What automated accessibility scanners can't tell you

There is a number that gets quoted at every accessibility conference. Automated tools catch somewhere between 30 and 40 percent of real-world accessibility barriers. That leaves the other 60 to 70 percent uncovered.

The remaining barriers are the ones a scan can't see.

That is the gap LumiLens is trying to fill.

What a scan can see

Automated accessibility scanners are good at the things computers are good at. They can reliably catch:

• -Images with no alt text
• -Form fields with no labels
• -Text with insufficient contrast against a solid background
• -ARIA attributes that point to elements that don't exist
• -Missing language declarations
• -Headings out of order
• -Buttons that are actually div elements pretending to be buttons

LumiLens uses five scanning engines to cover as much of this ground as possible: 64 proprietary rules on top of axe-core's defaults, including keyboard navigation tests and computed visual checks that most scanners skip.

But even when every one of those checks passes, the site might still be barely usable for someone who uses a screen reader.

What a scan can't see

The things a scan can't reliably catch include:

Whether alt text is meaningful. A scan can tell you that an alt attribute is empty. It can't tell you that "alt=image" is useless, that "alt=banner-final-v3.jpg" is worse than nothing, or that an image of a chart needs the chart's actual content described.

Whether a page makes sense to a screen reader. A page might have all the right landmarks, headings, and labels, and still feel like a maze when you can't see it. Logical reading order, sensible heading hierarchy, and clear language are judgement calls.

Whether interactions work. A scan can tell you a button has an accessible name. It can't tell you that the button opens a dialog where focus jumps somewhere unexpected, or that search results announce themselves so loudly they interrupt every screen reader user mid-sentence.

Whether the cookie banner is a barrier. Almost every cookie banner is a small accessibility disaster. Some scans flag the obvious ones. None can tell you whether the banner traps your focus, blocks the page until you accept tracking, or simply refuses to respond to keyboard navigation.

Whether the site works the way you, specifically, browse. NVDA, JAWS, VoiceOver, and TalkBack all interpret the same code differently. A site that works perfectly in VoiceOver can be confusing in JAWS. The only way to know is for someone who actually uses your screen reader to try.

Why this matters

If you have ever read an accessibility audit and thought "this site says it is compliant but I can't actually use it," this is why. Compliance documents the things that can be measured. Usability is the part that gets measured by the person using the site.

The accessibility industry has a long history of mistaking the first for the second. Sites that score 100 on Lighthouse and 0 violations on axe-core can still be unusable. We have seen it on banking sites, government services, and shops.

That is not because the scans are wrong. It is because scans are answering a different question to the one you, as a user, are asking. Scans answer "does this code follow the rules?" You are asking "will this site work for me?"

How LumiLens approaches the gap

We do three things to try to close the distance.

1. Personalised reports. When you tell us which assistive technology you use, the AI report adapts. A screen reader user gets screen-reader-specific guidance. A keyboard user gets focus and tab-order guidance. We can't simulate using the site as you do, but we can flag the things that are likely to cause problems with your specific setup.

2. Plain-language summaries. Every report is written for someone who wants to know "can I use this site," not "which WCAG criteria failed." Technical findings are translated into what they mean in practice. "Focus indicator removed" becomes "you may not be able to see where you are on the page when navigating by keyboard."

3. Community reviews. This is the part automated tools can't replicate. When someone who uses JAWS leaves a review of an online bank, they are documenting something no scan can see: what it actually felt like to use the site. Over time, community reviews on each site profile build up a picture of the real experience, sorted by which assistive technology the reviewer uses.

What you can do with this

If you find a site you depend on and want to know more than the scan can tell you, the LumiLens directory has reviews from other AT users on every scanned site. You can filter by assistive technology to find people with similar setups.

If you have used a site and have something to say about it, leave a review. It takes about a minute. Your experience is what makes the directory worth more than a list of automated scores.

What we are honest about

We say it on every report, but it bears repeating here.

LumiLens reports are an estimate, not the truth. Our scans are automated. The AI summary is generated. The score is an indicator, not a guarantee. A high LumiLens score is more likely to mean a site will work for you than a low one, but it is not a promise.

The truth about a site's accessibility is something only the people using it can tell you. We are building the place for them to do that.