Technical SEO, On-Page & Migrations

Fundamentals stable, the cost of ignoring them up; on-page is mathematical; crawlability and sequenced migrations are load-bearing.

This is the SEO Craft reference for the load-bearing technical layer: crawlability, on-page math, URL and redirect hygiene, and disciplined migrations. It synthesizes where multiple experts at the 2026 conference independently converged, then turns that into the checks and sequences you run on a live site. Read it as the "do the fundamentals correctly" companion to the on-page and anchor pages, and use the playbook section as a working checklist rather than background reading.

Across the 2026 SEO Spring Training event the technical and on-page sessions converged on a single uncomfortable idea: the fundamentals did not change, but the cost of ignoring them went up. Google "turned the knobs" (Kyle Roof, Ted Ives) rather than rewriting the rules, and a new reader (the LLM) now judges the same pages alongside the bot and the human (Simon). That makes crawlability, clean URL and redirect hygiene, structured on-page math, and disciplined migrations the load-bearing work, because mass-produced AI content fails the math and self-inflicted technical mistakes (bad migrations, crawl traps, missing contextual terms) are now the most common way sites lose traffic. This brief synthesizes the technical, on-page, and migration through-lines from Charles Taylor, Kyle Roof, Ted Ives, Brian William, Clint Butler, and Simon.

The through-line

• The fundamentals are stable; Google just "turns the knobs." Kyle Roof's repeated framing is that from his vantage point "nothing has changed" in the base of what SEOs do; Google adjusts weightings (he argues it "turned up the dials on LSI" to cheaply discount mass AI content). Ted Ives independently treats ranking as a probability range rather than a fixed number and says core mechanics (PageRank, referring domains) still govern. Charles Taylor's whole migration method assumes Google is "just a dumb algorithm, ones and zeros" you can predict and steer. The shared lesson: master the durable mechanics, do not chase every reported algorithm shift.

• On-page is mathematical, and AI content fails the math. Kyle Roof's testing (lorem ipsum pages ranking, keyword/contextual-term counts via Page Optimizer Pro and Cora) shows ranking turns on term coverage, not prose quality, and that LLMs score poorly out of the box (best model ~64 vs ~80 for a human-optimized article, contextual-term counts far below the 167 to 278 target for a 1,000-word page). His 10,000-page audit found 99.4% missing LSI. Simon arrives at the adjacent conclusion from the LLM side: AI slop earns no AI Overviews or citations because each section is chunked and judged on its own atomic facts and unique claims. Both reject "push-button" content.

• Crawlability is the precondition for everything, and crawl traps are self-inflicted. Charles Taylor's site-strategy pyramid puts Accessibility/Technical first (then Relevance, Authority, Quality) and warns that JS, parallax, and infinite scroll block Googlebot. Brian William names the specific failure mode (faceted-navigation <a> tags create an "infinite crawl" trap; fix by converting them to buttons with onclick). Ted Ives gives the diagnostics (indexed pages far exceeding real page count means Google found an infinite loop and quit; slow server response under crawl makes Google quit) and adds that Googlebot now snaps the rendered DOM about four seconds after document-ready, so content loaded slower than that is simply missed.

• Anchor text and link relevance are a precision instrument, not a volume game. Clint Butler's five-tier system treats anchor text as the relevance vote Google follows all the way down to tier 5 (his off-topic tier-5 set knocked a page out of results in seven days). Ted Ives reframes the same domain from the quantity side: PageRank is logarithmic, most small sites have a backlink quantity problem (not quality), and referring-domain count wins a keyword roughly 80% of the time when on-page is sound. Both agree links still work and both anchor the point to internal consistency. This is the same anchor discipline the Spearleaf anchor system encodes.

• Treat migrations and "mini-migrations" as engineered, sequenced events. Charles Taylor's core method is forcing Google to crawl old URLs before new ones via a phased launch, and he stretches the migration definition to cover any URL-structure change (CMS swaps, www removal, replacing thin pages with redirects). Clint Butler independently flags the same hazard from the link side: bad migrations leave old and new URLs coexisting for ~10 months and cause "calamity." The shared rule: never let Google guess which version is live.

Tactics & playbook

• Clean up the legacy site before migrating (Charles Taylor): up-to-date HTML/XML sitemaps, fix broken/3xx links and soft 404s, fix bad canonicals, remove duplicate content, fix titles/metas, enforce one H1 and correct heading hierarchy, use <p> not <span>. Whatever crawl issues exist before migration follow you to the new site.

• Find every URL, then decide Keep / Redirect / Kill (Charles Taylor): pull URLs from Search Console (16 months), Google Analytics landing pages, backlink sources, and an XML sitemap crawl, de-dupe into one master list. Keeping a URL unchanged is best (no equity loss); 301 if it must change; killing is acceptable but forfeits that page's authority.

• Run the sequenced "secret sauce" launch (Charles Taylor): dark launch the new site, point old-site canonicals at the new URLs, wait ~7-10 days, then fire single-hop 301s and the new XML sitemaps, wait again, then update internal/nav links and remove legacy sitemaps. Single-hop 301s only (no chains, no redirects to broken pages); don't forget images and PDFs.

• Keep 301 redirects effectively forever (Charles Taylor): if humans hit the old URL (his bigger worry) or Google hits it, redirect it indefinitely; keep the legacy domain in Search Console; skip the GSC Change of Address tool (no proven benefit, historically blinded you to legacy-URL data).

• Enforce URL hygiene (Charles Taylor): lowercase only (uppercase should 301 to lowercase), hyphens between words, no spaces/special characters, HTTPS everywhere, consistent trailing-slash and extension standards. Per John Mueller's rule he cites: trailing slash on the root does not matter, but on a path it does, and protocol/hostname/extension all matter.

• Use crawlable link code (Charles Taylor): <a href> not JavaScript onclick; an anchor without href is not crawled; prefer absolute over relative hrefs so Google does not wander into staging/HTTP.

• Select terms with a three-filter system (Kyle Roof): Avalanche Theory (trusted search volume), then KGR (Keyword Golden Ratio), then a keyword-difficulty cap of "medium or lower." Used together as a filter (not as standalone tactics) this surfaces terms you can win ~70-80% of with minimal SEO, lifting whole-site authority.

• Score on-page by contextual-term coverage, not word count alone (Kyle Roof): target roughly 167 to 278 important terms for a ~1,000-word page (the midpoint is the sweet spot), and never let AI strip terms out when "optimizing" (it usually improves formatting while dropping word count and terms). Cross-check against Google's NLP API.

• Become your own LLM source (Kyle Roof, Simon): weight selected terms so content satisfies query fan-out and can be cited by ChatGPT and similar engines; Roof reframes this as a higher-priced SEO package, not a new discipline. Simon's mechanism: build from atomic facts (One Subject + One Property + One Value) and earned signals so each chunk is quotable. See getting cited by AI.

• Run the LLM Multi-Gate Audit on every section (Simon): three gates (Retrieval, Extraction, Grounding/Evidence) checked with six per-section checks. Extractability check: a sentence using "it/this/we/they" with no named referent is unquotable. Substitution Test: if a competitor can swap their name into your sentence without lying, the LLM has no reason to cite you. Match format to information type (table for comparisons, numbered list for steps, bullets for parallel items, prose for relationships).

• Answer search intent in the first paragraph and structure crisis-first (Simon): for an emergency service page, lead H1 with the immediate help, then "stop the damage," then trust signals/licenses, then transparent pricing, rather than a self-centered "our history since 1983" sequence. Apply DRY (do not regurgitate a table in prose; "you don't need 4,000 words to cover a topic that needs only 800").

• Build an LLM Wiki knowledge graph (Simon): store each property as its own node (e.g. "Crown Lifespan") so every mention auto-backlinks and the graph can infer facts no one wrote, powering many articles from one fact source with zero drift.

• Fix e-commerce crawlability and naming first (Brian William): replace zero-targeting category names (bestsellers, new arrivals) and artistic product names ("Majestic Sunset Snuffler" for an orange silk scarf) with keyword-rich names; never build conjoined categories like "Socks and Shoes" (no search volume, usability failure); convert faceted-nav <a> links to onclick buttons to close the infinite-crawl trap. Merchant-center feeds, schema (including the new AI buy-form schema), and image SEO are the basics; internal linking moves e-com results faster than external links (external value takes ~3 weeks to register).

• Use crawl-rate as a diagnostic (Ted Ives, Brian William, panel): publish ~2 posts per week to keep Google returning; if indexed pages far exceed real page count suspect an infinite loop; distrust the GSC crawl report and verify with a log tool ("Glass"); splitting a heavy multi-language site into one domain per language got it fully crawled in 24-48 hours.

• Treat ranking as a probability and page speed as a filter (Ted Ives): repeat a search ~10 times in fresh sessions to see the true range and manage client expectations; page speed is a "filter, not a factor" (slow pages appear in fewer searches, especially on mobile and maps, but rankings do not move when they do appear), so report the impact in traffic, not rankings.

• Question your tool vendors on headless Chrome (Ted Ives): Google is openly flagging headless Chrome in the source code of both Search and Maps, so ask hard questions of providers rather than panicking.

Tensions & disagreements

• Word count: maximalist vs minimalist. Kyle Roof treats hitting (or exceeding) a target word count as protective (going below ~1,000 is "often very problematic"; Llama at 565 words was the worst) and pairs it with a high contextual-term target. Simon argues the opposite framing for the LLM era: "you don't need 4,000 words to cover a topic that needs only 800," and cites Google (Mueller and Sullivan) saying not to chop content into bite-sized chunks. They are not strictly contradictory (Roof targets term density, Simon targets intent efficiency) but the on-the-ground instruction ("write more" vs "write less") points in opposite directions.

• Schema as a ranking factor. Kyle Roof states flatly that schema "just for the sake of schema" was never and is still not a ranking factor (he tested it). Simon, Brian William, and Charles Taylor all treat schema as load-bearing (Simon's atomic-fact/entity structuring, Brian's "critically vital" merchant and the new AI buy-form schema, Charles's SEO-tags checklist). The reconciliation Roof himself offers (you can still rank with schema, and it can carry keywords) is implicit, but the speakers clearly weight schema differently.

• HTML meta refresh: legitimate tool vs footprint risk. Clint Butler builds his tier-3 link layer on an AWS-bucket HTML meta refresh and frames it as a legitimate redirect ("in the HTML guidelines"). Charles Taylor's migration method is built entirely on single-hop 301s and never endorses meta refresh as a redirect mechanism, and his guidance to avoid anything that confuses Google sits in tension with deliberately refreshing/redirecting for link manipulation. Different goals (white-hat migration vs off-page link engineering), but a junior reader should not conflate the two.

• How aggressive to be with crawl/link tactics generally. Clint Butler pushes back on the "white hat" framing itself (arguing guest posting and outreach are the same category as PBNs) and shares high-toxicity link tactics. Charles Taylor self-describes as a corporate white hat ("you name the sneaky tactic, I've probably tried it" notwithstanding) and explicitly will not migrate to shed toxic links. The event did not resolve the white-hat/black-hat line; it mostly de-positioned the labels.

Sources (conference sessions)

Conference session references, not pages on this site:

• Charles Taylor, Website Migrations With Zero Traffic Loss
• Marion, Austin, Roof, De-positioning, mindset, and keyword math vs unguided AI (Kyle Roof's on-page/testing block)
• Roof, Martinez, Buckner, AI-proof SEO, selling your agency, the growth flywheel (Kyle Roof's LSI/term-selection block)
• Day 2 wrap-up, link-building / e-commerce panel (Butler, Ives, William, Samuels)
• Simon, Talk Nerdy To Me (technical SEO for the LLM era)

Connect it to your system

This page sits inside SEO Craft alongside the other build-quality references. To put it to work:

• On-page execution: the term-coverage and contextual-term math here is operationalized in on-page; the schema page covers the markup the speakers split on (load-bearing for entities and e-commerce, not a standalone ranking factor per Roof).
• Anchor and link relevance: the anchor-text discipline (Clint Butler's tier system, brand-only tier-1 anchors, exact-match sensitivity) is encoded in anchors.
• AI readers: for Roof and Simon's be-your-own-source and atomic-fact tactics, see getting cited by AI, AI search visibility, and entities and schema.
• Run the migration: when you execute a URL-structure change, follow the sequenced site migration playbook (dark launch, single-hop 301s, sitemap and internal-link order).
• Where this fits overall: see the overview to start and strategy for how the technical layer feeds the wider system.

See also the off-page system this theme overlaps: DAS and cloud sites.