WCAG 3 Plain English Explorer

Simulated motion in sound can provoke nausea, disorientation or vestibular symptoms, and it makes speech harder to follow for people who process audio with one ear or with a hearing aid.

Vibration can cause pain for people with chronic pain conditions or tremor, and it can be startling or disorienting for people with sensory processing differences.

The size exemption assumes a small flashing area is safe, which does not hold for everyone or for every viewing distance. Removing it closes that gap for people with photosensitive epilepsy.

Flashing above these thresholds can trigger seizures in people with photosensitive epilepsy. This is one of the few accessibility failures that can cause immediate physical harm.

Where the exception applies, a user who cannot tolerate motion has no route through the content at all. This version leaves none.

Sustained motion can cause nausea, dizziness and migraine in people with vestibular disorders, and it pulls attention away from the content for people with attention-related disabilities.

Encountering distressing content without warning can trigger trauma responses, and people who need to prepare for or avoid such material cannot do so if nothing signals it is coming.

A single full-screen flash can trigger seizures and can cause pain or after-images for people with light sensitivity, migraine or eye conditions - the harm does not require a repeating flash.

A warning alone still forces a choice between the content and the trigger. Providing an alternative route means people with PTSD, anxiety disorders or phobias can get the information without the exposure.

Visually busy interfaces can overwhelm people with sensory processing differences, autism, migraine or cognitive disabilities, to the point where the content cannot be used at all.

People who navigate by searching for a remembered label - screen reader users, voice control users, and anyone with a memory or reading disability - lose their way when the same destination is called something different on the next page.

Once someone has learned that a link is the fourth item, changing the order makes every page a fresh problem. That cost falls hardest on screen reader, switch and voice users, for whom re-scanning is slow.

A predictable structure is what lets people move quickly to the part they want. Reordering it between pages forces everyone to re-orient, and forces assistive technology users to re-explore.

An error announced far from the field it belongs to can be missed entirely by someone using magnification, and leaves screen reader users to hunt for which field is wrong.

A message that disappears on a timer is gone before many people have finished reading it, particularly people who read slowly, use magnification, or are navigating with a screen reader.

Colour or an icon alone tells a screen reader user nothing, and 'invalid input' tells nobody what to change. A described error in text is what makes the problem fixable.

Knowing that a field is wrong is not the same as knowing what it wants. Suggestions are what make a form completable for people with cognitive or learning disabilities.

A programmatic association is what lets a screen reader announce the error when the user reaches the field, rather than leaving it stranded somewhere else on the page.

Colour alone is invisible to people with colour vision deficiencies and disappears in high-contrast modes. A second, redundant signal means the error still registers.

Validation that fires while someone is still typing produces errors for input that is merely unfinished, which is disorienting for screen reader users and discouraging for people who type slowly.

Most form failures are design failures. Reviewing the design removes errors before anyone has to recover from them, which matters most for people with cognitive and learning disabilities.

A mistake that can only be discovered after submission is expensive to undo. A review step is what makes an irreversible action safe for people who are more likely to mistype or misread.

Without an announcement, a screen reader user has no way to know whether anything happened, and may submit again - or leave believing a task is done when it failed.

When something is inaccessible, a feedback route is often the only way a disabled user can get it fixed. Its absence makes the barrier permanent.

Help that moves around is help that cannot be found in the moment it is needed, which is exactly when someone with a cognitive disability has least capacity to search for it.

Generic help pages put the burden of translation on the user. Help tied to the current task is what lets people with cognitive and learning disabilities finish it.

Some people cannot use speech and some cannot use typing. Offering both formats means the support route itself is not the barrier.

Charts encode meaning in position, colour and shape at once. Without an explanation or an alternative route to the data, that meaning is unavailable to screen reader users and to many people with cognitive disabilities.

A greyed-out button with no explanation is a dead end: the user cannot tell whether they have missed a step, lack permission, or hit a bug.

Help written without testing tends to answer the questions the authors imagined. Testing with the people who rely on it is what finds the gaps that matter.

A redesign resets the learned patterns people depend on. For users with cognitive disabilities that relearning cost can be severe enough to end their use of the product.

One fixed explanation suits one kind of reader. Being able to change the form of help - its length, its vocabulary, its medium - is what makes it usable across very different cognitive profiles.

Directions given purely by appearance or position are unusable by screen reader users, and often by people with colour vision deficiencies or those using a reflowed, magnified layout.

These are the moments when people get stuck and abandon a task. Support offered at that point, rather than in a manual, is what keeps the task completable.

High-stakes decisions are where poor comprehension turns into real harm. Extra support at those points protects people with cognitive disabilities from consequences that are hard to reverse.

A custom player can undo the accessibility of the media inside it. Naming the features it actually supports makes that checkable rather than assumed.

Captions and descriptions only help if the player exposes them and lets people adjust them. Caption styling in particular is what makes them readable for people with low vision.

A description in a language the viewer does not speak conveys nothing. This matters most for blind and low-vision users, who have no fallback to the picture.

When description and soundtrack are locked together, one drowns out the other. Independent control is what makes both audible for people who rely on the description.

Live events are exactly where a blind or low-vision viewer cannot catch up later. Without description, anything shown rather than said is lost.

Anything conveyed only by the picture - a document on screen, a gesture, who just entered the room - is simply absent for blind and low-vision viewers without description.

Description is welcome to some listeners and intrusive to others sharing the same screen. Control is what lets both use the same content.

A description that names what is on screen without conveying what it means leaves blind viewers with a thinner version of the content than everyone else gets.

Descriptions produced at volume, or automatically, drift from what the picture actually shows. A recorded review is the check that they still match.

Consistency across a library is what lets a regular listener build expectations. A style guide is how that consistency survives changes of author and vendor.

A description that collides with the dialogue costs the listener both. Timing is what keeps the described version coherent.

Whether a description works can only be judged by someone using it without the picture. Testing is what turns that judgement into evidence.

Dense visual content cannot be described in the pauses a soundtrack happens to leave. Pausing is the only way to convey it without cutting information.

The pause is bought at the cost of the viewer's time; an inadequate description wastes it and still leaves the content incomplete.

Meaning frequently sits in what nobody says. Without it, a blind listener gets the words of a scene and not its sense.

A door closing offscreen or an alarm can carry the plot. Deaf and hard-of-hearing viewers only get it if it is written down.

Without it, a listener cannot tell whether they have missed something or the speaker has simply switched language - and may not know a translation exists.

Voices alone are hard to tell apart without the picture, particularly for listeners with hearing loss as well as sight loss.

This is the core of the technique: what the picture carries and the soundtrack does not is exactly what a blind viewer would otherwise lose.

Being able to search, pause on or re-read a caption gives deaf and hard-of-hearing viewers the same control over pace that hearing viewers get from replaying audio.

Captions in the wrong language are no more use than no captions. This matters most where an audience shares a region but not a first language.

Fixed captions are unreadable for many low-vision viewers. Size, contrast and position are the settings that make the difference between legible and not.

Live is where there is no catching up afterwards. Without captions, a deaf or hard-of-hearing participant is excluded from the event as it happens.

Captions are the primary route into spoken content for deaf and hard-of-hearing users, and the route many other people use in noisy or silent settings.

Captions anchored to a fixed point in the scene vanish the moment the viewer turns their head, which in an immersive environment is constantly.

Reading captions is already a divided-attention task. Consistent conventions mean the viewer spends attention on the content rather than on decoding the captions.

Captions cover part of the picture and are a distraction to viewers who do not need them, so the choice has to sit with the viewer.

Auto-generated captions routinely drop names, technical terms and negations. An approximate caption can change the meaning of what was said.

Machine captioning is now the default, and its errors are systematic. A recorded review is what catches them before publication.

Caption conventions - how speakers are named, how sounds are written - only help if they are stable. A style guide is how they stay stable across teams.

Captions that run ahead of or behind the speech break the link between the words and who is saying them, and make dialogue hard to follow.

Captions sitting over on-screen text or a speaker's face force a deaf viewer to choose between two pieces of the same content.

Caption quality problems - timing, reading speed, speaker attribution - show up in use rather than in a checklist.

Spatial audio is the usual cue for where to look, and it is unavailable to deaf and hard-of-hearing users. A visual direction cue restores it.

Tone, sarcasm and emotion carry meaning that the words alone do not. Captions that omit them can invert what a scene means.

Significant sounds - a phone, a knock, music signalling a change - are part of the content, not decoration around it.

A deaf viewer cannot hear the switch. Naming the language is what prevents a translated line from reading as the speaker's own words.

Captions strip out voice, accent and volume, which is how hearing viewers tell speakers apart. Without attribution, a conversation becomes a block of text.

A caption that vanishes when focus moves cannot be read alongside the figure it explains - which is exactly how people using magnification or a screen reader read it.

Generated descriptions miss what the image is for in its context. Editability is what lets the person who knows the purpose correct it.

Announcing every decorative image buries the content in noise, and for a screen reader user that noise costs real time on every pass through the page.

Without one, a screen reader user gets a filename or silence where the content is - and if the image is a link or a control, may not be able to use it at all.

"Photo of a chart" tells a blind reader that information exists and withholds it. The alternative has to carry what the image is there to convey.

Alt text is written by many hands over time. A style guide is what keeps it purposeful rather than decorative or duplicative.

Knowing the kind of image sets expectations about what the alternative will contain and whether a longer description is worth seeking out.

Knowing an image exists is what lets someone decide to seek out its description. An undetectable image is invisible even to a user who would want it.

Text is the one form that can be rendered as speech, braille, larger type or a different colour. It is what makes content translatable to any user's needs.

For many deaf people a sign language is their first language and the written language of the page is a second one, so captions are not an equivalent route.

An interpreter window occupies part of the picture. Viewers who do not use it need it out of the way, and viewers who do need it present.

Live interpretation has to be booked ahead. A published route and deadline is what makes attendance possible rather than a matter of luck.

Icons, chart lines and diagram strokes carry meaning the same way text does. At low contrast that meaning disappears for people with low vision, and for anyone on a dim or glare-hit screen.

Around one in twelve men has a colour vision deficiency, and hue is also lost in greyscale printing, high-contrast modes and bright sunlight.

A beep signalling success or an error is silent for deaf users, for anyone with the volume down, and for anyone in a noisy place.

Spatial cues require hearing in both ears. They are unavailable to people with single-sided deafness or asymmetric hearing loss, and to anyone listening on one earbud or a mono speaker.

Stereoscopic depth needs usable vision in both eyes. People with monocular vision or amblyopia get no depth cue at all, and flat displays remove it for everyone.

A transcript is the one format that can be read at any speed, searched, translated, and rendered in braille. It serves deafblind users, who have no route through either captions or audio description.

A live transcript is what lets someone catch up on a point they missed, which captions scrolling in real time do not allow.

Reading at one's own pace is the difference between following a recording and merely playing it, particularly for people with cognitive or processing disabilities.

A transcript of words alone loses the tone and reaction that told the audience how to read those words.

For a reader who never hears the audio, an unmarked sound is simply an event that did not happen.

Without the marker, a reader cannot tell a translated passage from the speaker's own words, or know that a second-language version exists.

An unattributed transcript reads as one continuous voice, and the argument in a conversation becomes impossible to follow.

A transcript that summarises rather than conveys gives its readers less than everyone else receives from the same content.

A transcript nobody can find is not an alternative. Adjacency is what makes it a real choice at the moment the media is encountered.

Automatic transcription errors cluster on exactly the words that carry meaning - names, terms, numbers. Review is what catches them.

Transcripts are only easy to read when their conventions hold steady across a library.

Whether a transcript is usable depends on its structure and labelling, which shows up in use rather than in a spot check.

This is what separates a descriptive transcript from a dialogue list, and what makes the transcript a complete substitute for the media.

Fingerprint and face recognition fail for people whose bodies the sensors were not designed around, including people with tremor, limb differences, facial differences or certain eye conditions. A failed scan with no alternative locks the account.

A person who cannot use one biometric may well be able to use another; offering only one turns a single sensor's limits into a lockout.

Voice recognition fails for people who are non-speaking, who have dysarthria or a speech difference, or whose voice changes with a condition or treatment.

Requiring someone to move their whole body excludes wheelchair users, people with limited mobility and anyone seated at a fixed workstation.

Eye tracking assumes steady, controllable gaze. Nystagmus, strabismus, low vision and blindness all make it unreliable or impossible.

A tooltip that disappears when you move toward it cannot be read by anyone who uses magnification, has a tremor, or simply moves the pointer slowly.

A popup covering the field you are typing in is a trap for magnification users - dismissing it by moving away means losing your place.

Content that times out on its own is gone before slower readers, screen magnifier users and people using switch input have finished with it.

Many people move between keyboard, pointer, voice and touch during one session, often because of fatigue or pain. Being locked in forces them to stop.

Path gestures require fine motor control and a pointer that can be dragged accurately, which rules out switch users, many voice users and people with tremor.

When focus and pointer disagree, a user who alternates between mouse and keyboard loses their place, and a screen reader announces somewhere other than where the user is working.

One-way navigation means a mistake can only be corrected by starting the page over, which for a keyboard-only user can mean dozens of keystrokes.

An undocumented shortcut may as well not exist, and a user who discovers one by accident cannot tell what else is available.

Focus landing on something invisible strands sighted keyboard users, who can see the focus indicator disappear but not where it went.

Being dumped back at the top of a long page after closing a dialog can mean re-traversing the entire page, which is slow and disorienting for keyboard and screen reader users.

Focus that moves on its own takes control away mid-task: a screen reader starts reading somewhere else, and typed input can land in a field the user never chose.

The keyboard interface is what screen readers, switch devices, sip-and-puff systems and voice control all drive. Content it cannot reach is content those users cannot reach.

Reaching a control is not the same as being able to use it. A custom widget that only responds to a mouse event is a dead end for every keyboard-driven assistive technology.

A single-key shortcut that overrides a platform command can disable an assistive technology's own keys, which is worse than having no shortcut at all.

A keyboard trap ends the session: the user cannot move on, cannot go back, and often cannot reach the browser's own controls.

Content can be technically keyboard-operable and still take five times as many keystrokes. That gap is a fatigue and pain cost for people with motor disabilities and RSI.

Custom interaction models are not discoverable by exploration. Without a description, a keyboard user has no way to know the technique exists.

Duplicated links and buttons double the tab stops and the screen reader announcements for no added function, which is pure cost for keyboard and screen reader users.

Cancelling an accidental press is a skill users learn once. Changing the rule between pages means accidental activations stick.

People with tremor, limited dexterity or an imprecise pointer press the wrong thing regularly. Acting on release rather than press is what makes a mis-press recoverable.

Pressure-sensitive input assumes controlled force, which is not available to people with weakness, tremor, or those using a head pointer, stylus or switch.

Speed requirements like double-click timing or a fast flick exclude people whose movements are slow or variable, including many older users.

Multi-point and timed gestures are unavailable to people using a head pointer, switch, stylus or single finger, and to anyone whose movements are slow or imprecise.

Many non-speaking people communicate through a speech-generating device. A system tuned only to human voices rejects them at the door.

Real-time text lets a deaf or non-speaking person hold a live conversation at conversational pace, including in emergencies where a voice call is the default channel.

Requiring speech excludes people who are non-speaking, who have a speech difference, who use a communication device, or who are in a setting where they cannot speak.

A state change nobody announces - a button that has become pressed, a field now marked invalid - leaves screen reader users acting on stale information.

Constraints discovered only by failing are a guessing game, which is costly for people with cognitive disabilities and for anyone whose input is slow or effortful.

If the boundary of a control or the mark showing it is selected is too faint to see, the control is effectively invisible to people with low vision, even though its label is readable.

When every control looks equally weighted, users have to read all of them to find the one that matters - a real cost for people with cognitive disabilities or limited screen area.

Users rely on labels to know what will happen when they click or tap. Screen readers need programmatic names to announce controls accurately.

Flat design that removes button edges and link underlines makes people hunt by hovering, which is impossible on touch and unreliable for anyone with low vision or a cognitive disability.

Voice control users say what they see. When the accessible name does not contain the visible words, the command simply fails.

This is what assistive technology reads. An element with a wrong role is announced as the wrong kind of thing, and a missing state hides whether it is on, off, expanded or selected.

Spatial memory is how most people find familiar controls quickly. Moving them costs the most for magnification users, who see only a fraction of the screen at once.

Inconsistent behaviour forces relearning at every encounter, which is exactly the cost people with cognitive disabilities can least absorb.

Assistive technologies are built around conventional patterns. A custom component that ignores them may be unusable even when it is technically labelled.

The default indicator is designed to be visible against unknown backgrounds and is what users already recognise. Replacing it is where most focus indicators become invisible.

An indicator that cannot be distinguished from the background tells a sighted keyboard user nothing about where they are on the page.

A hairline outline is easy to lose on a busy page, particularly for people with low vision or when the page is magnified.

Focus styling is usually removed by accident, one component at a time. A style guide is what keeps it from eroding across a product.

When focus order and reading order disagree, keyboard and screen reader users get the page in an order it was never written to be read in.

Tabbing through invisible or repeated stops wastes keystrokes and makes the focus indicator seem to vanish - a common reason keyboard users abandon a page.

Escaping an overlay by tabbing through everything inside it is slow enough that many keyboard users give up on the task.

People who have enlarged or recoloured their cursor did so because they lose it otherwise. A custom cursor undoes that setting without asking.

Finding the pointer is a recurring problem for people with low vision, attention-related disabilities and anyone on a large or high-resolution display.

On touch and immersive platforms there is often no cursor at all, leaving no feedback about where an action will land.

Without confirmation, users repeat the action - which on a form or a payment is worse than no feedback at all.

A custom cursor that is smaller or fainter than the system one is harder to track for exactly the people who most need to track it.

Losing the pointer means losing your place. The exceptions cover the cases where a visible cursor would obscure the content itself.

Auto-hiding cursors force users to move the pointer to find it, which is a real cost for people with motor disabilities and for anyone with low vision.

Infinite scroll can put the footer permanently out of reach and makes a page impossible to get to the end of - a particular problem for screen reader and keyboard users.

A cookie banner or chat widget with no reachable close button can cover the content permanently for keyboard users and for anyone using magnification.

Conventional layouts are what let people apply what they already know. Departing from them without testing shifts the learning cost onto users with cognitive disabilities.

Clear visual grouping is how readers find their way without reading everything, and it is what keeps a page navigable when magnified or reflowed.

The programmatic part is what lets a screen reader jump between sections; the spacing is what lets a sighted reader see where one ends and the next begins.

Structure is decided while writing, not afterwards. A review process is what keeps clear structure from depending on which author happened to write the page.

Headings are the primary way screen reader users navigate a long page; a flat or scrambled hierarchy removes that route entirely.

Whether an information architecture works is not something its authors can judge - they already know where everything is.

Without position markers a screen reader user hears a list of items with no sense of sequence, which matters when the order is the information.

A label beside a field, a cell under a column header, a caption under a figure - all of these are obvious by sight and invisible to assistive technology unless the relationship is marked up.

Labelled regions are what let a screen reader user skip to navigation, search or main content instead of reading through the page to find them.

Knowing how much is left is what lets someone decide whether they can finish now, which matters for people with fatigue, pain or limited time on a shared device.

Without it, a user who is interrupted cannot tell where they are on returning, and a screen reader user may not be able to tell at all.

Knowing where you are is the precondition for deciding where to go next; it is easily lost in deep or dynamic interfaces.

In a single-page application nothing reloads, so a screen reader user may not be told anything happened and can continue reading content that is no longer there.

The title is the first thing a screen reader announces and how users tell tabs, windows and history entries apart.

A reliable way home is what makes exploring safe. Without it, getting lost means starting over.

A dependable route home is what makes it safe to explore an unfamiliar interface, and it is what a disoriented user reaches for first.

Models trained on data where disabled people are rare or stereotyped reproduce that in their output - and increasingly that output is the interface.

Algorithmic decisions about ranking, moderation or eligibility can disadvantage disabled people invisibly, without any individual decision looking wrong.

Naming what was reviewed and which disabilities were in scope is what turns a general claim of fairness into something that can be checked.

Consequences disclosed after the fact are not a choice. Presenting them beforehand is what makes consent meaningful for people who need time to process them.

Interfaces routinely present consequential choices as routine ones. An explicit description is what protects people who take the interface at its word.

Safety design usually assumes an average user. Disabled users are disproportionately targeted, and a threat model that leaves them out leaves them unprotected.

If turning on accessibility features marks a user out, weakens their security, or discloses their disability, the price of access is exposure.

Time limits assume a typical speed of reading, typing and decision-making. People using screen readers, switch input or voice, and people with cognitive disabilities, routinely need longer.

Most time limits are implementation convenience rather than necessity, and each one silently excludes whoever cannot work at that pace.

A limit nobody mentions is discovered by losing work. Knowing about it in advance is what lets someone adjust it or plan around it.

Without an extension, a session that times out mid-form destroys work, and the people most likely to hit it are those who were already working hardest to complete it.

Deceptive patterns work by exploiting the gap between what an interface appears to offer and what it does - a gap that is widest for users who read it literally.

Expert review catches patterns that individual users experience as their own mistake rather than as a design decision.

Manufactured urgency pushes people into decisions before they have understood them, and hits hardest anyone who needs longer to read or decide.

A pre-ticked box is a decision made on the user's behalf. If it is not announced, a screen reader user may agree to it without ever encountering it.

Blocking paste or autofill forces manual transcription of exactly the data that is longest and least memorable, which is a barrier for password manager users and people with cognitive or motor disabilities.

Cognitive tests like CAPTCHAs lock out users with memory loss, dyslexia, or brain injuries.

Copying is how people move text into a translator, a reader, a note or a password manager. Blocking it removes an adaptation the user had already made for themselves.

Requiring recall turns a task into a memory test, which excludes people with memory impairments, and is harder for anyone interrupted mid-process.

Discovering halfway through that a document is needed means abandoning the task. Knowing in advance is what lets people with limited time, energy or attention start at the right moment.

Processes that assume prior knowledge exclude first-time users, and disproportionately those who cannot infer the missing steps from the layout alone.

A process that stalls without saying why reads as a broken page. The user most likely to give up is the one who cannot see the unfilled field.

Correcting an earlier answer should not cost the whole form. Losing data on going back punishes exactly the users who most often need to check their work.

Re-typing the same data is effort and a new chance to make a mistake, which is costly for people with motor and cognitive disabilities and slow for anyone using speech input.

Being able to stop and come back is what makes a long task possible for people with fatigue, pain, attention-related disabilities, or an unreliable connection.

Every extra step is a place to give up. Which steps are unnecessary is visible in testing and invisible from inside the design.

Unexplained abbreviations stop readers who are new to a domain, reading in a second language, or have a language or learning disability - and screen readers often pronounce them as words.

Clarity is a property of how content is produced. A review policy is what makes it survive deadlines and changes of author.

Vocabulary is the most common barrier in text. Uncommon words stop readers with cognitive and learning disabilities, and readers working in a second language.

Without them, a word can be read several ways. Readers who are still learning the language, or who have a reading disability, cannot reliably pick the right one from context.

Each nested clause is something the reader has to hold in mind while the sentence finishes. That load is what makes otherwise simple information unreadable for people with cognitive disabilities.

Padding lengthens the path to the meaning, which costs the most for people who read slowly, use text-to-speech, or have limited working memory.

Non-literal language is routinely read literally by people with some cognitive disabilities and by autistic readers, and it rarely survives translation.

Statistics compress a lot of meaning into a form many readers cannot unpack, and a misread number can change a decision about health or money.

A summary is how a reader decides whether to spend the effort, and for people with cognitive disabilities or limited reading stamina it is often the only part they can get through.

Some relationships are far easier to see than to read. A diagram is often the difference between a process being followable and being merely described.

Readers with dyslexia, low vision or attention-related disabilities often need wider spacing or shorter lines. Layouts that break when those change force a choice between reading comfortably and reading at all.

Setting readable defaults means users do not have to know that adjustment is possible, which is most of them.

Long lines and tight leading cause readers to lose their place on the return sweep - a problem for everyone, and a barrier for people with dyslexia or low vision.

Many people read only with a specific colour combination - light text on dark, or a tinted background for scotopic sensitivity. Content that resists recolouring blocks the adaptation they depend on.

The minimum level is a floor, not a comfortable reading level. The enhanced level is aimed at people whose contrast sensitivity is significantly reduced.

Low-contrast text is hard or impossible to read for people with low vision, and for anyone in bright light.

Adjustments that vanish on export mean the saved copy is unreadable to the person who saved it - and they usually have no way to reapply them.

Enlarging text is the most common adaptation there is. Layouts that clip, overlap or scroll sideways at 200% take it away from people with low vision.

Readers with dyslexia often substitute a typeface, and all-caps or hyphenated text is markedly harder to read. Being able to change these is what makes text legible for them.

Defaults decide what most readers get. Setting them at the enhanced level removes the need for individual adaptation.

This is the floor below which text is difficult for a large share of readers regardless of any setting they might change.

Screen readers choose pronunciation rules from the language tag. Untagged or mistagged text is read in the wrong accent and is often incomprehensible.

"03/04" is two different dates depending on the reader, and a screen reader may voice a Roman numeral as a word. Context is what stops a number being misread.

Bold, colour and size are invisible to a screen reader unless marked up. Meaning carried only by appearance simply does not reach the user.

Text baked into an image or drawn on a canvas cannot be read aloud, enlarged, recoloured or translated. The programmatic equivalent is what restores all of those.

320 CSS pixels is what a 1280-pixel page looks like at 400% zoom. Content that forces scrolling in both directions at that size makes reading a line a two-handed operation.

A device mounted to a wheelchair or a bed is fixed in one orientation. Content that only works the other way round cannot be used at all.

Locking orientation overrides a physical setup the user cannot change, which is why this is a baseline rather than an enhancement.

Having to scroll sideways to read every line makes sustained reading impossible, which is the everyday experience of magnification users on unreflowed layouts.

This is the baseline the rest depends on: content that blocks or breaks assistive technology is inaccessible no matter how well it is labelled.

Interruptions break concentration and can steal focus mid-task, which is particularly costly for people with attention-related and cognitive disabilities.

People configure contrast, motion, text size and colour once, at system level. Content that ignores those settings makes them configure everything again, site by site.

Screen reader users read with the virtual cursor. Content that only responds to real keyboard focus is reachable but unreadable.

A change a sighted user sees at a glance is invisible to a screen reader user unless it is announced, so they act on a version of the page that no longer exists.

Unannounced focus movement is disorienting for sighted keyboard users and can leave screen reader users reading from an entirely different part of the page.

Being thrown into another app or device mid-task can drop assistive technology settings and abandon work in progress.

Texture behind text lowers effective contrast unpredictably and makes letter shapes harder to resolve, particularly for people with low vision or dyslexia.

Hierarchy expressed only by size collapses once a user sets their own text size - and it was never available to screen reader users in the first place.

Which alternative helps depends on the viewer. Control is what lets people sharing a screen, or with different needs, use the same media.

Search is what makes a long transcript usable for finding one fact, rather than a wall of text to read through.

Scrubbing a timeline precisely is difficult with a switch, a head pointer or a tremor. Chapters give structural navigation that does not depend on fine pointer control.

Background audio masks a screen reader's speech, making the page unusable until it stops - and the system volume control turns off both at once.