agentleFS
charanjit-singhcharanjit-singh/lovable-skills/skills/programmatic-seo-content-quality/SKILL.md

programmatic-seo-content-quality

Skill10 starsChanged 6 months agoMITView on GitHub

Use when designing the content layer for programmatic SEO pages: avoiding thin content, deduplication, AI generation guardrails, E-E-A-T signals, and quality thresholds. Not for one-off blog editing or full template design (use programmatic-seo-pages).

The file

We split this skill into the rules it gives an agent

8 rules. About 0 are common, 6 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈6

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    If not, do not publish that row.

    line 20
  2. Distinctive:

    If you can't hit this for a row, leave it unpublished.

    line 31
  3. Distinctive:

    If you use an LLM to draft prose:

    line 47
  4. Very specific:

    Human review: before published = true. Even a 30-second skim catches hallucinations.

    line 51
  5. Distinctive:

    Disclose AI: if your audience or regulation requires it.

    line 52
  6. Distinctive:

    Run a similarity check across pages before launch:

    line 57
  7. Distinctive:

    After 90 days in production:

    line 97
  8. Very specific:

    After 180 days with no recovery, set published = false and 301 redirect to the parent hub.

    line 101

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Programmatic SEO — content quality

Programmatic pages succeed or fail on whether each page deserves to exist on its own. Google's Helpful Content System actively penalizes thin, near-duplicate, doorway pages — and the penalty often hits the whole domain, not just the bad URLs.

Companion: programmatic-seo-pages, seo-meta-tags-spa, seo-structured-data.

The single rule

If a user landed on this URL and 9 nearly identical ones, would they find each useful?

If not, do not publish that row.

Per-page minimum bar

A programmatic page should have:

  • ≥ 300 words of substantive, page-unique content.
  • At least 3 unique data points (price, stat, testimonial, list item) that vary per row.
  • At least 1 UGC or hand-curated element (review, FAQ answered by a human, local note).
  • A reason a user would click your result over the existing top 3.

If you can't hit this for a row, leave it unpublished.

Data inputs — where uniqueness comes from

SourceExample
Public datasetsCensus stats, NOAA weather, Wikipedia openings
Your own dataAnonymized usage, top searches, customer outcomes
User-generatedReviews, Q&A, before/after photos
Computed"Compare X and Y on price, plan limits, ratings"
AI-augmented (with guardrails)Summaries of UGC, opening paragraphs

Mix at least three. A page built only from AI prose will read like every other AI page.

AI generation guardrails

If you use an LLM to draft prose:

  1. Feed it structured per-row facts in the prompt — never let it invent facts.
  2. Force a specific structure (intro, comparison table, 3 bullet pros/cons, 1 anecdote).
  3. Human review before published = true. Even a 30-second skim catches hallucinations.
  4. Disclose AI if your audience or regulation requires it.
  5. Re-generate on a schedule with refreshed data — don't ship 2024 prices in 2026.

Deduplication

Run a similarity check across pages before launch:

-- find suspiciously similar pages
select a.slug, b.slug, similarity(a.body, b.body) as sim
from public.location_pages a
cross join public.location_pages b
where a.slug < b.slug and similarity(a.body, b.body) > 0.80
order by sim desc;

(Use pg_trgm's similarity(); or compute embeddings + cosine if you need semantic dedup.)

Threshold: kill or merge any pair with > 0.85 prose similarity.

E-E-A-T signals

Show Experience, Expertise, Authoritativeness, Trustworthiness:

  • Author byline with photo, role, link to bio.
  • datePublished + dateModified visible and accurate (see seo-structured-data).
  • Citations for facts: link to the source.
  • Reviews / testimonials with names, dates, optional verification.
  • Contact / legal pages linked from footer (not orphaned).

Quality gates before publishing

Block published = true until each row passes:

  • ≥ 300 words page-unique text
  • No near-duplicate (similarity < 0.80) of another page
  • Title and meta unique
  • Hero image exists (real or generated for this row)
  • At least one human-reviewed paragraph
  • No invented facts (LLM output reviewed)

Enforce with a server check, not "the team will remember".

Pruning

After 90 days in production:

  • Identify pages with zero clicks + zero impressions (Search Console).
  • Set noindex = true and remove from sitemap.
  • After 180 days with no recovery, set published = false and 301 redirect to the parent hub.

This is the most important habit in programmatic SEO: cut what doesn't work.

Avoid

  • Shipping 5,000 rows on launch day. Start with 100 best; expand by 100/week.
  • Identical headline structure ("Best X in Y") with nothing else unique.
  • Auto-translated content without native review.
  • Stuffing locations into a single page (Atlanta, Boston, Chicago, …) hoping one ranks.
  • Using AI to invent reviews or testimonials (illegal in many jurisdictions; always against Google guidelines).

Checklist

  • Quality gate enforced at the published flag.
  • Deduplication script run before each batch publish.
  • Human review per row, recorded with reviewer + timestamp.
  • Monthly pruning of zero-traffic pages.
  • Author / source / date signals visible on every page.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.