Spearleaf · Position Zero Playbook v10 · 2026-06-16
Start here Changelog
Reference

"Talk Nerdy To Me: Technical SEO for the LLM Era (Simon / limeygent)"

Simon's engineer-minded walkthrough of winning local-business content in both Google and LLM search by building from atomic facts, credible signals, and a repeatable per-section LLM audit.

On this page

Simon (last name not given in the deck; GitHub handle "limeygent") delivered a technical, engineer-minded walkthrough of how to make local-business content win in both Google and LLM-driven search (AI Overviews, ChatGPT, Perplexity, Grok). The through-line: AI "slop" no longer ranks because LLMs do not read pages as documents, they chunk them into sections and judge each chunk on its own. The fix is to build content from atomic facts and credible signals, structure each section to pass an LLM's retrieval and extraction gates, and run a repeatable "LLM Multi-Gate Audit" over every section. This is a deck-only session (no recorded transcript). The day of the conference is not stated in the source.

Main takeaways

  1. AI slop no longer earns AI Overviews or citations. Push-button 4,000-word AI content is cheap, fast, and client-approved, but commodity content carries no unique perspective or data, so LLMs have no reason to cite it. The slide reads "AI slop leads to no AIOs."

  2. LLMs chunk content, they do not read it, so every section must stand alone. A section is judged in isolation: it either stands up or it doesn't. Audits built for the old two readers (Google's bot and a human skimmer) miss the new reader, the LLM.

  3. Every answer must pass three gates, checked with six checks per section. The "LLM Multi-Gate Audit" runs a Retrieval gate, an Extraction gate, and a Grounding/Evidence gate (the precise label for the third gate is inferred from the "Evidence and Citations: The Grounding Layer" slide). Six per-section checks tell you which gate failed. Pull any section, run the six.

  4. Extractability requires named referents and unique claims. If a sentence uses "it," "this," "we," or "they" with no named referent, the chunk is unquotable. The Substitution Test: if a competitor can swap their name into your sentence without lying, the LLM has no reason to cite you.

  5. Match format to information type. Comparison data goes in a table, sequential steps in a numbered list, parallel items in bullets, relationships in prose. There are eight "fingerprints of AI slop" (not individually enumerated in the deck); three or more in one section flags slop.

  6. Build content from atomic facts and signals. Atomic facts (Russell and Wittgenstein: the world is made of facts, not things) are One Subject plus One Property plus One Value, verifiable and objective. Signals (Michael Spence's signaling theory, asymmetric information) are credible, hard-to-fake proofs. Facts are inputs you control; signals are scores you earn.

  7. An "LLM Wiki" knowledge graph stores each fact once and reasons across articles. Make each property its own node (for example "Crown Lifespan"), so every mention auto-backlinks and the graph can infer facts no one wrote (a nightguard preserves Crown Lifespan via bruxism). One fact graph powers many articles with zero drift.

  8. Replace vibe coding with structured engineering. Write a PRD, decompose it into discrete tasks with defined inputs and success metrics, have one LLM code and a fresh LLM audit (to beat context rot), and loop via the Ralph Wiggum pattern (Geoffrey Huntley).

  9. LLMs pick local businesses by buckets of evidence. Intent match, local grounding, proximity/logistics, E-E-A-T credibility, verifiability/consistency, actionability, freshness, behavioral evidence, and spam checks. Use weighted two-step scoring (rough candidate list, then deep dive), with weights that shift by query type (emergency versus research).

  10. Replace WordPress themes and plugins with AI-written HTML. Themes ship bloated CSS/JS you never use; AI writes precise HTML and updates it faster than page builders. Reviews, schema, and shortcodes can all be AI-coded directly.

Key points

Single speaker (Simon / limeygent). The session is deck-only, so points are anchored to slide themes, not timestamps.

Framing: AI slop and the new reader

Search intent

LLM Multi-Gate Audit (the llm-audit skill)

LLM Wiki / knowledge graph

Blog production and Image Banger

DRY content and on-page bangers

Replacing WP themes and vibe coding

Atomic facts and signals (the theory)

Evidence Cards (worked roofing example)

LLM selection and scoring of local businesses

Resources

Slides

Slides (105) Slide 1 Slide 2 Slide 3 Slide 4 Slide 5 Slide 6 Slide 7 Slide 8 Slide 9 Slide 10 Slide 11 Slide 12 Slide 13 Slide 14 Slide 15 Slide 16 Slide 17 Slide 18 Slide 19 Slide 20 Slide 21 Slide 22 Slide 23 Slide 24 Slide 25 Slide 26 Slide 27 Slide 28 Slide 29 Slide 30 Slide 31 Slide 32 Slide 33 Slide 34 Slide 35 Slide 36 Slide 37 Slide 38 Slide 39 Slide 40 Slide 41 Slide 42 Slide 43 Slide 44 Slide 45 Slide 46 Slide 47 Slide 48 Slide 49 Slide 50 Slide 51 Slide 52 Slide 53 Slide 54 Slide 55 Slide 56 Slide 57 Slide 58 Slide 59 Slide 60 Slide 61 Slide 62 Slide 63 Slide 64 Slide 65 Slide 66 Slide 67 Slide 68 Slide 69 Slide 70 Slide 71 Slide 72 Slide 73 Slide 74 Slide 75 Slide 76 Slide 77 Slide 78 Slide 79 Slide 80 Slide 81 Slide 82 Slide 83 Slide 84 Slide 85 Slide 86 Slide 87 Slide 88 Slide 89 Slide 90 Slide 91 Slide 92 Slide 93 Slide 94 Slide 95 Slide 96 Slide 97 Slide 98 Slide 99 Slide 100 Slide 101 Slide 102 Slide 103 Slide 104 Slide 105

Source

This page is synthesized from the SEOST conference deck for Simon's ("limeygent") session, "Talk Nerdy To Me" (deck file: Simon - Talk NerdyTo Me.pptx, 105 slides). It is a deck-only session with no recorded transcript. Unknowns are marked inline (speaker's last name, conference day, and the exact label of the third audit gate).