"Kyle Roof: LSI as Google's AI-Content Filter, Avalanche Theory, KGR, and Being Your Own LLM Source"

Kyle Roof argues Google turned up LSI scoring as a cheap mass-AI-content filter, then shows a three-filter term-selection method (Avalanche Theory plus KGR plus keyword difficulty) for winning terms fast and becoming your own cited source inside LLM answers.

Kyle Roof (on-page SEO, PageOptimizer Pro) opened on a recurring failure pattern: sites humming along on pure AI content that fell off a cliff. His theory is that Google does not need to detect human-vs-AI directly; it can cheaply discount pages whose contextual (LSI) term coverage is far below expectation, then discount the rest of the site. His positive method is a three-filter term-selection system that surfaces low-competition terms Google already trusts you for, lifting whole-site authority and letting the same content satisfy LLM query fan-out so you can cite yourself.

Main takeaways

1. Google likely turned up LSI scoring as a cheap mass-AI-content filter, and almost no one passes it. Rather than detecting "human vs AI," Google can discount a page that is significantly below expected on LSI, then discount the rest of the site. LLMs are weak on LSI out of the box and people prompt poorly, so low LSI coverage is an easy, inexpensive signal to discount mass content. A 10,000-page audit showed 99.4% missing LSI.

2. Stack Avalanche Theory, KGR, and a keyword-difficulty cap as a single filtering system. Used together (not as standalone tactics) this surfaces low-competition terms you can win roughly 70 to 80% of with minimal SEO, which raises whole-site authority, builds a moat, and targets terms competitors overlook ("SEO almost without SEO").

3. Become your own source for LLMs. By weighting the selected terms, the content both makes Google trust the site fast (all boats rise) and satisfies the query fan-out for AI search engines, so you can get inserted into engines like ChatGPT and cite yourself. Roof reframes this GEO work as a slightly tweaked SEO package you can charge more for, not a new discipline.

Key points

The AI-content penalty theory (LSI)

• Recurring pattern in consultations: sites that were "humming along" (around 5 million impressions) fall off a cliff; their top pages were built with plain AI and owners bragged about their "money prompt."
• Theory: Google found "a very elegant, very inexpensive way to find mass-produced AI content" by turning up the dials on LSI, because LLMs are "awful out of the box on LSI" and people are not prompting properly.
• Mechanism: if a page is significantly lower than expected on LSI, Google discounts that page, then discounts the rest of the site. "Those are the sites that got hit."
• Of twelve consultations on hit sites: ten admitted to "pure AI"; two claimed handwritten but their counts "looked exactly the same" as AI ("hand-wrote their way into content that looked like AI").

Term-selection method (three filters)

• Goal: pick terms via a numbers game, avoiding worry about intent or query fan-out.
• Start by putting a concept into any keyword/content tool and pulling roughly 800 candidate words/terms.
• Filter 1, Avalanche Theory (search volume): Over a three-month span, take low clicks/day and high clicks/day and average them to derive the search volume Google will trust you to go after; keep only terms within that volume. A QR code on Roof's slide links to a free course explaining Avalanche Theory (URL not stated in the source).
• Filter 2, KGR (Keyword Golden Ratio): Apply KGR as a second filter. Both Avalanche Theory and KGR were "really hot" then "fizzled away" because alone they are incomplete; combined as a filter they are "extremely effective."
• Filter 3, Keyword difficulty: Run the survivors through any keyword-difficulty tool and keep only those rated "medium or lower."
• Result: a "powerful set of terms" you can launch with minimal SEO and win roughly 70 to 80% of. "Win" means page one or page two within about a month with minimal SEO.
• Benefits: gets clicks/impressions, raises whole-site authority, "builds a moat," targets terms competitors overlook.

Be your own source for LLMs

• Weight the selected terms (a method Roof says he has "taught for years," that "worked for a decade," and was "even better" in the last two years).
• This satisfies two things at once: Google immediately trusts the terms (all boats rise), and the content satisfies query fan-out for AI search engines, so "you can become your own source."
• PageOptimizer Pro (POP) examples where his own pages are the cited source in ChatGPT: "eChecker for SEO" and "best LLMs for SEO."
• Reframe: GEO work is "basically just an SEO package that's slightly tweaked, but you can charge more for it." Joke: "the content blaster five thousand... now becomes the AI content blaster seven thousand."
• Productivity angle: a proprietary keyword score built from the weighting, plus a "one-click silo" tool that crunches the numbers and outputs a silo / content plan in about 5 minutes.

The 10,000-page audit

• Ran the last 10,000 unique pages through POP (about one week of data); all first runs (a baseline, the first time a small-business owner or SEO put a page in).
• 45.4% had no schema.
• 99.4% were missing LSI.
• Word count averaged about 1,500 on pages; the target was about 2,200.
• Conclusion: people over-rely on AI; AI does not give them structure or contextual terms (which opens them to a penalty) and brings "a knife to a gunfight" on word count.
• Note: these are POP-tool baselines from one week of first-runs, not independent third-party data.

Closing metaphor

• "The Chess Players" painting (late 1700s, hung in the Louvre while the artist was alive): a man plays Satan for his soul and appears in checkmate, and the Louvre renamed the placard "Checkmate." About 100 years later a grandmaster found not just an escape but a path to victory. Roof's point: with AI overwhelming you, there is an escape, and it is a path to victory.

Source

Synthesized from the conference knowledge notes for the Day 1 Part 2 session (Kyle Roof, Chris Martinez, Chase Buckner); only Roof's portion is included here. No deck file was supplied for Roof's talk (he presented on-screen slides, but no deck text was provided), so no slides are embedded. Roof's target word count (about 2,200) and audit percentages are PageOptimizer Pro baselines from about one week of first-runs (roughly 10,000 pages), not independent third-party data. The Avalanche Theory free-course URL was not stated in the source.