A customer opens a product guide to decide whether the product will work for them. The instructions are inside an image. On a narrow screen, the words are hard to read.
They enlarge the image, then try to reach the rest of the instructions. This should be an ordinary part of shopping. Instead, they have to work around the page before they can make a decision.
Recent accessibility work on a storefront brought me back to that simple expectation. Adding the missing text was only part of making the guide usable. We also had to make sure someone could read it to the end.
Let’s follow that shopper through an illustrative example. The implementation details and test results come from the actual work.
Give the customer another way to read
Our shopper needs the instructions, but making the image larger leaves them moving around a picture to read its contents.
We added ordinary text beside the guide in the product gallery and in the enlarged view. This kept the instructions near the product, where someone could consult them without leaving the page.
That choice also gave us a constraint. The enlarged view already used an image viewer called PhotoSwipe. The caption had to work inside it.
On a narrow screen, a long caption cannot all fit at once. The customer needs to scroll. Giving the text a scrollable area looked like a reasonable solution, until we tried to use it with the viewer running.
PhotoSwipe handled the mouse wheel and vertical arrow keys to move the enlarged image. Input intended to scroll the instructions could be taken by the image viewer instead.
The customer could see the beginning of the text and still be unable to reach the rest.
Let them reach the last instruction
The caption needed to receive keyboard focus, so a person using the keyboard could move into it. It also needed to handle scrolling within that area without letting the image viewer take over those controls.
This is where the surrounding component changed the work. A caption that scrolls on a plain page can behave differently inside a viewer with its own keyboard and mouse controls.
We tested with the viewer bundled in the repository and the caption code from the pull request. The earlier version failed the scrolling comparison. With the revised handling, arrow keys and the mouse wheel reached the end of the text at a narrow width.
Now let the shopper move to another image, then come back to the guide. The instructions need to follow the right image. Closing and reopening the viewer should not add a second copy of the caption.
We checked those transitions too. They are ordinary things someone does while comparing products, and they can expose problems that opening the guide once will miss.
Keeping the existing viewer avoided replacing the whole gallery for this change. It also meant taking responsibility for how the added text interacted with the viewer. The local checks supported that implementation; live-page and assistive-technology checks still had their own work to do.
Follow the description all the way to the page
Someone using a screen reader needs the page to convey information they may not get from the image. That information has to survive the path from the content editor to the rendered page.
During the same accessibility work, we looked at images reported as missing alternative text, a written description of the image’s relevant information. Our first scan searched for images without an alt attribute.
It overstated what remained unresolved.
Some images were rendered by Replo, a page builder, and supplied a name through aria-label. We checked the browser’s accessibility tree, which shows the names and roles exposed to assistive software. The images we checked had names there. I had also tested the approach with a screen reader.
A missing attribute in our scan did not always mean the customer received no description. Other images were still unnamed, so those findings stayed open. The result applied to the components we checked; it was no reason to replace ordinary image alternative text across the site.
The content editor could be misleading in the other direction, too. A description saved in Shopify’s image library was not necessarily the description used by the page-builder component.
For the person shopping, the saved value helps only if the page delivers it. Before editing a description, I want to know which component controls what reaches them. Otherwise, we can make a reasonable change in the wrong place and leave their experience untouched.
Tell them what happened next
Reading the product information is one part of the purchase. Later, the shopper might enter a discount code and wait for the price to change.
If the code is invalid, a visible message can explain why the discount did not apply. Someone who cannot see that message needs another way to receive the answer while using the field.
In a separate Safari check, the page displayed the invalid-code error and associated it with the discount field. Focus stayed in the field, ready for another attempt. From those observations, it was tempting to think the interaction had enough information to work.
But when we repeated it with VoiceOver, Apple’s screen reader, the new error was not automatically announced. It read the field and code. Applying and removing a valid discount worked with the keyboard, but the invalid-code response still had an unresolved announcement problem at that point.
Without that answer, a shopper may wait longer or try the same code again. They should not have to guess whether the store rejected the code or is still working.
The field’s relationship to the error message was useful technical evidence. Listening to the interaction answered the remaining question about what was announced.
Let the customer continue
I start with what the person needs to do next. After reading the guide, they may compare another product and return to the instructions. After a rejected discount code, they need enough information to correct it or continue without the discount.
Automated checks help find repeated problems, but they cannot establish that the whole interaction works. W3C’s accessibility evaluation guidance makes that limit clear. We still need to follow the action through on the page the customer uses.
Our shopper should be able to reach the last instruction and decide whether the product suits them. If their discount code fails, they should hear why and choose what to do next. Neither step should require someone else to read the missing information for them.

