CONCEPT

8-Pass AI Writing Pipeline

Properties1
tagsai-content, burstiness, perplexity, ai-detection, content-humanization, caleb-ulku

2 min read

8-Pass AI Writing Pipeline

Caleb Ulku’s multi-pass process for generating AI content that gets indexed, ranks, and passes AI-detection — used inside his Core30 AI Tool.

Definition

Instead of one prompt → one article (which “reads like AI wrote it because it did”), Caleb’s Core30 agent runs eight passes, each a separate model call:

  1. Research synthesis — compress Reddit/PAA/competitor/local research into a content brief
  2. Strategic outline — full page architecture (every H2, section purpose, flow)
  3. Section drafting — each H2 written by an independent API call, producing natural tone variation
  4. Burstiness — vary sentence/paragraph length to break AI’s uniform rhythm
  5. Perplexity injection — replace AI-tell words (robust, leverage, streamline), strip em-dashes
  6. Human bookends — rewrite the first two and last two sentences as conversational/opinionated (claimed to be weighted heaviest by Google and read first by users)
  7. Conversion — inject CTAs, phone numbers, persuasive lines (customized per content type)
  8. Final QC — re-check against brief/outline/word-count and flag leftover AI patterns

In parallel it generates FAQ, meta tags, Schema Markup (Structured Data), images, and external authority links. Quality is gauged by an AI-detection score (a demo batch averaged 39%; pieces scoring high, e.g. 65%, get another pass).

Consensus: creator-opinion. “Burstiness” and “perplexity” are real terms from AI-text detection, but this specific pipeline is Caleb’s proprietary method, and it explicitly optimizes for detection evasion — which sits in tension with his own repeated claim that “Google rewards value, not human-sounding content.”

Key Properties

  • Multi-call, section-by-section drafting is the core trick (tone variation vs. one-shot uniformity).
  • Burstiness (length variation) + perplexity injection (word-choice unpredictability) target AI-detection signals.
  • “Human bookends” — opening/closing sentences treated as highest-value.
  • AI-detection score used as a QC gate (not as a ranking signal per se).
  • Tension flag: dismisses AI detection rhetorically while engineering around it.

Examples from Sources

ExampleSource
Demo run: 29 articles, 53K words, 39% avg AI-detection score; editor re-passes a 65% pieceThis AI Tool Will Make You RANK FAST (Full Walkthrough)
Human editor mandatory to catch hallucinated numbers/regulationsTRAIN Any AI To Recommend You (This Study Proves How)

In the Sources

SourceContext
This AI Tool Will Make You RANK FAST (Full Walkthrough)Full 8-pass breakdown inside the Core30 tool.
TRAIN Any AI To Recommend You (This Study Proves How)Related multi-step Claude content process + human review.

Concepts: Programmatic SEO, 250 Authority Protocol, Schema Markup (Structured Data), Core 30 Entities: Core30 AI Tool, Claude, Caleb Ulku