← Back to overview

Socrates Documentation

Quality & GEO Standards

The 2026 SEO/GEO bar and the scored anti-slop gate.

GEO Quality Standards (2026)

The quality bar every piece must clear, plus the standards behind it. Current as of 2026; the search landscape moves fast, so treat the specifics as a living standard.

Why this matters (the stakes)

Ranking no longer equals being cited: the overlap between top Google results and AI-cited sources has collapsed (one estimate, ~70% down to under 20%), and the bulk of AI-Overview links now come from pages outside the organic top 10. And readers punish AI-feeling copy — roughly half disengage when they suspect content is AI-generated, and some platforms suppress its reach. So quality is not cosmetic; it's whether the piece gets cited and converts.

The scored de-slop gate (mandatory before "final")

After the Optimized draft is written, score it with an LLM-as-judge against the rubric below before presenting it. Auto-revise anything failing, then re-score. Only a passing draft is offered as the Final Draft. This runs on top of the stop-slop checklist, not instead of it.

Rate 1–10 on each dimension:

Dimension Question Pass
Directness States things plainly, not throat-clearing announcements? ≥7
Rhythm Sentence length varied, not metronomic? ≥7
Specificity Concrete facts/examples, not vague declaratives? ≥7
Authenticity Reads human; first-person/experience present where apt? ≥7
Trust/E-E-A-T Named author, sourced claims, dated facts? ≥7
Density Nothing padded or cuttable? ≥7
Brand fit Matches the client's voice and banned-word list? ≥7

Gate: total ≥ 49/70 and no single dimension below 7. Below that, revise and re-score. Log the final score to the run record.

AI writing tells to eliminate (2026)

The classic tells still apply (see stop-slop), plus these current ones:

Source-tier & recency discipline (research step)

When selecting facts to cite: - Prefer high, diverse tiers — government, peer-reviewed/academic, major news, official primary sources — over aggregators and SEO blogs. AI engines weight diverse authoritative domains more heavily. - Prefer original data (studies, surveys, first-party numbers) — citation-worthy specifics beat generic claims. - Prefer recent and stamp it — include "Last updated" and dated statistics; freshness is a citation factor. - Every cited fact keeps its real source URL for the source list and the no-fabrication guardrail.

GEO structure (drafting step)

Schema (ships with every piece)

Don't just "recommend" schema — generate it. Call flood-schema-generator to produce matching JSON-LD (Article/BlogPosting, FAQPage, LocalBusiness/Service as relevant, author/Organization) and include it in the deliverable. Schema is a named technical-signal layer for AI citation.

AI-crawlability check (onboarding + when relevant)

On-page work is wasted if AI bots can't fetch the page. Verify: - robots.txt does not block GPTBot, ClaudeBot, PerplexityBot, Google-Extended, OAI-SearchBot. - Cloudflare (or other CDN) isn't blocking AI bots by default — this silently shut off AI traffic for many sites. - Pages are server-rendered/crawlable (clean semantic HTML, sitemap present). Screaming Frog can pull robots directives and rendering status. Record findings in the Competitors & Gap doc as fixes for the client's dev team.

Off-site awareness (gap analysis + delivery)

2026 citation depends heavily on what the wider web says about the brand, not just owned pages. The engine can't do PR, but it must surface the opportunity: - In Competitors & Gap Analysis, capture off-site citation gaps — where rivals get cited that the client doesn't (Reddit threads, YouTube, G2/Clutch, expert roundups, news). UGC sources like Reddit and YouTube are disproportionately cited by AI engines. - With every finished piece, include a short distribution/repurposing note: how to extend it off-site (a LinkedIn post, a genuine Reddit/community answer, a source-pitch angle for a publication). Owned content is one layer; entity authority and third-party mentions are the others.

llms.txt (optional, low-leverage)

Generating a per-client llms.txt is a cheap nice-to-have that removes a minor crawl-friction point. It is not a citation driver and Google's Search team doesn't use it — don't oversell it or spend real effort on it.