GPTMap Article Maintenance Guide: Re-testing, Version Sync, and Sunset

Publishing is the starting line. Below is GPTMap's editorial standard for maintaining a published article — turning one-shot content into a durable, citable asset.

1. Two triggers for maintenance

Schedule trigger (six months)

Six months is the hard floor. Going longer creates two compounding risks:

• Code examples break: OpenAI rotates endpoints, SDK method names, parameter shapes — old code blocks no longer run as written.
• AI summary engines demote: stale softwareVersion / lastTestedAt tell ChatGPT/Perplexity the source is outdated; citation weight drops.

Event trigger (overrides schedule)

Three events, any one of which kicks off immediate re-test:

1. OpenAI releases a new major family (e.g. GPT-5.6)
2. A model or API field used in the article is deprecated / retired
3. A link in officialReferences goes 404 or is significantly rewritten

Event triggers outrank schedule triggers.

2. Re-test execution

The following sequence takes ~1 hour per article.

2.1 Re-read and produce a change list

Read the full article. List every claim that should have changed: price, version, API field, official reference. Read with skepticism — code you wrote six months ago probably doesn't run anymore.

2.2 Bump three frontmatter timestamps

If the OpenAI model version changed, also update openaiVersion — it auto-flows into the Article schema's softwareVersion, which AI summary engines use for freshness scoring.

2.3 Edit body + append update log

Apply the change list to the body. For large changes, rewrite a paragraph. Append a ## Update log entry at the end of the article:

## Update log

- 2026-08-08: GPT-5.6 family refresh — code examples re-tested, openaiVersion bumped to gpt-5.6
- 2026-02-12: Migrated links to developers.openai.com
- 2026-01-05: Initial publish

Newest first — gives readers a direct version signal.

2.4 Run the SEO/GEO mandatory checklist

Per .agents/skills/add-content step 8: title / description / canonical / robots / hreflang / OG / Twitter / JSON-LD (softwareVersion / applicationCategory / FAQPage / Breadcrumb / Person / Organization / mainEntityOfPage / image / citation / inLanguage / speakable) — all must emit. H1 = 1.

pnpm lint && pnpm build must pass.

2.5 Link health check

Crawl all outbound URLs:

# Example: check every URL in the post's frontmatter
grep -oE 'https?://[^"]+' apps/web/content/posts/<slug>.zh.md | sort -u | xargs -I {} curl -s -o /dev/null -w "%{http_code} {}\n" {}

Fix or replace 4xx / 5xx; fix wrong-redirect targets.

3. Sunset decisions

Don't delete articles — that creates 404s. Three options:

generateStaticParams is false by default, so to truly delete a page you must explicitly remove its slug from the code's static-params list.

4. The metadata signals your maintenance emits

4.1 Schema.org lastReviewed

When frontmatter has lastReviewedAt, GPTMap's Article schema emits a lastReviewed field. Google Quality Rater Guidelines treat "most recent human review" as an important E-E-A-T signal.

4.2 softwareVersion auto-updates

The openaiVersion field drives both:

• Manual annotation in frontmatter (you write it)
• softwareVersion in Article JSON-LD (schema emits it automatically)

Change it once, both update. That's why the SEO/GEO mandatory checklist includes "rendered HTML softwareVersion must match frontmatter openaiVersion byte-for-byte" — the consistency check catches drift.

5. Recommended rhythm

6. Common errors and troubleshooting

• Six months pass, frontmatter untouched → AI summary engines mark as stale; citation weight drops.
• Body updated, timestamps forgotten → schema doesn't change; search engines think you didn't maintain it.
• Link rot unnoticed → users click into 404s; SEO score drops.
• Article deleted outright → inbound link equity evaporates; future references dead-end.
• World-state changed, related articles not synced → same fact contradicts itself across the site.

7. What's Next

• The GPTMap editorial checklist: 12 steps to a publishable OpenAI article — the writing-stage counterpart
• OpenAI Models Release Notes (Living Document) — example of a maintenance target
• Deep Research prompt patterns — standard method for citation verification

Update log

• 2026-08-08: Initial publish