The scope questions every UX audit engagement negotiates, answered. Deliverables, methodology, framework selection, severity scoring, accessibility, presentation, timeline, pricing, remediation — the operational reference, not the marketing overview.
A defensible UX audit includes seven components: a signed scope agreement, an evidence pack (screenshots and session recordings per template), a heuristic evaluation against Nielsen's 10 or Google's HEART, a WCAG 2.2 AA accessibility layer, a prioritised findings list with severity scoring, a recommended remediation guide, and a shippable report format (executive summary, findings appendix, presentation deck).
Audits that skip any of these seven components produce output the client can't act on — the audit becomes a document rather than a change-driver. The methodology guide covers each component in operational detail.
Five deliverables in a shippable audit engagement. First, an executive summary — one page, top three findings, business impact, recommended remediation timeline. Second, a full findings list — every issue with WCAG or heuristic citation, severity score, screenshot evidence and remediation guidance. Third, a prioritisation matrix — impact × effort quadrants for direct roadmap conversion.
Fourth, a raw evidence appendix — automated scan outputs, manual review notes, AT test logs. Rarely read; essential to have when clients or auditors want to interrogate the methodology. Fifth, a stakeholder presentation deck — for the walkthrough session where the audit is presented live rather than emailed as a PDF.
The worked audit report example shows all five deliverables applied end to end.
Nielsen's 10 usability heuristics are the default framework for the usability layer. They've survived 30 years because they cover structural failure modes with little overlap and few gaps. The heuristic evaluation template provides an interactive scoring tool for the ten. For accessibility, WCAG 2.2 AA is the operational baseline in 2026 — referenced in UK PSBAR and in most private-sector procurement questionnaires. The accessibility audit checklist is scoped against WCAG 2.2 AA directly.
For conversion audits, add CRO-specific heuristics (Peep Laja, MECLABS, ConversionXL frameworks). Never invent a custom framework for a single engagement — auditors get pushback on personal opinion but rarely on framework citations. Named frameworks end debate.
Nielsen's 0-4 scale is the industry default. 0 — not a usability problem (drop from report). 1 — cosmetic (fix if time). 2 — minor (users work around it). 3 — major (must fix before next release). 4 — catastrophe (users cannot complete the task; data loss; accessibility exclusion).
The rubric must be published in the report so the client can interrogate any score. Most audits severity-inflate — everything gets marked 3 — which dilutes prioritisation and the roadmap becomes a flat list. Reserve high severity for around 10-20% of findings; if more than 30% score high, the calibration is wrong.
Accessibility is the highest-leverage category in any UX audit because the findings are objective (WCAG citations end debate), the findings carry legal weight (Equality Act 2010, PSBAR 2018, European Accessibility Act enforcement began June 2025), and the fixes disproportionately improve the product for all users.
Most UX audits include accessibility as a category alongside usability, forms, content and performance. For engagements where accessibility conformance is the primary deliverable — regulatory reporting, VPAT preparation, procurement responses — a dedicated accessibility audit against WCAG 2.2 AA is more appropriate than a general UX audit with an accessibility section.
A focused single-flow audit takes 3 to 5 working days end to end. A template-level audit covering 8 to 15 unique page templates takes 8 to 15 working days including AT testing. A full site audit with component library review, cross-device testing and remediation coaching takes 3 to 6 weeks.
Slippage almost always happens at scoping (stage 1) or presenting findings (stage 6) rather than during the audit work itself. Get scope signed off in writing before starting; get calendar time for the presentation session blocked before the report is delivered.
Independent senior consultants: £2,500 to £6,000 for a focused audit; £8,000 to £15,000 for a template-level audit with full AT testing. Boutique agencies: £10,000 to £30,000 for full audits including component library review and remediation guidance workshops. Enterprise accessibility consultancies: £25,000 to £75,000+ for full-programme audits with retest cycles.
Cheap audits — under £2,000 for anything above 10 pages — almost always mean automated scanning dressed up as a full audit. They miss 60-70% of real issues and produce false confidence. If the fee is meaningfully below market rate, ask specifically what proportion is automated versus manual review.
The audit should include remediation guidance for every finding but should not include the remediation work itself — those are two different engagements with different pricing and skill profiles.
A well-structured audit hands off cleanly to a remediation phase, often 3-10 times the audit's own effort, which can be delivered by the auditing team, the client's own team, or a specialist implementation partner. Clients who expect the audit fee to cover the fixes are working from a misunderstanding of what an audit is — clarify this in the scope agreement before the engagement starts.
Do not just send the document. Every audit should end with a live presentation to the stakeholders who will act on the findings — engineering, product, design leadership. The presentation opens with the top three highest-severity findings, walks through each with screenshot evidence, and closes with a prioritised roadmap.
Engineering will surface implementation constraints during the session; product will surface roadmap constraints; the recommendations get sharper and the team takes ownership. Audits that arrive as PDFs in shared drives without a presentation are the audits that don't change behaviour. Block calendar time for the walkthrough as part of the scope agreement.
Automated scanning: axe DevTools, WAVE, Lighthouse, Pa11y for the accessibility baseline. Contrast: the WCAG contrast checker or Colour Contrast Analyser. Assistive tech: NVDA on Windows, VoiceOver on macOS/iOS, TalkBack on Android. Behavioural evidence: Crazy Egg or Hotjar for heatmaps and session recordings on high-traffic sites. Structured audit: the 60-point UX audit checklist, the accessibility audit checklist, and the heuristic evaluation template as interactive working tools.
LLM assistants (Claude, GPT-4) accelerate content quality review at scale but require every AI-generated finding to be manually verified before shipping. The AI UX audit reference covers the tool stack in operational detail.
A UX audit is expert-led against frameworks — a senior practitioner evaluates the product against Nielsen's heuristics, WCAG success criteria, and behavioural best practice, producing a prioritised list of findings. A usability test is user-led against tasks — real users attempt scenarios while researchers observe and capture friction.
Audits surface known failure modes cheaply and fast; usability tests surface the unknown failure modes frameworks miss. Run audit first — it catches the obvious problems that would otherwise dominate the usability test — then test to discover what the audit couldn't see. The usability testing guide covers the test side end to end.
Not in 2026. AI accelerates specific audit stages — automated scanning, session-replay clustering, content-quality review, initial pattern extraction — but cannot replace the judgment layer. Severity calibration, cross-finding synthesis, remediation prioritisation, and client-context adaptation all require human expertise.
AI-only audits miss 40-60% of high-severity findings and misprioritise the remainder. AI-assisted audits (human auditor with AI acceleration on specific stages) are typically 20-30% faster than fully manual audits; they are not dramatically cheaper. The AI UX audit reference covers where AI genuinely helps and where it produces false confidence.