DEJAN AI Reverse Engineering

Overview

Use this skill for DEJAN-style AI SEO and search-system reverse engineering. It gives a compact working model of Dan Petrovic's recent ideas around grounding, selection, machine readability, entity bias, and model-aware visibility.

When To Use

• The user mentions dejan.ai, Dan Petrovic, DEJAN, SRO, AI Rank, Tree Walker, grounding snippets, or selection rate.
• You need DEJAN-style framing for Google AI Mode, Gemini grounding, AI Overviews, citation mining, or brand-association analysis.
• The task is generally about reverse engineering AI search, grounding, snippet extraction, or answer construction even if DEJAN is not mentioned by name.
• You want a local snapshot of recent DEJAN articles without re-reading the full archive.

Workflow

1. Read references/dejan-ai-notes.md first for the core models, recent studies, and default heuristics.
2. Read references/dejan-ai-archive-2025-2026.md when you need article lookup, dates, or a quick timeline.
3. Read references/dejan-models-and-systems.md when the task touches AI Mode internals, browser embeddings, classifiers, fan-out models, or DEJAN's build patterns.
4. Read references/dejan-legacy-technical-seo.md when the task needs older DEJAN technical SEO heuristics from the pre-AI era.
5. Read references/reverse-engineering-checklist.md when the task is operational and you need a step-by-step diagnostic workflow.
6. Separate sourced DEJAN claims from your own inference. Cite article title and date when leaning on a DEJAN-specific concept.
7. Treat older DEJAN material as historically useful heuristics, not automatic truths. Verify time-sensitive claims against current systems.
8. If the task is time-sensitive or you need newer material, browse dejan.ai again. This pack is a snapshot captured on 2026-03-20.

Operating Lens

• Ranking is necessary but not sufficient. Ask what sentence or claim survives into grounding.
• Track selection rate, not just clicks or rankings.
• Assume grounding budgets are limited and transient. Prioritize clear, self-contained statements near the top of the page.
• Treat model output as a mix of latent bias and retrieved evidence, not retrieval alone.
• Optimize for machine readability: distilled main content, clean blocks, explicit claims, and tool-friendly structure.
• Expect hidden fan-out or implicit sub-queries behind a visible prompt.
• For brand work, inspect associations and uncertainty, not just exact-match mentions.
• For reverse engineering, prefer artifacts over theory: snippets, chunks, citation positions, prompt variants, and source-page diffs.

Core DEJAN Concepts

• SRO = Selection Rate Optimization, DEJAN's AI-native successor to CTR.
• Selection Rate = frequency with which a grounded candidate is chosen into the answer.
• Primary bias = the model's latent relevance perception before grounding effects are applied.
• Grounding snippets/chunks = the small text units actually supplied to the model.
• Transient grounding = a grounded passage may shape the answer without becoming durable model memory.
• LLM as presentation layer = retrieval and ranking still sit underneath the model layer.
• Association networks and Tree Walker = probes for brand adjacency, token confidence, and model worldview.

References

• references/dejan-ai-notes.md
• references/dejan-ai-archive-2025-2026.md
• references/dejan-models-and-systems.md
• references/dejan-legacy-technical-seo.md
• references/reverse-engineering-checklist.md