# AI website QA before client review

> Turn a repeatable website launch checklist into AI-assisted QA, then keep release approval with the team and client.

- Canonical: https://gptnavi.com/workflows/ai-website-qa-before-client-review
- Category: Coding
- Difficulty: Intermediate
- Setup time: 2 hours
- Estimated time saved: 3-8 hours
- Last materially updated: 2026-08-24
- Best for: Web teams, Agencies, Marketing teams, Product teams
- Tools: Superflow, Stagehand, Steel, Linear, Slack

## Quick answer

AI is useful for finding broken links, missing alt text, inconsistent copy, and key-flow failures. It should surface evidence, while people decide what is acceptable to ship.

## When to use it

- Site relaunch
- Campaign page
- Client handoff
- Weekly regression QA

## Steps

1. **Turn standards into a checklist** — Write the required pages, critical flows, accessibility checks, brand rules, SEO basics, legal copy, and explicit release owner. Tool: Linear. Expected output: A versioned launch checklist.
2. **Run broad content and page checks** — Ask review agents to scan the staging site for links, spelling, metadata, image text, and approval-sensitive content. Tool: Superflow. Expected output: A prioritized findings list.
3. **Automate critical paths** — Define a small number of stable, user-visible flows and run natural-language actions plus structured extraction against each one. Tool: Stagehand. Expected output: Repeatable critical-path checks.
4. **Capture replayable evidence** — Run the browser checks in isolated sessions and save screenshots or replays for failures that need reproduction. Tool: Steel. Expected output: Reviewable QA evidence.
5. **Triage and sign off** — Assign only verified findings, track fixes, and require a named human and client approver before production release. Tool: Slack. Expected output: An accountable go-live decision.

## Prompt templates

### Launch QA checklist

Create a website launch QA checklist for this site. Include critical user paths, responsive checks, accessibility, metadata, analytics, consent, broken links, copy accuracy, visual regression risk, approval owner, and release blockers. Site context: [paste]

### Finding triage

Review these automated QA findings. Separate release blockers, important follow-ups, false positives, and subjective feedback. For each blocker, state evidence, owner, verification step, and release risk. Findings: [paste]

## Common mistakes

- Treating AI findings as final truth
- Testing only the homepage
- Letting an automated check publish a production change

## Related workflows

- https://gptnavi.com/workflows/spec-first-ai-feature-delivery
- https://gptnavi.com/workflows/agentic-engineering-delivery-loop
- https://gptnavi.com/workflows/product-analytics-to-growth-experiments