Template-to-Page QA Sampling (QA)

Template-to-Page QA Sampling (QA)

1. 01

   Freeze the template schema before any batch render runs

   Lock the field list (city, service, price band, review count, inventory status) and version it. Any field added mid-batch invalidates comparison across pages and hides duplicate-content patterns.

2. 02

   Draw a stratified sample of 30 to 50 generated pages per batch

   Split the sample across top-traffic cities, mid-tier locations, and the thinnest data rows. Random sampling alone misses the low-data pages where thin content concentrates.

3. 03

   Score each sampled page on unique-data density, not word count

   Flag any page where fewer than three fields carry client-specific values (real pricing, live inventory, verified reviews). Generic AI filler around a swapped city name is the failure mode that triggers algorithmic suppression.

4. 04

   Run a near-duplicate check across the full batch, not just the sample

   Compare body text similarity across sibling pages and set a threshold (for example 70 percent overlap) that forces a rewrite or a noindex decision before publication.

5. 05

   Verify indexation and answer-engine citation status after 14 days

   Pull Search Console coverage for the batch and query ChatGPT, Perplexity, and Google AI Overviews with the target long-tail phrasing. A page that ranks but is never cited in AI answers is a different defect than one that never indexes.

6. 06

   Log every defect against the template version that produced it

   Attribute failures to schema fields rather than individual pages so a single fix corrects the whole batch. This is what makes the retainer defensible when a client asks why 4,000 pages underperformed.

7. 07

   Hold the next batch until the defect log is cleared or explicitly accepted

   Publishing volume on top of an unresolved template defect multiplies the cleanup cost. Document accepted risks in writing so the client owns the tradeoff.