Website audit: findings tied to the page

A bounded audit of grex-ai.com, three evidence-linked findings, and the deployed fix that closed each one.

Objective

Inspect a defined website scope and return findings a reader can check against the actual page, not generic advice.

Input scope

grex-ai.com, five public pages (home, how-it-works, product, pricing, interest), desktop and mobile renders. Public read-only checks only — no login, form submission, transaction, or accessibility/security certification.

Working editorial angle

Can an AI team turn a defined site review into an action list whose claims a reader can check — and does the resulting fix actually land?

The final headline follows the evidence.

Case-study record

What the published record will show

Each field separates what Grex was asked to do from what it actually produced and what the evidence supports.

Objective

What the run was meant to establish.

Inspect grex-ai.com across five public pages, on desktop and mobile, and return reproducible, evidence-linked findings.

Input scope

The sources, boundaries, exclusions, and time window supplied to the work.

Five pages on grex-ai.com (/, /how-it-works/, /product/, /pricing/, /interest/), captured 2026-09-16 07:10 UTC, both viewports. No login, forms, transactions, or accessibility/security certification.

Actual deliverable

The work product that was actually handed back.

An audit_report with three medium-severity "cta" findings — home, how-it-works, and pricing each lacked a link or button with a clear action label — each with a page, viewport, repro step, proposed change, and a verification criterion.

Source / effect evidence

The evidence supporting the finding and any observed effect.

Each finding cites a page capture and an independent verifier pass that reopened the flagged page before the finding was accepted. No CTA quota was invented — /product/ and /interest/ were checked and cleared with no finding.

Disposition

The decision made from the evidence.

Accepted and fixed. A "Tell us you're interested →" primary call to action, linking to /interest/, was added to exactly the three flagged pages. Live-checked after deploy: all three return HTTP 200 with the exact CTA text and link present.

Elapsed time

How long the bounded run took from start to settled outcome.

Mission window: 2 hours (2026-09-16 07:01–09:02 UTC). The audit_report was produced 9 minutes after the mission started; the fix and its live verification were separate, subsequent engineering work.

Metered cost + pricing gaps

What was metered, what was not priced, and which cost basis applies.

$0 of a $1 metered spend cap on the audit mission itself. The subsequent fix and deploy ran on the operator's existing engineering subscription, not separately metered.

Models / runtime used

The models, runtime, and compute context that produced the work.

The website-audit Manipulus's browser-capture worker and an independent verifier actor that reopens each flagged page before a finding is accepted, running on Grex's Timmy Nidus.

Operator interventions

Human approvals, corrections, reruns, or other changes during the run.

None during the audit mission itself. The fix that followed the finding was authored and deployed as separate engineering work, then independently re-checked live against the finding's own verification criterion.

Known limits

What the evidence does not establish.

This is a presence check, not a conversion-rate claim — it establishes that a CTA now exists where one did not, not that it changes visitor behavior. No accessibility or security certification is claimed.

Publication gate

This case study remains unpublished as a claim until the actual demo output, source links, cost record, and limitations are reviewed. A missing deliverable stays missing in the final account.

Request a walkthrough