabilities. It should also recognise temporary and situational barriers: a bright room, an injured hand, a slow connection, a new language or a device used without sound. Inclusive design often improves the experience for many customers, not only those who identify as disabled.

Use AI to create a prioritised work list

Ask the system to group findings by customer impact and effort, not just by technical category. A missing form label may block a task. A slightly awkward sentence may be worth improving but less urgent. A decorative image with unnecessary alternative text may create noise rather than a serious barrier.

Require evidence for every suggestion. The work list should identify the page, the observed issue, why it matters, a proposed change and the person responsible for checking it. If the system cannot point to the relevant page element, mark the item for investigation rather than accepting it as fact.

Use plain language when communicating findings to the team. “The submit control is not understandable without seeing the screen” is more actionable for many owners than a technical label alone. Keep the technical detail where a developer needs it, but connect it to the customer task that may be affected.

Review text, images and structure carefully

AI can suggest clearer headings and shorter paragraphs, but readability is not the same as simplicity for every reader. Preserve the meaning, legal conditions and important qualifications. Do not remove a necessary warning merely because a shorter sentence looks cleaner.

For images, decide whether the image conveys information, supports the surrounding text or is decorative. Write alternative text that communicates the relevant purpose without guessing. If an image contains a complex chart or process, provide the information in the page text as well. Do not ask AI to invent details that are not visible or confirmed.

For page structure, check heading order, lists, tables, links, labels, error messages and focus order. A page can have good-looking headings that are not marked correctly in its underlying structure. A link called “click here” may be understandable in context but less useful when someone navigates a list of links. The fix should be tested in the real page, not accepted because an AI explanation sounds plausible.

Keep manual testing in the loop

An accessibility report is a starting point, not a certificate. Use the keyboard to move through the selected journey. Check that focus is visible, the order makes sense and no control traps the user. Test zoom and reflow. Listen to important pages with a screen reader where the team has the skills to do so, and check whether the words, controls and status messages make sense.

Automated and AI-assisted checks may miss whether instructions are understandable, whether an error message helps a person recover, whether a timed interaction is fair or whether a complex service is usable in practice. They may also flag a harmless pattern or miss a barrier created by a particular browser, device or assistive technology.

Invite disabled people to test the journey and pay them appropriately for their expertise. Ask what they were trying to do, where they became uncertain, what workaround they used and what would make the task easier. Do not ask a participant to disclose medical information that is not necessary. Treat feedback as product and service evidence, not as a performance test of the person.

Understand standards without overclaiming compliance

Accessibility guidance and standards can help a business define good practice, but a scan does not prove compliance with every requirement or legal duty. New Zealand government digital standards apply in the public-sector context, and the expectations for a private business can depend on its service, audience, contracts and circumstances. Get tailored advice when a website is part of a regulated or high-stakes service.

Avoid publishing a claim such as “fully accessible” based only on an AI score. A more honest statement is what the business has tested, which barriers it is addressing and how customers can report a problem. Make sure the contact route works and that someone is responsible for responding.

Do not use accessibility as a marketing badge while leaving a known barrier unresolved. A short, transparent improvement plan is more credible than a certificate-like graphic with no evidence behind it.

Protect content and customer information

Website audits may include unpublished copy, customer journeys, support messages or data from forms. Use only information the selected tool is approved to handle. Remove personal information from examples and do not upload private customer correspondence merely to make a wording suggestion.

Check who can access the audit, where it is stored, how long it is retained and whether prompts or uploaded material may be reused. Give the team a simple rule for reporting a possible disclosure. Keep the final accessibility backlog in an approved location with a clear owner and review date.

Be careful with AI-generated language about disability. Avoid stereotypes, pity-based framing and claims that a design choice will work for everyone. Prefer specific, respectful descriptions of barriers and user needs. When the subject is complex, seek input from people with relevant lived experience.

Make accessibility part of normal publishing

Accessibility improves when it becomes part of the content process rather than a once-a-year scan. Add checks before publishing: meaningful headings, descriptive links, useful alternative text, keyboard-friendly interactions, clear error messages, sufficient contrast and a way to test important changes. Keep a small set of representative pages for regular review.

Record what changed and retest the journey after a redesign, new form, new content type or platform change. A fix on one template may not reach older pages. Track unresolved issues by impact and avoid closing an item simply because an AI tool stops reporting it.

Used this way, an AI accessibility checker NZ businesses can manage becomes a practical assistant for finding patterns and explaining work. The goal is not a score. The goal is a customer who can understand, navigate and complete an important task with dignity, choice and a fair opportunity to get help.

Frequently asked questions

1. What is an AI accessibility checker?

It is an AI-assisted tool or workflow that looks for possible barriers in website content, structure or interaction and suggests improvements. Its findings still need human verification.

2. Can an AI checker prove that a website is accessible?

No. Automated checks can miss real barriers and report harmless patterns. Manual testing and feedback from disabled people are needed for a more complete view.

3. What should a small business test first?

Start with one important customer journey, such as contacting the business, booking a service or completing a purchase. Test the journey from start to finish on a real page.

4. Does a website accessibility scan prove legal compliance?

No. A scan is not a legal opinion or a universal compliance certificate. Requirements depend on the business, service, audience and circumstances; seek tailored advice where appropriate.

5. Should AI write all image alternative text?

No. AI can suggest wording, but a person must decide what the image means in context and whether it is informative, decorative or part of a complex explanation.

6. Why is keyboard testing important?

Some people navigate without a mouse, and keyboard testing can reveal missing focus, illogical order or controls that cannot be reached. It checks a real interaction, not just page code.

7. How should a business involve disabled people in testing?

Invite people with relevant experience, define the task, protect their privacy, pay them fairly and listen to the barriers they describe. Do not require unnecessary medical information.

8. How often should a business repeat an accessibility review?

Review important journeys regularly and whenever templates, forms, content types or major design elements change. Keep a small representative test set and retest fixes rather than relying on one annual scan.