Known limitations 

Automated testing does not replace a manual audit 

Automated tools can only detect problems that are machine-detectable. Depending on the site, automated checks typically surface a minority of the barriers that a full audit finds. The following always require human judgement and cannot be decided by AQG:

• whether alternative text actually describes the image in its context,
• whether link text is meaningful for the target,
• reading order, focus order and keyboard operability,
• visible focus indication and interaction states,
• correct use of headings as a document outline rather than as styling,
• whether colour is the only carrier of meaning,
• whether captions, transcripts and audio descriptions exist and are correct,
• error prevention, error identification and form usability in practice,
• behaviour with screen readers, magnification and speech input,
• plain language and comprehensibility.

Use AQG to remove the mechanically detectable problems early and continuously, and plan a manual accessibility audit, including assistive technology testing, for the conformance statement itself.

Some rules encode best practices and are not a hard WCAG failure in every context. Rules with the severity Needs review are explicitly undecidable by the automated check. Review every finding in its context.

Content scan 

• Only fields that are discovered from TCA and enabled in the settings are analysed, see Scanned fields. Run Re-scan TCA after TCA changes.
• The content scan works on stored records. Problems that only appear when TYPO3 assembles the page — duplicate id values across content elements, a missing main landmark, a missing lang attribute — are found by the rendered page check, not by the content scan.
• Content produced by extensions that do not store its text in scanned TCA fields is not covered.
• The phrase lists of the text-based rules are English by default. Configure the dictionary settings for other languages, see Site settings (site sets).

Rendered page check 

The rendered page check analyses server-rendered HTML only. It does not execute JavaScript, does not wait for AJAX or lazy-loaded content, does not interact with cookie banners, does not take screenshots and does not run axe-core. It evaluates one page at a time and never follows links: a site scan repeats the check per page instead of crawling.

It is skipped for changed-only runs, for subtree CLI and Scheduler runs, for page doktypes that do not deliver a frontend page, and when it is disabled in the ruleset settings.

Frontend scan 

• The crawler requests the site from the public internet. Installations behind a VPN, an IP allowlist or a bot filter cannot be scanned.
• Free Remote Preview is limited to a small daily allowance of single-page scans and does not include screenshots, TYPO3 record mapping, scan history or PDF export. The extension does not define the allowance itself: the scans and pages used, the remaining quota and the next reset time are reported by the AQG service and displayed in the module.
• Licensed scans are limited by the page budget of the plan.
• Only one remote scan per site can run at a time.
• Mapping crawler findings back to TYPO3 records requires the AQG frontend markers, see Site settings (site sets). Without them, findings are reported per URL.

Quality gate 

• The gate evaluates stored findings. A page that has not been scanned since its last change is evaluated against outdated data. Schedule regular scans, see Automation.
• Blocking mode requires an active licence. Without one, the gate can only warn.
• The gate is an editorial safeguard. It is not a security control and not a statement about the accessibility of the published page.

AI suggestions 

Suggestions are proposals that an editor must review. AQG rejects unsafe outputs, but it cannot verify that a suggested text is factually correct for the image or link it describes. Suggestions are never applied automatically and never write to RTE bodytext.