Skip to main content
// service · AI-Built QA

AI-built and vibe-coded website testing

Independent QA for websites and apps built with AI tools.

AI-built website testing is independent, human-led QA for a website or web app created with AI coding tools such as Lovable, Bolt.new, Replit, Cursor, Claude Code or v0. We test it the way real customers, and curious outsiders, will meet it: public pages, real journeys and, with your permission, signed-in accounts and configuration. Every finding comes with evidence, a severity and a recommended fix.

We test projects built with Lovable, Bolt.new, Replit, Cursor, Claude Code, v0 and similar tools. BugFinder Labs is independent and not affiliated with any of these platforms.

// at a glance
AI-Built Website QA
Starts from
Your live or staging URL
Format
Evidence-backed report
Access
Agreed per test level
Pricing
Scoped USD quote
// quick answers

Short, factual, quotable.

Direct answers to the questions people (and AI assistants) actually ask about ai-built website qa.

What is vibe-coded website testing?

Vibe-coded website testing is QA for products built mainly by prompting an AI coding tool rather than writing code by hand. It focuses on problems these builds commonly ship with: secrets in front-end code, access control that exists only in the browser, forms and payments that only appear to work, leftover placeholder content, and mobile, SEO and accessibility gaps.

Can a public-URL audit prove my app is secure?

No. A public review can find exposed keys, debug pages, missing headers and broken journeys. It cannot confirm that one user is unable to read another user's data, or that your database rules are correct. That needs test accounts, read access to configuration, or an explicitly authorised security test. Every finding we report states the access level it was verified at.

Does it matter which AI tool I used?

The tool tells us where to look. Many Lovable and Bolt projects use Supabase, so row-level security matters; others use Firebase rules or a custom API. But every finding is verified against your deployed build, not assumed from the platform.

// who it's for

Built for teams that

  • Founders about to send real users or paid traffic to an AI-built MVP
  • Teams whose AI-built app already holds sign-ups, customer data or payments
  • Agencies and freelancers delivering AI-assisted builds to clients
  • Owners who want an independent check before trusting what the tool generated
// problems it solves

If any of this sounds familiar

  • It works in the demo, but nobody has tested what a second user can see
  • An API key, database key or admin page might be reachable from the public site
  • Forms say “Success!”, but you're not sure anything reaches your inbox or database
  • Sample data, placeholder text or invented testimonials made it to the live site
  • Nobody has checked mobile layouts, page titles, link previews or keyboard access
// what we test

Four access levels, from public pages to authorised testing.

P

Public website review

Part of every audit · browser-only, non-intrusive

  • Exposed keys and secrets in front-end code
  • Forms and checkout that only look finished
  • Mobile, SEO and accessibility gaps
  • Placeholder content, demo data and unsupported claims
  • Broken links, deep-link refreshes and missing-page handling
O

Signed-in testing

With test accounts you create, ideally on staging

  • Whether one user can see or change another user's data
  • Admin pages and actions protected on the server, not just hidden
  • Logout, session expiry and password reset
  • Payments: failed, cancelled and repeated events
S

Configuration review

By agreement, with read-only access

  • Supabase row-level security or Firebase security rules
  • Secrets in repository history and build settings
  • Handlers that always report success, TODO stubs and disabled checks
  • Payment webhook signature verification
T

Authorised security testing

Only with separate written authorisation · staging preferred

  • Injection testing (SQL and NoSQL) within an agreed scope and time window
// how we verify

Every finding says how it was verified.

A finding is only as strong as the access behind it. We label each one with the level it was verified at, so you can see what was proven and what still needs checking.

Read the full verification standard
  • Four access levels: Public (P), Owner test accounts (O), Source & configuration (S) and Authorised security testing (T).
  • Each finding has a severity (Critical, High, Medium or Low) and a confidence (Confirmed, Likely or Possible).
  • Checks that need access we didn't have are reported as not tested, never as passed.
  • Checks that don't apply to your product, such as payments on a site without checkout, are marked not applicable.
// checklist

20 high-impact checks, verified where they apply.

These are 20 high-impact checks from a broader, scoped audit. Which ones apply depends on your website, its features and the access available.
#CheckAccess
01No secrets in the shipped front-endP
02Row-level security is on and strictS
03Users can't reach each other's dataO
04Users can't change each other's recordsO
05Authorisation enforced on the serverS
06Private uploads stay privateS
07Users can't set protected fieldsO
08Admin routes are protectedO
09Logging out ends API accessO
10Costly actions are rate limitedO
11Handlers really write dataS
12Contact submissions arriveO
13Critical handoffs are actually wiredO
14Failed payments don't grant accessO
15Webhook signatures verifiedS
16Privacy policy covers visible third partiesP
17Account deletion and export workO
18Crawlers see real contentP
19Nothing clipped or scrolling sideways on phonesP
20Core journeys work from the keyboardP

Access: P public · O owner test accounts · S source & configuration · T authorised security testing. Representative examples, not the full list.

// what you receive

Evidence you can act on.

  • Findings with severity, confidence and the access level they were verified at
  • Reproduction steps, affected URL, device and browser for each finding
  • Evidence: screenshots, with request and response excerpts redacted
  • A recommended fix for every finding
  • A coverage statement: what was tested, what wasn't, and why
  • Any secrets we come across shown redacted, never in full

Re-testing of fixed findings is included only where it's agreed in your quotation.

See an illustrative sample report
// limitations

What this audit does not prove.

A public review is not a penetration test, a compliance certification or proof that your app is secure. It reports what we observed, at the access level we had. We'll tell you which risks remain unverified and what access would be needed to check them.

// how it works

From enquiry to report.

  1. 01

    Tell us about your project

    Share the public URL and what you built it with. Please don't send passwords, API keys or other secrets through the contact form.

  2. 02

    Agree scope and access

    We agree which pages, journeys and test levels are in scope. Signed-in, configuration and security testing need your written confirmation that you own or are authorised to test the product.

  3. 03

    Quotation and invoice

    You receive a USD quotation for the agreed scope, followed by an invoice.

  4. 04

    Access, if in scope

    You create least-privilege test accounts or read-only access, ideally on staging, and share them through the channel we agree, not the form.

  5. 05

    Testing and report

    We test at the agreed levels and deliver the report with evidence for every finding.

  6. 06

    Re-test, if agreed

    Fixed findings are re-tested where that's part of your quotation. Remove any test accounts and access once we're done.

// frequently asked

Common questions, straight answers.

Do you need my source code?

Not for a public review. For a configuration review we ask for read-only repository access or an export, never production passwords.

How do I give you access safely?

Create dedicated test accounts with the minimum permissions needed, preferably on staging. Share them through the channel we agree after scoping, not the contact form, and remove them when we finish.

What happens if you find an exposed key?

We don't use it. We tell you as soon as we've confirmed it so you can rotate it, and the report shows it redacted.

Is this a penetration test?

Not by default. Intrusive security testing happens only with separate written authorisation covering scope, environment and timing.

How is it different from the free Quick Scan?

The Quick Scan runs automated checks on public pages. This is a human-led audit that can include signed-in journeys and configuration, and labels every finding with the access level it was verified at.

How is it priced?

We quote in USD after agreeing the scope, then invoice. There's no online checkout for this service.

// related services

Often paired with

// ready when you are

Scope an AI-Built QA engagement today.

Tell us what you built. We'll confirm scope and access before any testing starts.