Research methodology
How we built the 1,450+ UX rule database.
UXAudit.Now launched in 2025, but the methodology behind it goes back 15+ years. This page documents exactly where the 1,450+ rules come from, how new ones get added, and how we keep the database current.
How the 1,450+ rules break down per platform
Each platform has its own rule set, sized to the surface. E-commerce is the deepest (drawing on Baymard's published research + our own prior consultancy); Conversational AI is the newest and fastest-evolving.
E-Commerce
700+ guidelinesCurated from Baymard Institute's published e-commerce research (the industry gold standard), cross-validated with patterns our team observed across years of enterprise commerce engagements — checkout flows, PDPs, search & filters, account flows, post-purchase.
SaaS & Web Apps
250+ guidelinesBuilt from NN/g heuristic frameworks + dashboard / onboarding patterns we've audited across B2B SaaS products. Skews toward B2B SaaS; consumer apps are a smaller subset of the database today.
Corporate
200+ guidelinesDrawn from corporate-site audits we've delivered to enterprise brands (Eczacıbaşı, Dynavit, and others) — first-impression clarity, information findability, contactability, IR signal hygiene.
Landing Pages
200+ guidelinesSynthesized from conversion-optimization research (message-match, single-CTA hygiene, social proof placement) and from our team's prior paid-traffic landing-page consulting engagements.
Conversational AI
100+ guidelinesThe newest and fastest-evolving rule set. Built from emerging LLM UX research and from patterns observed in production conversational AI products throughout 2025. Updated monthly as the surface evolves.
Four principles behind every guideline
Every rule traces to a source we can name
We don't add a guideline because a designer said it should be there. Every rule in the database cites its origin — Baymard, NN/g, WCAG, CrUX, or our own prior UX consulting engagements with enterprise clients. No invented heuristics, no AI-fabricated rules.
Severity is calibrated by impact, not opinion
Critical / High / Medium / Low severity tiers are calibrated by observed business impact (conversion delta, task-failure rate, support-ticket generation) — not by how visually unappealing a pattern is. A subtle finding with measurable lift outranks a loud finding that doesn't move metrics.
Platform-specific guidelines, not flattened heuristics
A bad checkout pattern isn't the same problem as a bad dashboard empty state. Each platform has its own guideline set tuned to the surface — 700 for e-com, 250 for SaaS, 200 each for Corporate and Landing, 100 for Conversational AI. Generic 'UX heuristics' don't substitute for surface-specific research.
Continuous updates, not annual revisions
The Conversational AI set sees weekly additions as the surface evolves. E-commerce and SaaS see monthly updates. Stable surfaces (Corporate) update quarterly. Every guideline carries a 'last reviewed' date and a research summary so the trail is visible.
How a new guideline gets into the database
Recruit a representative panel for the surface
Per surface (checkout / dashboard / landing / etc.), we recruit a panel matched to the actual user demographic — geo, role, prior experience. No 'tech-savvy generic users'; specific cohorts for specific surfaces.
Observe task-completion sessions
Moderated and unmoderated sessions where users attempt realistic tasks on real production sites (theirs and competitors'). We measure time-on-task, drop-off points, hesitation, verbalized confusion, and ultimate task success.
Cluster observations into patterns
After enough sessions on a surface, observed struggles cluster around repeatable patterns — e.g., 'guest checkout buried below account-create reduces completion by X%'. Patterns become guideline candidates.
Validate the pattern in independent sessions
A candidate guideline is validated by checking it against fresh, independent sessions before being added to the database. If the pattern doesn't reproduce, the guideline doesn't ship.
Map to severity + write the recommended fix
Validated patterns get a severity tier (calibrated against business-impact data we've collected on the cluster) and a recommended-fix pattern. Both go into the guideline record alongside the research summary.
Continuous review + retirement
Guidelines that no longer match observed reality are retired or updated. We don't ship a stale database — the Conversational AI set particularly evolves with the LLM UX landscape.
Sources we cite and cross-validate
Original research isn't all of it — we lean on the standards-bearers of the field and cross-validate every overlap. Honest attribution beats invented novelty.
Baymard Institute
Foundational e-commerce UX research. Our e-commerce guideline derivations cite Baymard heavily; we cross-validate every overlap.
Nielsen Norman Group (NN/g)
Heuristic frameworks, foundational usability principles. Where NN/g and Baymard converge, we anchor; where they diverge, we test.
WCAG 2.1 / 2.2
Accessibility rules surface through the axe-core Smart Plugin and through manual UX Review guidelines for visual hierarchy + content readability.
Google Chrome UX Report (CrUX)
Real-user performance data — surfaces in Smart Plugins for every audit, layered alongside synthetic Lighthouse-equivalent scores.
Our team's prior UX engagements
Before UXAudit.Now launched in 2025, the team spent 15+ years auditing UX for enterprise brands — Eczacıbaşı, Gratis, Dynavit, D-Smart, and many more. Recurring patterns from those engagements feed the database alongside the published research above.
Research FAQ
Can I see the research behind a specific finding in my audit?
Yes on Pro and Enterprise — each finding shows the guideline ID and a research summary. Click through to the methodology page (this one) for the full framework. We don't publish the raw session transcripts (privacy of the recruited panel), but the synthesized patterns are visible.
Is the database open source?
No — the database is proprietary, and it's what makes UXAudit.Now different from generic AI scans. The methodology (this page) is public; the specific guideline corpus is what you subscribe to access.
How does this compare to Baymard's research?
Baymard is the gold standard for e-commerce UX research. We cite them and cross-validate every overlap. Where we go further: SaaS (250 guidelines), Corporate (200), Landing (200), and Conversational AI (100) — surfaces Baymard doesn't cover.
Who decides what becomes a guideline?
Our research team. Senior UX researchers review observed patterns, validate them against independent sessions, and assign severity. Subject-matter experts (e-commerce, SaaS, conversational AI) sign off on platform-specific additions.
Can I contribute findings from my own research?
Customers on Enterprise can contribute observed patterns; we review and may incorporate them with attribution (and credit). Talk to us if you have ongoing UX research you'd like to feed back into the database.
Stop guessing. Start auditing.
Run your first UX audit in 5 minutes. No credit card. 30 free credits to explore every platform — SaaS, E-Commerce, Corporate, Landing pages, and Conversational AI.
Run a free UX audit
