SpamBrain rules — what pseolint detects.

Per-template aggregation — how rules feed verdicts

In v0.6, the engine audits by template rather than by URL. Phase 1 detects URL templates (filter ≥1% of URLs, ≥5 URLs, ≥2 survivors after deduplication). Phase 2 samples K=10 URLs per template and runs all 32 rules. Each rule's output per template is summarised as a uniformity score (0–1) and a top driver — the single rule responsible for the most findings on that template. The site verdict is determined by siteVerdictFromTemplates: the worst template that covers ≥5% of the site's URLs (spec §15.1). Three aggregation patterns apply:

• Per-page → template uniformity score (most spam/* and content/* rules). The rule fires on each sampled page; the fire rate becomes the template's score for that rule.
• Corpus-wide (spam/near-duplicate). Computed across all sampled pages regardless of template — it surfaces cross-template duplication, not just within a single template.
• Per-page → template-level signal (aeo/* rules). A rule that fires on 8/10 pages of a template reports one template-level finding, not 8 individual URL findings.

How the rules map to SpamBrain

The rule set clusters around the major axes Google's SpamBrain classifier scores against. Spam/* (8 rules) covers the patterns the March 27, 2026 core update demotes most aggressively — the most recent classifier shift to hit pSEO, tightening scaled-content signals on date-stacked corpora — building on the March 5, 2024 scaled-content-abuse update that first targeted thin content under 300 words, doorway clusters with shared boilerplate, near-duplicate templates with >85% lexical overlap, templates that don't vary their structural skeleton, and corpus-aware publication-velocity (the threshold scales with corpus size, so a 50,000-page directory and a 50-page blog get appropriate cutoffs). Content/* (4 rules) checks unique value, meta-description uniqueness after entity masking, author signals, and E-E-A-T markers. Aeo/* (8 rules, shipped April 21, 2026) audits answer-engine readiness — citable facts, atomic Q&A blocks, freshness signals, AI-crawler access, and the things Perplexity, ChatGPT, and Google's AI Overviews actually extract.

The remaining categories are links/* (6 rules — orphan pages, dead ends, cluster connectivity, link depth, unreachable-from-root, and host-section-divergence — the last one detects sub-sections that ride a host's reputation without integrating into it, which is the May 2024 site-reputation-abuse policy target), tech/* (4 rules — canonical consistency, sitemap completeness, soft-404, and redirect chains), schema/* (3 rules — JSON-LD validity, required-fields by type, and cross-page consistency), data/* (2 rules — missing-binding and identical-across-pages, fired when --data-source is set), and cannibal/* (1 rule — url-pattern; title-overlap and keyword-collision were dropped in v0.4 due to high false-positive rates).

What makes a rule "AEO-aligned"

The 2022 SpamBrain rebuild changed what enforcement looks like — instead of waiting for a manual reviewer to hit a domain with a policy action, the classifier silently suppresses pages it scores as spam-like at query time. That means the old "wait for the manual action notice" playbook is dead; you have to anticipate the scoring. An AEO-aligned rule is one whose detection logic also predicts whether AI Overviews and answer engines will cite the page — because the same signals (entity grounding, citable facts, atomic structure, schema integrity) drive both classical ranking and extraction by LLM-powered SERPs.

The full rule registry is open source at github.com/ouranos-labs/pseolint, and Google's underlying spam policies are documented at developers.google.com/search/docs/essentials/spam-policies. Every rule in pseolint links back to the specific policy paragraph it implements, so you can see exactly which Google guideline a finding maps to.