Accessibility testing tools: 10 free tools and what none can check
A demand letter or a buyer’s accessibility questionnaire arrives, and the first answer in the room is that the site passes a checker. The second question follows quickly: how did a failure get past it? Another checker would have found a different set of failures, and none finds them all. Accessibility testing software decides what can be decided from code, and a large part of WCAG needs a person with a keyboard and a screen reader.
This guide lists ten free accessibility testing tools, the job each one does and what it cannot see. It also shows, from W3C’s own data, how many WCAG 2.2 criteria have an automated W3C test rule, and the order in which a team uses the tools so that each result is trusted for exactly what it measures.
No accessibility testing tool decides WCAG conformance on its own
The standard’s own publisher is direct about this. W3C’s overview of evaluation says that “no tool alone can determine if a site meets accessibility standards” and that “knowledgeable human evaluation is required to determine if a site is accessible.” Its guide to choosing an accessibility evaluation tool sets out the limits a buyer has to plan around:
Tools cannot check all accessibility aspects automatically. Human judgement is required. Sometimes evaluation tools can produce false or misleading results. Web accessibility evaluation tools can not determine accessibility, they can only assist in doing so.
That passage is from Selecting Web Accessibility Evaluation Tools, and the makers of the tools say the same about their own products. WebAIM’s help page for WAVE states that “the absence of errors DOES NOT mean your page is accessible or compliant.”
The Department of Justice draws the same line in its 2022 guidance on web accessibility and the ADA. It says automated checkers and overlays “can be helpful tools” but that, like spelling or grammar checkers, they “need to be used carefully,” and it warns that a “clean” report “does not necessarily mean everything is accessible.” A clean scan is one part of the record a business keeps under the ADA; what the ADA requires of a website sets out the rest.
The gap is easy to see in a single criterion. A tool can find an image with no text alternative. It cannot judge whether the alternative that is present describes the image. The UK Government Digital Service named this case in its test of automated tools: potential barriers that tools noticed but that needed a person to check, “like whether alt text descriptions were accurate.”
Put one sentence into the test plan before any tool runs: a scan with no errors is recorded as “no automatically detectable failures in the pages scanned,” never as “conforms to WCAG.” That wording keeps a clean dashboard from becoming a claim the organization cannot support.
W3C’s ACT Rules automate tests for 21 of the 55 WCAG 2.2 A and AA criteria
W3C publishes Accessibility Conformance Testing (ACT) Rules: written test procedures for parts of WCAG, each with examples that a tool must pass, fail or mark inapplicable. The ACT Rules page states that the rules are informative and are not required for determining conformance. Its filters show how far they reach. Read on 3 November 2026, with approved and proposed rules for WCAG 2.2 Level A and AA selected, the page shows 84 rules covering 31 of the 55 success criteria.
Narrow the filter to rules with fully automated implementations, and the page shows 54 rules for 21 of the 55 criteria. Narrow it again to approved rules only, and it shows 35 rules for 17 criteria. The remaining 24 criteria at Level A and AA have no ACT rule of any kind.

View the data as a list
- 21 criteria: at least one ACT rule with a fully automated implementation (54 rules).
- 10 criteria: ACT rules, none with an automated implementation.
- 24 criteria: no ACT rule.
- Approved rules only: 35 automated rules for 17 criteria.
These counts are neither a ceiling nor a score. W3C’s implementation pages note that tools “often have test procedures not yet described using ACT Rules.” axe-core, for one, runs a rule for links distinguished by color under 1.4.1 Use of Color by default, and Lighthouse switches on axe-core’s rule for 2.5.8 Target Size (Minimum); both criteria are among the 24 with no ACT rule. In the other direction, a rule mapped to a criterion tests one condition of it, so a criterion with an automated rule can still fail in ways the rule does not look for. The image case shows the gap: a rule can confirm that a text alternative exists, and only a person can confirm that it is right.
The UK Government Digital Service tested automated accessibility testing tools against a page of planted defects, and its results point the same way. Its published results report that it “deliberately introduced 142 accessibility barriers to a page of content, and ran them through 13 automated tools.” The best tool fully detected 40%. The worst detected 13%. Counting the items a tool flagged for a person to check, the best reached 50%.

View the data as a table
| Tool | Barriers detected | Including items flagged for a person |
|---|---|---|
| SortSite | 40% | 40% |
| Tenon | 34% | 34% |
| AChecker | 31% | 35% |
| WAVE | 30% | 39% |
| aXe | 29% | 30% |
| Siteimprove | 29% | 36% |
| ASLint | 28% | 35% |
| FAE | 28% | 50% |
| Asqatasun | 25% | 47% |
| HTML_CodeSniffer | 20% | 34% |
| EIII | 17% | 17% |
| Google ADT | 17% | 17% |
| Nu Html Checker | 13% | 16% |
The test dates from 2017, and the results page was last updated on 13 April 2018, so the percentages describe those versions, not today’s. In the first run, every tool missed 42 of 143 barriers, among them “links identified by colour alone,” according to the GDS write-up. Engines have changed since, and axe-core now runs its check for color-only links by default, so the list records that test rather than today’s tools.
The decision rule for a buyer follows from W3C’s ACT implementation tables. A tool counts as “consistent” with a rule when it maps failures to the right success criteria and produces the expected outcome on every example; automated tools may answer “cannot tell” on some examples and still count. The figures are self-reported: W3C states that they are “derived from data published by the vendors” and that “the data is taken as-is, and is not verified by the W3C.” Ask each vendor which ACT rules its engine implements consistently, check the answer against the rule-by-rule report W3C links for that engine, and ask separately which checks the engine runs outside the ACT set.
Ten free accessibility testing tools cover four jobs
The tools below are free and each one is documented by the organization that maintains it. They fall into four jobs: rule engines that run in code and in continuous integration (CI), inspectors a person uses on a page, guided manual assessments, and the screen readers that real users run. A team needs at least one tool for each job, and no single tool covers all four.
| Tool | Job | Where it runs | What it cannot tell you |
|---|---|---|---|
| axe-core | Rule engine | Code, test suites, CI | Anything its rules do not test |
| Pa11y | Rule engine runner | Command line, Node, CI | Anything its chosen engine does not test |
| IBM Equal Access Accessibility Checker | Rule engine and inspector | Chrome, Firefox, Edge, Node | Judgment calls such as meaningful alt text |
| Google Lighthouse | Inspector and score | Chrome DevTools, command line, Node | Its score leaves out manual checks |
| WAVE | Inspector | Chrome, Firefox and Edge extensions | Whether the page is accessible |
| ANDI | Inspector | Bookmarklet in desktop browsers | The outcome of a full manual test |
| Accessibility Insights for Web | Guided assessment | Chrome and Microsoft Edge | How a screen reader announces the page |
| NVDA | Screen reader | Windows | Behavior in JAWS or VoiceOver |
| VoiceOver | Screen reader | macOS | Behavior on Windows readers |
| TalkBack | Screen reader | Android | Behavior on iOS or desktop |
axe-core. An open-source rule engine distributed under the Mozilla Public License 2.0, with rules for WCAG 2.0, 2.1 and 2.2 at Levels A, AA and AAA plus a set of best-practice rules. Its WCAG 2.2 rule, for target size, ships switched off and has to be enabled. W3C lists version 4.14.0 as consistent on 30 ACT rules. Best-practice results are advice, not WCAG failures, and the guide to which axe-core results are WCAG failures separates the two before a report goes out.
Pa11y. A command-line runner for CI. Its documentation shows it running HTML_CodeSniffer (htmlcs) by default, with axe as a second runner, and a threshold that fails a build: the documented example returns exit code 2 on a page with ten or more errors.
IBM Equal Access Accessibility Checker. Browser extensions for Chrome, Firefox and Edge, plus a Node package that, in IBM’s words, runs “automated accessibility testing within a continuous integration pipeline” and can “validate results against baseline files.” W3C lists it as consistent on 21 ACT rules.
Google Lighthouse. It runs in Chrome DevTools, from the command line, or as a Node module. Its accessibility gatherer runs axe-core inside the page with its own selection of rules: those tagged for WCAG 2.0 Level A and AA, minus a few it switches off, plus a set it switches on, target size and several best-practice rules among them. The scoring documentation says the score “is a weighted average of all accessibility audits” and that manual, low-impact and best-practice audits do not affect it.
WAVE. WebAIM’s inspector presents the page with embedded icons and indicators, and its browser extensions for Chrome, Firefox and Edge run entirely in the browser. WebAIM’s help page notes that WAVE “is most effective when used by someone knowledgeable about web accessibility” and that “people who are not web accessibility experts can also benefit from WAVE.” WebAIM runs the same engine for its annual analysis of one million home pages.
ANDI. A free bookmarklet from the Social Security Administration that will “reveal what a screen reader should say for interactive elements” and “give practical suggestions to improve accessibility and check 508 compliance.” It is open source.
Accessibility Insights for Web. Microsoft’s extension for Chrome and Microsoft Edge has two main modes. In its overview, FastPass “helps developers identify common, high-impact accessibility issues in less than five minutes,” and Assessment walks a person through 24 tests against WCAG 2.2 Level AA.
NVDA, VoiceOver and TalkBack. NVDA is a Windows screen reader that NV Access describes as “100% free to use globally.” Apple’s VoiceOver is “the screen reader built into macOS,” and Google’s TalkBack is “the Google screen reader included on Android devices.”
Use the table as the first page of a test plan: one named tool for each of the four jobs, with an owner beside it. The free versions of these web accessibility testing tools are enough to start. Judge a paid product by what it adds to these four jobs, such as reporting or scale.
Two scanners on the same engine give one opinion
A team that runs Lighthouse and an axe browser extension, and reads agreement between them as confirmation, is counting one engine twice. Lighthouse’s source code calls axe.run on the page, and its package file pins axe-core 4.14.0, so a Lighthouse audit and an axe scan run largely the same rules from one engine, chosen slightly differently.
A second engine adds information only where its rules differ. W3C’s implementation pages list which ACT rules each engine implements consistently: 30 for axe-core, 21 for IBM’s checker, 23 for QualWeb and 27 for the fully automated configuration of Alfa. Compare the lists rule by rule, because totals do not show overlap. The GDS write-up found that “using tools in combination could help you pick up more barriers, but doing this can be harder and less cost-effective for teams.”

A photo gallery on a contractor’s home page shows where two engines part. One photo carries the text alternative “0M1A0299.jpg”, an image file name. axe-core’s image rule passes it, because an alternative exists. WAVE raises a “Suspicious alternative text” alert, because its algorithm flags an alternative that “is an image file name,” and W3C lists file names used as text alternatives as failure F30. Neither tool can say whether a better text describes the photo; that judgment goes to a person, which is how the GDS test recorded it.
Set the rule in writing. Run one engine as the gate in CI, with its name and version recorded in every result. Run a second engine on a schedule as a cross-check, not on every build. Treat agreement between tools that share an engine as a single result, and never count it twice in a report.
Screen reader testing starts with the readers people use
Screen readers are accessibility testing tools as well, and they are the only ones on the list that behave the way a blind user’s software does. WebAIM’s Screen Reader User Survey #10, run in December 2023 and January 2024, received 1,539 responses. Asked for their primary desktop or laptop screen reader, 40.5% named JAWS, 37.7% NVDA and 9.7% VoiceOver, and in North America JAWS led NVDA by 55.5% to 24.0%. WebAIM notes that “the sample was not controlled and may not represent all screen reader users.”
For a product with US users, that puts JAWS in the plan beside NVDA. NVDA is free on Windows and makes a quick first pass, VoiceOver ships with macOS, and TalkBack ships with Android for mobile checks.

A keyboard needs no tool at all. On another contractor’s site, the “Schedule free estimate” button in the header takes focus from the Tab key and looks exactly as it does without it, which leaves a keyboard user unable to see where they are on the page.
Write the minimum matrix into the plan: keyboard only on every key task; JAWS and NVDA on Windows and VoiceOver on macOS for the same tasks; TalkBack for a mobile web or Android flow. Record which reader, which browser and which version produced each finding, because a defect report without them cannot be reproduced.
Website accessibility tools for monitoring answer a different question
Crawlers and dashboards are website accessibility tools of a different kind. They run an engine across hundreds or thousands of pages and track the count over time. That answers the question “are the errors a machine can see going down?” It does not answer “can a person complete the task?”
On home pages, the errors a machine detects are the same few year after year. The WebAIM Million, its February 2026 automated analysis of one million home pages, reports that “these most common errors have been the same for the last 7 years” and that 96% of all detected errors fall into six types: low contrast text (83.9% of pages), missing alternative text (53.1%), missing form labels (51%), empty links (46.3%), empty buttons (30.6%) and a missing document language (13.5%).
A monitoring platform can track those six counts down, and four of the top ten WCAG failures on business websites overlap with them: alt text, contrast, forms and links. WebAIM states the limit of its own method: “Absence of detected errors does not indicate that a page is accessible or conformant.” Monitoring belongs beside accessibility regression testing in the release process, as described in this guide to accessibility monitoring.
Before buying a monitoring platform, ask five questions and keep the answers in the contract file:
- Engine. Which rule engine and which version run the scans?
- ACT consistency. Which ACT rules does that engine implement consistently, as listed by W3C?
- Standard. Which WCAG version and level are the results mapped to?
- Evidence. Can each finding be exported with its criterion, page, element and date?
- Manual items. How does the platform mark the checks it cannot decide, and who closes them?
A test stack by stage keeps each tool on its own job
The four jobs map onto the stages of delivery. Design is checked for contrast and focus styles before anyone writes code. During the build, a rule engine runs in component and integration tests. In CI, one engine with a stored baseline fails the build on new errors. In quality assurance (QA), a person runs Accessibility Insights Assessment, WAVE or ANDI on new templates. Then the key tasks are run with a keyboard and the screen readers in the matrix, and an audit against a named WCAG version closes the loop with a retest.

View the data as a list
- Design: contrast and focus styles checked by a person.
- Build: a rule engine in component and integration tests.
- CI gate: one engine with a stored baseline that fails the build on new errors.
- QA: Accessibility Insights Assessment, WAVE or ANDI on new templates.
- Assistive technology: keyboard, JAWS, NVDA, VoiceOver and TalkBack on the key tasks.
- Audit and retest: people testing against a named WCAG version, with a retest after fixes.
Federal work adds a defined method at the audit stage. Section508.gov describes the DHS Trusted Tester Process as “a manual test approach that aligns with the ICT Testing Baseline,” and states that “agencies that adopt the Trusted Tester Process only accept test results from individuals who have been certified as Trusted Testers.” The ICT Testing Baseline, maintained by the ICT Testing Baseline Working Group and hosted by the Access Board, is “a benchmark for test processes that determine Section 508 conformance,” and its page states that “the Baselines are mapped only to Section 508 (and WCAG 2.0) requirements.” A federal test, in other words, is measured against WCAG 2.0 Level A and AA, whatever version a scan reports. How that testing works for Section 508 is set out in what Section 508 compliance testing checks.
The stage view also explains why the old debate between automated and manual testing has no winner. The two methods answer different questions, and the comparisons of manual vs automated accessibility testing and what automated testing catches and misses set out the trade-offs in detail. The full audit, step by step, is in the digital accessibility audit guide for ADA, WCAG and Section 508.
Use the stages as a checklist. For each one, name the tool or method, the owner and the record it produces, and file every finding in the same accessibility bug report template so that a scan result and a screen reader result read the same way to a developer.
Your next step
List the accessibility testing tools the team already runs and the job each one does. Where two share an engine, keep one as the CI gate and move the other to a scheduled cross-check. Then book time to run the three tasks that matter most to revenue with a keyboard and two screen readers, and record what a scan said about the same pages.
Two decisions sit with others. The WCAG version and level the organization is held to come from its regulation, contract or solicitation, and counsel reads those; the map is in which WCAG version each US rule requires. Whether a set of test results is enough evidence for a contract is the buyer’s or the agency’s call. A free checker, whose score is explained in this guide to ADA compliance checkers, cannot settle either question, and the factors behind an audit’s cost depend on both.
ADA Compliance Pros starts where the tools stop. As part of its web accessibility services, a person works through the site with a keyboard and a screen reader on the flows that carry revenue, with JAWS, NVDA and VoiceOver, keyboard-only operation and magnification run against the pages and components agreed in scope. Every issue is tied to the WCAG success criterion it fails, with its location and how it was observed, and the findings come ranked by how badly they block a user and how quickly a reviewer will find them. After the fixes ship, each issue is retested and its status updated. Bring the scan results you already have to a call, and the work will be scoped with you there as a WCAG audit and testing engagement.