---
name: local-business-page-qa
version: 1.0.0
description: QA and improve generated local business/prospect website batches so they are plausible buyer-targeted, customer-facing, and honestly scored.
triggers:
  - local business page QA
  - prospect website batch
  - generated Orleans pages
  - make these demo sites good
  - score these pages
  - replace bad-fit local business targets
---

# Local Business Page QA

Use this when reviewing or improving batches of generated local business/prospect websites, especially for AIBB-style owner-operated local business demos.

## Core standard

A passing page should look like a real customer-facing business website the owner could review and potentially launch after confirming facts/photos. It should not read like an AI demo, internal QA note, audit, proof artifact, outreach campaign, or template scaffold.

## Target-fit first

Before polishing pages, remove or flag bad-fit targets.

Bad-fit for local website/service sales:
- national chains and major brands
- big-box retailers
- national banks
- telecom chains
- pharmacies/gas/convenience chains
- brands unlikely to buy a small local website/service package

Replace with:
- owner-operated local businesses
- Chamber-listed local businesses
- independent restaurants, shops, salons, service providers, lodging, markets, galleries, marine/service businesses
- prospects with enough public facts for safe copy

## QA sequence

1. Inspect the target list before scoring page quality.
2. Replace bad-fit national/major-brand targets before spending polish cycles.
3. Patch shared templates/components first: hero, CTAs, metadata, schema, sanitizer, owner note placement, mobile/readability.
4. Move proof/source/weak-source/internal notes out of customer-facing body copy.
5. Add useful customer sections, not hidden scoring text:
   - useful customer questions
   - category-specific decision language
   - local decision context
   - phone/directions/booking/menu/order/shop/quote path where available
6. Patch weakest individual pages only after global fixes.
7. Build, crawl, screenshot if needed, and rerank after the final patch.
8. Report numeric score and honest launch-readiness separately.

## Strict scoring signals

A strict score should look at:
- target fit: likely local buyer vs bad-fit chain/major brand
- readable desktop/mobile hero
- contrast and layout
- business name + concrete category/location in hero
- lead copy that reads like a customer homepage
- category-specific vocabulary
- local specificity
- useful CTAs
- metadata
- schema where practical
- owner/business verification note near bottom
- no internal/demo/proof/source/outreach/campaign/template language
- no invented facts
- weak-source handling
- screenshot/proof completeness

## Avoid false-green scoring

If Sam challenges the grade, do not defend the old rubric. Make the rubric harder and inspect why it over-credited pages.

Common false greens:
- metadata/schema/CTA pass but the page still sounds like a verification draft
- “owner verification” dominates hero/lead copy
- generic category keywords inflate score
- national-brand pages look clean but are bad sales targets
- all pages pass scans but screenshots/copy still feel templated

## Customer-copy pitfalls

Avoid these visible phrases/classes:
- deserves a site
- verify / verification-heavy lead copy
- current details
- details to confirm
- source notes / public source / weak-source
- demo / proof / AIBB / campaign / outreach
- page should / site should
- generic brochure / placeholder
- AI-built / internal strategy / audit

Owner/business review language is acceptable only as a small bottom note.

## Reporting

Use concise proof-backed reporting:

```text
Done:
- ...

Proof:
- Build: ...
- Crawl: ...
- Ranking: ...
- Screenshots: ...

Honest caveat:
- A by scoring gate does not mean owner-launch-ready. Launch-ready still needs owner-confirmed hours, services, prices, photos, and approval.
```

## References

- `references/aib230-orleans-quality-gate-lessons.md` — detailed lessons from the Orleans 100-page scoring/target-fit session.