Why this matters

AI-detection signals are real, and the platforms increasingly down-weight content that reads like an LLM wrote it without an editor. We have measured this across our engagements: pages that pass our human-signal checklist outcite pages that do not by 1.4–2× across ChatGPT and Perplexity, controlling for content quality.

The LLM-detector arms race is not the only reason to do this. Readers can also tell. A page that opens with “In a world where…” loses trust in the first sentence; a page that opens with “We measured this across hundreds of pages and the math is consistent…” keeps it.

The blocklist

Forty markers we ban in generation prompts and catch in post-edit. The headline ones:

  • Nouns: tapestry, realm, landscape, journey, toolkit, testament, manifold, multifaceted, nuance, quest, roadmap, significance, whimsy
  • Verbs: delve, embark, embrace, empower, ensure, foster, harness, navigate, resonate, revolutionize, seamlessly, transcend, underscore, weave
  • Adjectives: comprehensive, crafted, crucial, curated, groundbreaking, paramount, pivotal, profound, robust, seamless, transformative, vibrant
  • Phrases: “in a world where”, “in today’s [X]”, “in conclusion”, “it’s important to note”, “paving the way”, “dive into”, “not only X but also”, “step-by-step guide”, “the rise of”, “when it comes to”
  • Metaphors: any journey/tapestry/toolkit/landscape construction

The list is not theoretical. We pulled it from analysing what major AI-detection systems flag. Every banned word was a top-10 signal in at least one detector we tested against in Q1 2026.

The required human signals

Every published page must show at least four of these:

  • One em-dash (—) used naturally (not as a hyphen substitute)
  • 1–2 contractions (it’s, can’t, we’ve, there’s, you’ll)
  • One sentence starting with But / And / So
  • One paragraph that is a single short sentence
  • One concrete number or date
  • One explicit opinion or provocative claim
  • Mixed paragraph lengths (1 line, then 4, then 2)
  • Oxford comma dropped in 1–2 places naturally

These are not stylistic preferences. They are the pattern markers human writers naturally produce and LLMs systematically do not. Hitting four out of eight tips the AI-detection scoring back into the human range.

The five-axis rubric

Every Answerly page is scored 0–5 on five axes before publish:

  • Hook — does the opening earn the next sentence?
  • Humanness — do the human signals show up?
  • Specificity — concrete numbers, named entities, real examples?
  • Voice — does this read like a particular author, not generic-house?
  • CTA — clear next step?

Publish only at ≥20/25. Anything below gets sent back for a rewrite. Two-thirds of the first drafts our writers ship score 17–19; the rewrite pass takes 30–60 minutes per page and is non-optional.

The pipeline

Every page goes through:

  1. Generation prompt with bans baked in — the writer (Claude API or human + Claude assist) generates against an explicit ban list
  2. Regex post-edit — automated check for any banned markers; flagged for human review
  3. Claude editing pass — Claude re-edits the draft to enforce required human signals (em-dash insertion, contraction conversion, paragraph-length mixing)
  4. Human editor review — Vika (or her team) scores against the rubric and either approves or sends back
  5. Schema validation — final page validates against Schema.org and Google Rich Results Test before publish

This is what we mean when we say AI-assisted content; it is not AI-generated content. The human editor pass is non-negotiable.

What you should do for your team

If you are running content production with AI assist and not enforcing this checklist — your AEO citation rate is probably 30-50% lower than it should be, and your AI-detection risk is high.

The cheapest fix: take the ban list above and the human-signal checklist and add them to the system prompt of whatever generator your team is using. Add a regex pass for the markers. That is one afternoon of work and lifts citation rate within sixty days.

If you want the full pipeline including the editor pass: that is the Growth tier at $2,400 / month. We ship one new FAQ page per month and three priority-page rewrites — every piece passes the rubric before it goes live.