Screen reader research: let participants use their own tools
Plan research with screen reader users using their own tools, meaningful tasks and clear notes, without treating one session as a conformance audit.
Article language: English

In this article
Imagine a research team setting up a laptop, the latest browser and a screen reader. A participant arrives with her own computer, different settings and a way of working nobody in the room has seen before. The first research decision comes before she opens the website: let her work in her usual way, or start by training her on the company laptop? The second option might end up testing her patience with the organisers more than anything else.
A screen reader conveys interface content through speech or a braille display, among other means. How someone navigates depends on their tool, settings and habits. A study should let the participant carry out a meaningful task under agreed conditions. It does not require guessing a diagnosis or judging whether someone uses their computer the way the script anticipated.
In the fictional Aneks document library, we want to examine how a participant downloads the instructions for a particular product variant. She normally finds similar documents as part of her work and uses a screen reader. The examples below are a session plan, not an account of research that has taken place. They cover preparation, facilitation and recording a barrier; they do not establish that the library is fully accessible.
Agree on the conditions before booking a room
Ask how the person would like to take part, which device they use, what support they need and what format works for the organisational materials. The answer should help you prepare the session. You do not need a detailed health history to find out whether an invitation or consent form needs a particular format. Describe the task and working conditions, and leave room for the participant to explain their needs in their own words.
GOV.UK’s guidance on running research sessions with disabled people recommends that participants use their own assistive technology because personal settings are hard to recreate on another device. It also suggests visiting a participant who cannot bring their technology. This is organisational guidance, not a requirement to hold every session in the same place.
For Aneks, a remote session or a meeting in the participant’s usual workplace could be considered. Check beforehand whether she can operate the call software and whether screen sharing interferes with the task. If the researcher cannot hear the screen reader’s output, agree on another way to capture it, with her consent. Discuss any changes to audio settings before the test starts.
Prepare the account and task data so the participant does not have to reveal a private password or open personal correspondence in front of observers. Check that the test data fits the task scenario. If the session requires switching accounts or stopping screen sharing, agree on that before the test. Do not later count these organisational steps as difficulties with the service being studied.
Allow time to set up equipment, take breaks and agree on how you will communicate. The pace should not be dictated by a desire to fit the same number of tasks into every hour. If technical preparation takes longer, reduce the scope. Rushing can prevent you from understanding the very problem you organised the study to investigate.
Choose participants with the actual service in mind. Someone who knows a screen reader very well may be using a technical document library for the first time; someone else may know the products but be learning the tool. Those are different kinds of experience. Recruiting the right UX research participants helps you describe both without reducing a group to the single label “screen reader user”. For more context, see the article on recruiting UX research participants (in Polish).
If a helper or interpreter is present, agree on how you will work together while addressing the participant directly. Explain the roles of observers and note-takers. In the consent information and session description, say what will be recorded and who will use the material. The participant should be able to ask about recording or stop taking part without having to find an inaccessible button in yet another form.
Keep the task grounded in everyday work
The Aneks task might read: “You are preparing to install model L-40, version B. Find the instructions you would use.” Do not begin with “open the filter, choose the variant and click download”. That gives away the route you wanted to evaluate. Agree beforehand on what the participant knows at the start and what counts as successful completion.
Downloading the right file and reading its instructions are separate stages. If the test ends at the library, record that boundary. If it also includes the document, prepare the actual version and check the route to opening it in the participant’s usual tool. Choosing the right button does not confirm that the file is accessible, and difficulty in the document does not necessarily come from names in the catalogue.
Provide the task in a form the participant can return to while working. Reading the whole prompt aloud may not be enough, particularly while the screen reader is also speaking. Ask how they would prefer to receive it. Do not make someone memorise a fictional product code while discovering an unfamiliar interface if, in the everyday task, that code would be written in a customer’s message.
Before inviting participants, check whether the material supports the interaction you want to study. A mock-up made of images may show the visual layout, but it will not automatically reproduce the names, relationships and behaviour of a real website. Establish which questions it can answer. If the aim is to study screen reader navigation through the library, you need a version with appropriate content and behaviour, not just a screenshot.
Remove known blockers that would prevent someone reaching the area under study. If test login does not work with a keyboard, a session about choosing instructions may stall before the library. Record any organisational workaround, such as the researcher providing a signed-in session, as part of the conditions. Do not later attribute the entire journey to the participant when someone else performed a part of it.

A person’s usual approach may include a heading list, a search within the page, a different reading speed or skipping some content. Do not interrupt simply because it looks different from reading visually from the top. Ask what the person expects at a particular moment. “What are you looking for now?” gives you different information from telling someone who knows their own tool which shortcut to use.
This approach connects with observing a website in the workplace. The surroundings and a person’s own tools are part of the task. An additional document or note is not necessarily a workaround for a website defect. Record why the participant uses it before treating every action outside the browser as a sign of difficulty. The article on observing website use at work (in Polish) explores related considerations.
Listen for expectations and record what actually happens
When the participant pauses at the third button called “Download”, note the name announced by the screen reader, the element’s surroundings and what she says. Is the product, the variant or the effect of the action unclear to her? Those are three possible problems. A note saying “the screen reader struggles with buttons” is not enough. The person investigating needs to know which element to check and under what conditions the problem occurred.
Record the browser, assistive technology, software versions and settings relevant to the observation. Do not transcribe the computer’s entire configuration. If the participant uses a heading list, that route matters when reproducing the problem. An operating system name without an account of what the person tried to do will not give the person investigating enough to check the element’s behaviour.
W3C WAI’s guidance on analysing accessibility issues considers how content, browser and assistive technology work together, including how they are used. It also warns against generalising a limited study to all users. That is why the configuration and scope of the observation matter: one success or difficulty cannot stand in for an assessment of the entire service.
Keep three separate records: what you observed, what the participant said and what the researcher suspects. “She chose the third item called Download” describes an action. “I expected the instructions for version B” is a statement. “The button names probably do not distinguish the variants” remains a hypothesis to check against the element and in further testing.
If you want to help, agree on the help first. Taking over the mouse or changing the reading position may remove the problem from the screen, but it can also interrupt an independent attempt. Record when you intervened and what you did. The conversation afterwards may still be valuable; what changes is the conclusion you can draw about completing the task independently.
Do not drown out the tool’s announcements with more questions. Agree with the participant how to signal that you would like clarification and when to take a short pause. A moderated usability test calls for neutral facilitation, but neutrality does not mean remaining silent when something is unclear. Ask what an action means while keeping an explanation distinct from a hint about the solution. The topic is also discussed in the article on moderated usability testing (in Polish).
A long silence may mean the person is listening to content, waiting for the page to respond or thinking about a document name. Do not automatically treat it as uncertainty. At an appropriate moment, ask them to explain and record the answer. If a recording does not show the cause, leave that part unresolved. An honest gap in the notes is better than a confident explanation added later by someone who did not see how the person worked.
Check what a recording actually captured. Cursor footage alone may not show how someone navigated with a screen reader. A conversation transcript, meanwhile, may mix the tool’s speech with the participant’s words. Label those sources and check unclear passages before quoting them. An automatic transcript does not remove the need to establish who or what said something.
Check the fix without rewriting the session’s history
For Aneks, a possible solution might be to connect the instruction name clearly with the model and variant. Record the intended effect: the person can recognise the right document before downloading it. Then check the implemented change under the agreed conditions. Adding text to a design alone does not confirm that the information is available through the way the participant uses the service.
If the same participant returns to the revised version, account for her familiarity with the earlier task. She may remember the right item even if the new label is still unclear. Depending on the question, change the example or invite someone who has not seen the previous version. Do not present an easier repeat as an independent first-contact test.
For a follow-up session, choose a specific problem rather than asking the same person to go through the entire site after every small change. Recreate the conditions needed to assess the fix and agree on participation again. The participant shares experience from their own work; attending once does not make them a permanent tester of every future version. The scope of the next test should follow the research question.
Combine observations with an assessment of website accessibility. Research into experience and technical checks provide different information. An AI-assisted accessibility audit may cover selected rules, but its result does not reproduce the participant’s entire work. A green report does not, on its own, explain why a person selected instructions for the wrong variant. For a closer look, see the article on business website accessibility (in Polish) and the article on the scope of an AI-assisted accessibility audit (in Polish).
After the session, share the agreed information about next steps in a form the participant can use. If you want to confirm the meaning of a statement for the report, quote the specific passage and let them explain it. Do not require someone to sign in to a new tool just to answer the organiser’s question. This part of the interaction should also respect the communication method agreed earlier.
In the final material, identify the tasks, configurations, service version and help provided. Describe the effects of barriers without judging the participant’s abilities. You do not need to promise conformance across the whole site or determine the cause of every observation in a single conversation. Specific problems can inform a project with Aeneia Design. For the Aneks library, a useful improvement would let someone find the right instructions in their own way, without guessing what the button’s author was thinking.










