Spokes.wiki Search About

Synthesis — Search, SEO & Performance Marketing

The evolving thesis. Spun out 2026-06-01 as "AI Search" (the AI transformation of discovery); broadened 2026-06-03 (human directive) to own the whole search/SEO/PPC field — the traditional discipline and its AI disruption.

Scope: one continuum

search-marketing (organic SEO + paid PPC + affiliate — ops, strategy, measurement) is the established discipline; the ai-search-shift is its disruption; generative-engine-optimization is the successor sub-practice. The traditional concerns don't disappear — they mutate: structured-data "eligibility signals" (seo-commissioning-workflow) become GEO's machine-readable substrate; measurement stays the cross-cutting hard part (PPC incrementality b2b-ppc-metrics ↔ GEO opacity google-io-business-visibility); affiliate reputational signals (seo-affiliate-alignment) feed AI recommendations. So the wiki reads as traditional search marketing → AI transformation → GEO, one story at different points in time. The software layer beneath the practice now has a node too (2026-07-10): all-in-one suites (Semrush/Ahrefs) vs budget/niche tools. Two caveats travel with it — most tool comparisons are vendor listicles with a conflict of interest (semrush-alternatives-answersocrates ranks its own tool #1, echoing Google's "we don't endorse third-party tools"), and question/intent research is the shared input to classic keyword SEO and GEO — so the tooling sits on the same traditional→AI-era seam as the strategy.

Current thesis (the AI-transformation half)

The catalyst is Google's May 2026 search overhaul (AI overviews over links; agentic-commerce previews at I/O) — the ai-search-shift. It splits into three reactions — two demand-side (below) and a third platform-side (the platform asserts authority — see "How the thesis filled in"):

  1. Consumers contest it. duckduckgo-no-ai-search documents a real, measurable flight to no-AI search (DuckDuckGo, Kagi; ~84% above baseline) — the AI-default results page is not settled, and the market is fragmenting on a "do I want AI in my results?" axis.
  2. Brands adapt to it. With clicks-to-your-site no longer guaranteed (google-io-business-visibility, agentic-commerce), the new game is being recommended by AI — generative-engine-optimization. The mechanism (per brand-depth-ai-recommendations) is brand depth: entity salience + coherence + relationship density win you both parametric weight (be known inside the model) and retrieval survival (pass retrieval-augmented-generation filters). And per ai-content-seo-visibility, AI content volume alone fails — you need good first-party inputs, operationalized.

The unifying move: discovery's contested ground shifts from ranking on a page to being present in a model's answer. Two consequences thread all four sources — opacity (you can't see why an agent didn't consider you) and consolidation of the funnel (the agent/overview owns the journey end-to-end).

How the thesis filled in

Threads added by later sources, folded into the core thesis above rather than left as a running log.

The operating-model consequence

The visibility debate has a labor/org corollary: seo-no-longer-drives-growth argues the traditional SEO deliverables (packaged keyword research, high-volume content, standalone on-page) no longer drive growth because they've commoditized, and teams must reallocate to five non-commoditizable capabilities — entity/brand building, original research, distribution & PR, AI-search visibility, analytical depth — captured as the seo-operating-model-shift. This is the what-work-by-whom layer beneath the what-to-optimize-for visibility pages (in-house: senior strategists over production headcount; agencies: capability offerings over retainer deliverables). A task-level instance (2026-07-08): ai-content-gap-analysis-workflow runs a classic content-gap-analysis by handing the commoditized part — collating Semrush/GSC/GA4 reports and sorting thousands of gap keywords — to an LLM (Claude, via CSV or MCP), while the analyst keeps the strategic call of which opening serves the business. It's the operating-model shift seen from the bottom up: not "fire the SEO," but move the row-scrolling to the model and validate its output against first-party GSC data. It also lands on the clarity-over-volume side — preferring a qualified-visitor topic over a high-volume-but-hard keyword.

The operating-model shift has so far been an organic-SEO story (commoditized content production moving off human desks). groas carries it to the paid leg: an autonomous AI product that runs a whole Google Ads account — bids, keywords, budget, ad copy, intent-matched landing pages — and is sold not as assistance but as a full replacement for the PPC manager or agency team ("the difference isn't skill, it's math"). Two things fold in. First, it's the aggressive pole of the automation axis: where ai-content-gap-analysis-workflow keeps the human's strategic call and hands the LLM the grunt work, groas claims the whole loop — the same axis, pushed to substitution. Second, it lands the wiki's measurement caveat on a live product. groas optimizes toward ROAS and conversions — the exact metrics b2b-ppc-metrics warns "may be lying to you" (quadruple-counted conversions, false ROAS, average-vs-marginal CPA). A model trained to maximize reported ROAS inherits whatever that metric mis-measures and scales it faster than a human would; "trained on $500B of profitable spend" only relocates the question to whose profit, measured how. So autonomous PPC doesn't dissolve the incrementality problem — it raises the stakes on getting the objective right before handing over the bidding. (T4 throughout — the $500B/$100m+/500-client figures are unaudited vendor copy.)

The platform asserts authority — the third reaction

Above the consumer/brand split: the platform owner moves to control the discipline itself. google-guidance-seo-authority documents new Google Search Central guidance that (a) makes Google's own docs the benchmark all SEO advice is measured against, (b) claims AEO/GEO (answer-engine-optimization, generative-engine-optimization) fall under Google's purview rather than being vendor-defined, and (c) tells practitioners third-party tools lack Google's internal ranking data — use Google Search Console (the "first-party tool") instead. Two consequences fold into the existing thesis:

Traditional technical-SEO ops — first-party rigor (the underrepresented half)

The spoke's traditional half is mostly operational rigor, and domain-migration (google-domain-migration-requirements) is a clean instance: Google tightened site-move guidance to require a Change of Address for every verified domain variant (each subdomain, www/non-www, even unused ones), because stray linking makes Google crawl forgotten permutations. It folds into the platform-authority thread from the technical side — the authoritative procedure lives in GSC + Google's docs, not third-party folklore — and a botched migration surfaces first as search-indexation loss across variants. A reminder that beneath the AI-visibility debate the field still runs on redirects, index health, and variant hygiene. A second, smaller instance: x-frame-options-seo — Mueller's view that, of the HTTP security headers, only the iframe-blocking one (X-Frame-Options / CSP frame-ancestors) plausibly touches SEO, and defensively (stop a third party from ranking your content inside a frame on their domain). Both pieces are the same genre as hyphenated-domains-seo — a Googler issuing the canonical answer to a narrow technical-SEO question on social/Reddit, which the trade press relays — reinforcing the "Google as SEO authority" thread (google-guidance-seo-authority) from the small-tactical-fact end rather than the formal-doc end.

Foundations & mechanisms — Search Essentials + E-E-A-T + llms.txt

First-party/standard pieces ground the trade-press thesis. google-search-essentials is the canonical ruleset the platform-authority thesis points at: Google's own three-pillar document (technical requirements + 16 named spam policies + best practices). Two threads land directly on it. Opacity from the source: Google states outright that meeting every requirement "doesn't mean Google will crawl, index, or serve" a page — the rules are necessary, never disclosed-sufficient. The AI-content stance, primary-sourced: the scaled content abuse policy prohibits "using generative AI tools … to generate many pages without adding value," i.e. Google penalizes mass-produced rank-gaming, not AI content per se — the first-party grounding for the "AI volume alone fails" thread. e-e-a-t (Google's Experience/Expertise/Authoritativeness/Trust framework) is the older, Google-authored root of the "be a credible entity" mechanism — and a clean opacity example (raters assess it; "not a direct ranking factor"). llms-txt is a concrete GEO mechanism: a curated /llms.txt markdown map for AI at inference time — the GEO counterpart to robots.txt/sitemaps, though (like all GEO) its payoff is unmeasured. Together they sharpen "is GEO just SEO rebranded?": GEO is partly old credibility signals (E-E-A-T) + Google's standing anti-mass-production rule re-aimed at models, partly genuinely new plumbing (llms.txt).

The product and the meter — AI Overviews + Search Console

Two pieces ground the thesis's two most abstract points in concrete, dated objects. ai-overviews is the ai-search-shift as a product: Gemini-powered summaries atop results, on >48% of searches by March 2026 (from ~6.5% a year earlier), occupying ~67.1% of the desktop screen when paired with a featured snippet. That last figure is the hard supply-side data under the click-collapse thread and the reason GEO exists — both poles of the optimize-vs-abandon fork rest on the same fact. Publisher lawsuits (Chegg, Penske) and the Jan-2026 health-summary restriction make the opacity + platform-power theme adversarial. google-search-console is the meter behind the recurring "the only authoritative signal is the one Google exposes" observation: it reports real Search queries/impressions/clicks/position (first-party, not modeled) — but does not expose whether an AI Overview or GEO answer considered you. So even the authoritative tool leaves AI-recommendation unmeasurable, which is precisely why "optimize for AI recommendation" risks being unfalsifiable (the measurement-gap open question, now pinned to a specific tool). Together: Google owns the answer (ai-overviews) and the meter (google-search-console) — the platform-authority thesis (google-guidance-seo-authority) in two artifacts. The gap has a measurable floor, though: search-indexation — whether a URL is in the index at all — is checkable (GSC's index-coverage report, or third-party index checkers per google-index-checker-use-cases). The split is clean: the traditional technical layer (indexed? ranking?) is verifiable; the AI-recommendation layer (considered?) is the part that vanishes from measurement. The meter's reach is also widening (2026-07): gsc-platform-properties adds a GSC property type for social/video accounts (Instagram, TikTok, X, YouTube), so Google now reports how your off-site posts perform on Search & Discover — content it doesn't even host. It deepens "the platform owns the meter" (now beyond your own domain) and nods to discovery fragmenting across channels, while staying inside the same bound: it meters Search/Discover clicks, not AI-Overview/GEO consideration.

The apparatus behind "Google as SEO authority"

The platform-authority thread had a gap: we kept citing "Google says X" without a node for the thing that says it. google-search-central fills it — Google's Search Relations program (ex-"Webmasters"), the institutional spine that issues the guidance across four surfaces at once: the formal docs (google-search-essentials et al.), the YouTube channel, the search-off-the-record podcast, and SEO Office Hours. It also names the humans — john-mueller and martin-splitt — who were already the wiki's most-quoted voices (hyphenated-domains-seo, x-frame-options-seo) but had no page. The provenance payoff is a tier ladder on the same first-party authority: a formal doc is T1; the same team saying the same kind of thing on a podcast or Office Hours, relayed by SEJ/SEL, lands T2–T4 — first-party in substance, informal in form. That ladder explains the wiki's source-base shape better than "trade press vs primary" did.

A live instance arrived with the apparatus: gsc-validate-fix-guidance (Mueller/Splitt on a search-off-the-record episode) explains Validate Fix — and the explanation is quietly deflating. The button most operators read as "verify my fix" does neither verifying nor fixing: it samples URLs and accelerates a recrawl, nothing more, and most flagged issues self-resolve as Google recrawls anyway (Splitt: the index-coverage report surfaces patterns, it isn't a to-do list to zero out). This sharpens two existing threads. On the meter (google-search-console): even GSC's most prominent "act on this" affordance is, by Google's own account, usually unnecessary — the tool observes more than it lets you do. On crawl economics (crawl-budget): the only lever the button pulls is recrawl timing, the exact resource Google rations — so search-indexation recovery is gated by re-crawl, not by a click. The operator's side of the crawl-budget story, from the meter's own UI.

Citations are outcomes, not drivers (the recurring reframe)

"Citations are outcomes, not drivers" (brand-depth-ai-recommendations): models mention brands far more than they cite them (only 6–27% overlap). GEO is therefore about structural presence in the model, not chasing links — a genuinely new optimization target vs. classic SEO.

Volume inverts — density beats coverage (the supply-side mechanism)

publishing-volume-hurts-seo supplies the supply-side mirror of brand depth: where brand-depth-ai-recommendations says coherence wins AI recommendations, this says publishing volume now actively destroys it. The old "5,000 pages beat 50" coverage logic inverts once systems retrieve chunks and weight embeddings — near-duplicate pages cause semantic dilution and internal vector competition ("competing for embeddings, not just rankings"), so the lever becomes authority-density ("clarity, not volume"), raised by consolidation + structural clarity for extractability. Two existing threads gain a mechanism: it generalizes classic keyword-cannibalization (keyword overlap → semantic/vector overlap — a clean traditional→AI-era bridge), and it upgrades the "AI volume alone fails" thread from doesn't help to measurably hurts. On the central fork it lands firmly optimize-the-channel (consolidate and win retrieval), opposite the abandon-the-channel pole. Same caveat as the whole density family: the harm and the remedy are conceptual models, unmeasured — no data that consolidation lifts AI consideration.

The supply-side half of the inversion — crawl economics (2026-07-14). scaled-ai-content-crawl-economics adds the mechanism the volume-inversion thread was missing. Where publishing-volume-hurts-seo explains harm on the retrieval/demand side (thin pages lose in embedding space), this explains it on the supply side: Google "does not have infinite computing power," so it spends crawl budget as an investment gated by inventory-vs-utility, demand, and authority — and simply won't crawl, index, or retain a flood of thin pages. The concrete, near-falsifiable detail is a lifecycle: burst-crawl → freshness boost → decay → de-indexation for pages not re-crawled within ~75–140 days, with low-value clusters getting reduced crawl frequency so the decay compounds. Three consequences fold in. (1) It gives search-indexation a downward dynamic — the "measurable floor" recedes; thin pages get indexed in a burst then dropped. (2) It reframes authority-density as what buys continued crawling, not only what wins embeddings — the same "concentrate value" instruction at the resource layer. (3) It's the resource-economics reading of scaled content abuse: Google penalizes mass AI production because it's a bad crawl investment, not because it's AI. This nudges the density family off pure conceptual reasoning toward documented crawl mechanics — though still T3 (an SEJ VIP-contributor analysis of Google's crawl-budget docs; the specific day-window is the author's number, not a cited Google figure), so it narrows the "unmeasured" caveat without erasing it.

The mechanism under the paid half, and the arc it repeats

The paid corner ran for two months on what an advertiser measures — the CPC in b2b-ppc-metrics, the return in groas — with no page on what sets those numbers. search-ad-auction now holds it, on gsp-auction-paper for the theory and google-ad-rank for what the platform says today.

The history is this spoke's own thesis, told twenty years earlier in a different half of the domain. Overture's 1997 generalized first-price auction had no stable resting point: any bid could be undercut by a cent, so the return went to whoever re-bid fastest, advertisers bought robots, and one automated bidder against slow humans could hold the engine's revenue at 2.02 cents per click with true values of $10. Google's 2002 generalized second-price design — pay the next bid down plus a cent — removed that incentive by construction. Mechanism rewards gaming → industry invests in gaming → platform changes the mechanism. The organic version is publishing-volume-hurts-seo and scaled-ai-content-crawl-economics; the generative version is c-seo-bench finding the advice ineffective. The auction is the case where the platform's counter-move is documented in an economics journal.

Two things worth holding against the rest of the wiki. First, the most repeated claim about this mechanism is wrong: GSP is not the Vickrey auction Google's early marketing invoked, and truth-telling is not an equilibrium of it, so "bid what a click is worth" is folk advice rather than a result. Second, the transparency ran backwards. In 2002 quality score was the estimated click-through rate and rank was bid × quality; today google-ad-rank names six inputs, publishes no weights, and quality has become a gate — an Ad Rank threshold can keep an ad out of the auction at any bid. The spoke's standing complaint about google-guidance-seo-authority — the party being optimized against also writes the rules — has a twenty-year precedent in the paid half.

And one asymmetry that reframes the GEO argument: the ad auction has a peer-reviewed analysis from outside the platform, which is exactly what generative-engine-optimization lacks. That is a difference in the evidence available, not only in the maturity of the practice.

The outbound half

The wiki's scope has always run traditional → AI-era, and every page on both ends assumed the buyer moves first: you rank, you get indexed, you get cited, they come. awesome-ai-lead-generation is the first source where the seller moves first. It catalogs 24 tools that watch public conversation for a buying signal, resolve the poster to an email address, and contact them — lead-generation, with cold-email-deliverability as the gate on the contact step.

Why it belongs, on the reading that admits it. Someone posting "what tool does X?" in a subreddit has run a query nothing indexes, and the demand behind it is the same demand search-marketing competes for. The ai-search-shift says clicks are draining off the results page; outbound is one account of where the sellers went when they did. Monitoring replaces ranking as the way to reach the same buyer at the same moment. The shared input is concrete rather than metaphorical: question-and-intent research already feeds classic keyword SEO and generative-engine-optimization (seo-tools) and content-gap-analysis mines it to decide what to publish — social listening mines it to decide whom to message.

Why it strains, stated plainly. The header says search, SEO and performance marketing. Outbound is performance marketing by budget and by team, and it is not search by any reading. The router parked this source as out of scope on 2026-08-05 and the curator directed it here the same day; the decision is recorded rather than reasoned away, and one T3 catalog does not establish a corner. Either a second outbound source arrives and this becomes a real sub-domain, or this stays a single page cluster and the header is the thing that was right.

The transferable pattern, offered as a hypothesis. cold-email-deliverability and google-search-essentials describe the same structure: a gatekeeper with its own economics, rules against scaled generic output, enforcement by scoring the sender rather than judging the item. On the search side the outcome is documented — publishing-volume-hurts-seo and scaled-ai-content-crawl-economics show cheap generation meeting that gate and density beating coverage. If the mailbox is the same kind of gate, warming more inboxes is the outbound equivalent of publishing more pages, and it ends the same way. Nothing here sources the email half of that, and no source in the corpus documents mailbox-provider policy. It goes in the open questions, not the thesis.

Demand discovery as a method

Almost everything in this spoke is about competing for demand that already exists — ranking, being cited, being recommended. emerging-category-search-signals works one step earlier: how to tell that a category is forming, before the tools can price it. Paged as category-emergence.

The useful inversion is that the early signals are shape, not size. Volume arrives last and is visible to everyone; what fires first is the authentic phrasing being absent from keyword tools, formal vocabulary (standards, regulations, job titles) moving before plain language, several labels competing at similar volume, keyword difficulty sitting below demand, and a results page where two-person consultancies outrank the Big Four. The discipline that makes it a method rather than trend-chasing is three rules against self-deception: fixed keyword cohorts, a twelve-month window, and commercial-intent terms before you believe the demand has budget.

It lands on two of this spoke's standing threads.

It gives seo-operating-model-shift a concrete non-commoditizable task. That argument says to move teams off commoditized production; noticing a market before the instruments can measure it is a specific example rather than a category.

And it sharpens the measurement gap rather than closing it. Everything here is counted in searches and rankings. The spoke's #1 edge is a consideration metric for GEO — share of AI recommendation, not clicks — and keyword difficulty is silent on that. If the ai-search-shift thesis is right, this is a sharper instrument pointed at a shrinking surface. The article does not say so; the corpus should.

The evidential limit, which the source shares with the rest of the field: every case is told forward from a category that emerged. Nobody reports the clusters that showed all five signals and went nowhere, so there is no false-positive rate and the method stays a practice rather than a finding.

Cross-spoke. The primary case is AI governance, a spoke in this hub — so those figures ("AI governance framework" 40 → 3,600 in twelve months, ISO 42001 610 → 3,600) are demand-side evidence that the field ../ai-governance-wiki documents is in early-adopter growth. The method is also an empirical instrument for the diffusion-of-innovation theory ../research-wiki holds (Rogers, Moore, the hype cycle, Bass): a way to watch an adoption curve while it happens, where that spoke has the models of it.

Open questions

Who is allowed to resell a SERP? serper sells Google result pages by the thousand and says nothing about the legal basis; Google's terms have long discouraged automated querying. The corpus holds no enforcement action, no licensing arrangement and no ruling either way — only a live price list. — needs: a court decision, a terms-of-service enforcement case, or a scraping vendor's own account of its legal position. (Added 2026-09-10.)

Growth edges

Ranked; each names the kind of source that would close it (see ../QUALITY.md → Growth edges).

  1. A consideration metric for GEO — REOPENED 2026-08-08, hours after being closed. geo-paper (Aggarwal et al., KDD 2024, T1) was ingested as closing this edge: it defines visibility over the generated answer (position-adjusted word count), ships GEO-bench, and reports up to 40% gains. Then c-seo-bench (Puerto et al., NeurIPS 2025, T1) was read properly and says the metric is the problem — word count "does not measure the LLM preference, contrary to the citation ranking that we use." Across six domains and multiple competing actors the GEO interventions are "largely ineffective" and often harmful (the Statistics method lowers ranking in 19 of 24 settings), and what dominates is plain retrieval ranking. So the corpus acquired a metric and then learned it does not measure the thing. The edge stands. — needs: a measurement of citation or recommendation share on a live platform. Note what the closure attempt cost: it was closed on a proxy nobody had checked, which is precisely the failure this edge was written to prevent.
  2. Mailbox-provider policy, first-party. Whether the scaled-content lesson transfers to cold email is sourced on the search side and unsourced on the email side. — needs: a first-party mailbox-provider policy document (T1/T3 — a primary vendor fact), not trade-press interpretation of one.
  3. Conversion data for intent-based prospecting. The claim is awesome-ai-lead-generation's editorial line and carries no conversion figures, no cost comparison and no case study. — needs: any measured outbound comparison; absent one, the outbound thread stays an assertion.
  4. Is the no-AI backlash durable? — half-closed 2026-08-14. no-ai-search-share-trend (StatCounter via Backlinko, T2) supplies the multi-year global/US series the edge asked for: DuckDuckGo's share rose through 2022, fell in 2023–2024 below its 2021 level, rose again in 2025, dipped slightly in 2026. Answer is neither durable nor a spike — a series that has already reversed once, so a single post-overhaul week can't settle the question either way. — needs now: the same series read again a year past the May 2026 overhaul, to see whether this leg holds or fades like the last one.
  5. A false-positive rate for category-emergence. (new 2026-08-08) Every case in emerging-category-search-signals is told forward from a category that emerged, so the five signals have no measured precision. — needs: any account of clusters that showed the signals and did not become categories — a failed-bet retrospective, or a cohort study over a fixed keyword set.

Coverage edges (added 2026-08-08, at the curator's request for a wider backlog). These widen what the spoke covers instead of answering an open question above; one ordinary solid source closes any.

  1. How the price is set. CLOSED 2026-08-12 with both halves the edge asked for: gsp-auction-paper (Edelman, Ostrovsky & Schwarz, AER 97(1), T1, read in full) and google-ad-rank (Google Ads Help, T3, first-party), synthesized in search-ad-auction. Successor, and it is the familiar one: nobody outside the platform has measured the auction as operated. The theory is twenty years old and the current mechanism publishes six named factors with no weights. — needs: an independent estimate of how much of a CPC difference Quality Score explains, or where the Ad Rank thresholds sit. Half closed 2026-08-16 (research pass), from the auctioneer's side rather than the advertiser's. quality-score-auction-estimation (Kim & Pal, Management Science 71(12) 2025, T1) recovers advertiser values nonparametrically from bids in a quality-weighted GSP — Yahoo! Webscope data, 77.9M observations cut to 1.61M, five product categories, Jan–Apr 2008 — and then simulates the platform changing its own rule. Score squashing raises auctioneer revenue 8.5%–24.3% at the cost of advertiser profit and consumer welfare, with the revenue-maximizing exponent differing by market; tuning the precision of quality scores adds a further 0.4%–3.0%. So the quality weight is measured as a revenue instrument, by outsiders, with magnitudes attached. What is still open is exactly the advertiser's question: how much of a CPC gap a Quality Score explains, on a live platform, today. The industry's own figures for that (a QS-1 keyword costing ~400% more than QS-5) trace to vendor blog posts with no published method, and this paper does not touch them. — needs now: an advertiser-side estimate from account data with a stated method, or a platform-era replication of Kim & Pal on Google rather than Yahoo! 2008. Half closed again 2026-10-03 (research pass), from the advertiser's side this time. return-on-quality-improvements-sem (Abou Nabout & Skiera, J. Interactive Marketing 26(3) 2012, T1, read at abstract level, closed access) tests the practitioner belief that a better score means a cheaper click. Of all quality improvements in its data, 5% raised the price per click, 53% raised total costs and 37% lowered profit, while 100% raised clicks. The reason is the mechanism: a higher weighted bid lowers the price only if rank stays put, and a better rank is priced by a stronger competitor. So the spoke's unsourced claim is now contradicted in direction by a peer-reviewed source, before anyone has sourced its size. Residue (6b): the magnitude is still open. How much of a CPC gap a Quality Score explains, with a stated method, on today's Google. The sample behind the 2012 paper is in the full text and is not held here.
  2. Measurement after the cookie. The spoke names measurement constantly and holds no method: GA4, consent mode, and incrementality testing are how any of these claims would be checked. — needs: a published incrementality study or the analytics platform's methodology.
  3. Links — HALF CLOSED 2026-09-01 (quality cycle). Google's own guidance is in: google-link-spam-policy (T1, first-party), which defines link spam by purpose rather than mechanism, names the schemes, and permits paid links outright provided they carry rel="sponsored" or rel="nofollow". The rule is a labelling rule. — needs, unchanged: one study of link-market effects — pricing, prevalence, or what discounting actually does to a site. The policy says what is forbidden and nothing about what happens, and Google cannot be the source for that half.
  4. Search that is not Google. YouTube, Amazon, the app stores and TikTok are where large slices of commercial search happen, and ai-search-shift argues about attention without them. — needs: platform documentation or a measured share study.
  5. What a small publisher actually earns from AI licensing (new 2026-08-17). The money layer arrived with four T1 sources and not one figure for an independent site. Every number the corpus holds comes from a vendor selling the mechanism (pay-per-crawl's floor price, cloudflare-agentic-internet-report's deal count) or from a publisher large enough to negotiate (wikimedia-enterprise's $8.3M across 13 customers). The advice in ai-content-monetization is therefore marked as inference. — needs: any independent site's reported pay-per-crawl or licensing revenue over a stated period, or an rsl-collective royalty distribution — the small-publisher twin of the measurement gap in edge 1. Unmoved 2026-08-29: gemini-notebook-expert-intelligence adds a fifth billing event and still no figure — Google's post states no publisher terms beyond the ordinary book sale. Five mechanisms are now documented and the earnings question has never once been answered.
  6. What a per-copy AI licence pays the publisher (new 2026-08-29). per-copy-licensing is described entirely from the reader's side. Whether an AI-enabled title earns its publisher anything different from an ordinary sale is unstated by the only source. — needs: a publisher's or author's own account of the Expert Intelligence terms, an industry-press report on the deals behind the six launch publishers, or a comparable arrangement from another store.

The quality weight is a revenue dial, and someone finally turned it (2026-08-16)

search-ad-auction has held two halves of the mechanism since 2026-08-12: the theory (gsp-auction-paper, 2007) and the platform's own description of the inputs (google-ad-rank, six named factors, no weights). quality-score-auction-estimation adds the part neither could supply — what happens when the weighting itself is changed, estimated from outside the platform.

The finding that reframes the paid corner: the quality score is not only a relevance filter that happens to affect price. It is a dial the auctioneer can set for revenue. Raising quality to a power θ < 1 — "squashing" it toward irrelevance — raises the platform's revenue by 8.5%–24.3% while advertiser profit and consumer welfare fall, and the best θ is different in each product category. Yahoo!'s own chief economist described the purpose as handicapping: an advertiser with a very high click probability is "very hard to beat, so it's not a really competitive auction."

A second dial follows from the same logic and is stranger: the platform can tune how accurate its quality scores are. More precise, user-attribute-based scores raise click rates but cut competition between advertisers, so maximum accuracy is not revenue-maximizing — optimally degrading precision is worth 0.4%–3.0%.

What this does to the spoke's standing complaint. This corner has repeatedly recorded that the paid half runs on practitioner sources and platform documentation. It now has an independent, T1 estimate of the mechanism's central parameter. The complaint does not go away, it relocates: the paper measures the auctioneer's trade-off on Yahoo! 2008 data, and the number every practitioner quotes — a low Quality Score costing several times a high one — remains sourced to vendor blog posts with no method. The corpus can now say what the quality weight does for the platform and still cannot say what it does to an advertiser's bill.

The paid half had one pole, and now it has two (2026-08-17)

groas arrived in July as the aggressive end of the automation axis — an agent sold as a replacement for the PPC manager — and stood unopposed for five weeks. google-ads-ai-agent-roadmap is the other end, from the same trade and with the same commercial exposure: the goal isn't autonomous marketing. Agents do the repetitive, data-heavy work (auditing, monitoring, trend analysis, surfacing opportunities); strategy and judgment stay human; trying to automate all of Google Ads is named as the biggest mistake its author sees. Neither side has a number, so the fork is recorded below rather than resolved.

What the disagreement is actually about is narrower than it sounds, and paid-search-automation is the page that narrows it. Bid (Smart Bidding), query (broad match) and placement (Performance Max) were handed to Google's models over the preceding decade, each step reported at a coarser grain than the last. So an advertiser's agent — whether it assists or replaces — supervises an already-automated account. It does not execute a campaign. Read that way, google-ads-ai-agent-roadmap's step 3, joining ad performance against stock, margin, pricing and CRM data, is an attempt to rebuild the visibility Performance Max removed using data the platform does not hold. Automation now runs in two directions at once in the paid corner: the platform absorbs execution, the advertiser builds instrumentation back on top.

And the delegation arc rhymes with the auction's. search-ad-auction records quality score in 2002 as a published formula, bid × estimated CTR; google-ad-rank today names six inputs, publishes no weights, and makes quality a gate that can exclude an ad at any bid. Legibility fell over the same twenty years that execution moved to the platform. The complaint this spoke files against google-guidance-seo-authority — the party being optimized against writes the rules and meters the result — is the paid half's oldest pattern, not something AI search introduced.

The organizational half is new here. agentic-adoption-ladder holds the four-rung sequence (context → off-the-shelf → custom → people) and its load-bearing claim, that AI does not compensate for bad process but runs it faster. That is the sequencing layer under seo-operating-model-shift's reallocation argument and ai-content-gap-analysis-workflow's task-level split: what a team must own before either produces anything. The assertion that most teams have not exhausted rung 2 — a general assistant plus MCP connectors — is easy to survey and nobody has.

A container note, recorded on search-engine-land: the publication is owned by Semrush, an SEO/PPC tooling vendor this spoke's pages cite as a data source, with its promotions running inline in the articles. That changes no page's tier — which still turns on what the individual piece rests on — but the vendor-incentive caveat this spoke applies to bylines now applies to the masthead too.

The spoke acquires a money layer (2026-08-17, curator-directed research pass)

For a year this spoke has documented the click collapse from the marketer's side and stopped at the edge of the obvious next question: if the visit is gone, what gets billed instead? Four T1 sources answer it, and they open a cluster the wiki did not previously cover — visibility was the whole subject, and payment is now beside it. ai-content-monetization is the node.

The event being sold moves. Per visit (ads, affiliate, leads) is the incumbent model, and the one pew-ai-overviews-clicks and great-content-no-longer-works have been recording the death of. Three successors bill different events: per fetch (pay-per-crawl — HTTP 402, $0.001 floor, per-domain), per use (pay-per-inference and pay-per-training terms in rsl-standard; Cloudflare moving from pay-per-crawl toward pay-per-use; x402 and the Monetization Gateway as the rails), and per access (serviced feeds — wikimedia-enterprise).

The supply-side numbers now exist, and they are the counterpart to Pew's demand-side ones. cloudflare-agentic-internet-report: AI training crawling is 52% of all crawler requests as of June 2026, up from 22% in spring 2025; mixed-use bots are over 36%; non-human traffic passed half of all internet traffic; human traffic to the most-crawled categories fell up to 40% in a year. crawl-to-refer-ratio prices the exchange: Anthropic ≈50,000:1, OpenAI ≈887:1, Perplexity ≈118:1 (Aug 2025, whole network), tightening to 2,500:1 / 152:1 / 32.7:1 for news publishers. The ordering is stable across every cut even if the magnitudes swing — search crawlers return traffic, training crawlers do not.

Scarcity is a default setting, not a market. The mechanism that makes any toll collectable is Cloudflare blocking crawlers by default, extended on 2026-09-15 to mixed-use bots on ad-supported pages. Notice what that leaves alone: search-only crawlers such as Google's pass through, so the arrangement preserves the traffic-sending crawl and prices the rest. One vendor's default setting is doing the work of the entire licensing economy — and the vendor publishes the measurements that justify it. The vendor-incentive caveat this spoke applies to SEO trade press applies here at a larger scale.

Leverage decides who collects, and it is concentrated. Three counts say the same thing from different directions: the total licensing market is ~50 bilateral deals since 2023; wikimedia-enterprise — the one audited case of a wiki earning from AI companies — has 13 paying customers and cleared $8.3M (+148%) with a cumulative profit of $646K after four years; and the rsl-collective exists at all because a single site cannot negotiate. The crawl toll's floor price pays $1 per thousand fetches, which rewards whoever is already crawled hardest. Every mechanism here was built by and for large publishers.

What transfers to a small site, marked as inference. Wikimedia's transferable result is the product shape rather than the price: what a machine buyer pays for is a clean structured corpus with a service guarantee, not a website. Declaring RSL terms costs nothing and creates a claim; llms-txt and a feed keep the content usable; pay-per-crawl is defensible as cost control against the re-fetch waste (over half of AI crawler traffic re-fetches unchanged pages) rather than as revenue. None of that is measured. No source in this corpus reports what an independent publisher earns from any AI-licensing mechanism, which is now growth edge 10.

A fifth billing event, and the first that bills the reader (2026-08-29)

The four events above all bill the AI company. Google's Expert Intelligence bills nobody new: a Play Books purchase authorizes Gemini Notebook to ground answers in that book, and a collaborator on a shared notebook must buy their own copy or lose access to it. 100,000+ books at launch from six publishers, including Penguin Random House, Macmillan and O'Reilly Media. per-copy-licensing is the node.

Read against the leverage finding directly above, it is the first mechanism here that needs no bilateral deal. That is the barrier keeping the other four away from anyone small — ~50 negotiated deals in three years, a crawl toll that pays whoever is crawled hardest — and a storefront transaction skips it entirely. Two qualifications keep that from being good news yet. The storefront is the AI company: Google is bookseller, model provider and the party deciding which titles get the badge, so a negotiation is replaced by a platform dependency rather than by nothing. And the post states no publisher terms at all, so what a publisher earns from an AI-enabled title is exactly as unknown as it was under the other four — see edge 10, which this source leaves standing, and new edge 11.

Its reach is also narrower than the others. Per-fetch and per-use were invented for web pages, which have no copy to own. Per-copy licensing works only where the content is already sold as discrete copies — books now, with third-party subscriptions, research reports and textbooks named as next — so it answers the paid-content half of the problem and none of the open-web half this spoke started from.

Where it sits against the existing thesis. great-content-no-longer-works's fork — optimize the channel or exit it — gains a third branch that neither camp named: sell to the machine instead of competing for the human. It does not resolve the fork, and on present evidence it pays almost no one, but it is a different move from both GEO and product differentiation, and the infrastructure for it shipped this year.

The spoke had no page on links until today, which is odd for an SEO corpus and worth naming as it is fixed. Reading the policy first-party (google-link-spam-policy) corrects a thing the trade-press version of this subject gets backwards: Google does not prohibit buying links. It prohibits buying them without saying so — a paid link carrying rel="sponsored" is compliant by the policy's own sentence.

That puts links in the same shape as the two other Google rules this spoke holds first-party. Scaled content abuse is not a ban on AI writing but on mass production without value (google-search-essentials). E-E-A-T is not a score but a description of what raters are told to look for (e-e-a-t). In all three the platform regulates declared intent, and the enforcement question — what actually happens to the undeclared case — is left unstated in every one of them.

One detail is new rather than a restatement: site reputation abuse is enforced differently inside the EEA, where such pages are ranked on their own merits instead of drawing a manual action. The spoke's standing observation is that Google owns both the rules and the meter; here is a place where a regulator has visibly moved the rules, and it is the first instance the corpus holds.

The crawl economy has a second half, and nobody is negotiating it (2026-09-10)

serper sells Google result pages as JSON — organic results with positions, knowledge graph, People-Also-Ask, ten surfaces including Scholar and patents — at $0.30 to $1.00 per thousand queries, live, uncached, with a claimed 1–2 second response. It is a vendor landing page and T3, and it still changes two arguments here.

It prices the input the whole seo-tools market runs on. The suites in that landscape are sold per seat at $129–$140 a month; the raw material underneath is a fraction of a cent per rank check. Building a rank tracker no longer means building crawling infrastructure, which is a statement about where the value in that market actually sits, and it is not in the data.

And it is the mirror image of the spoke's money layer. The paid half assembled since 2026-08-17 — pay-per-crawl, rsl-standard, per-copy-licensing, wikimedia-enterprise — is publishers charging machines to read their content, with standards, a collective and audited revenue. Serper is the same transaction inverted: a third party selling machine access to Google's result pages, priced by the thousand, with Google neither the seller nor a party to it. One half of the crawl economy is being negotiated in public; the other has a checkout page.

The asymmetry is worth stating precisely, because it is not hypocrisy on anyone's part — it is a gap in what the corpus can see. Publishers can meter crawlers because a crawler must fetch their origin. Google's SERP is assembled from other people's pages, and the party best placed to meter access to it is the one whose terms have always discouraged automated querying without, on this evidence, stopping a priced market from forming. Nothing here establishes the legal position, and the vendor page does not raise it. It goes in the open questions below rather than into a claim.

The customer list points somewhere else again. LangChain, CrewAI, Haystack and Jan AI integrations, and "your next AI chatbot" as the first named use case. The buyers of rank data now include systems that never rank anything — agents that need the web, for whom a SERP is a retrieval index. ../agentic-tooling-wiki holds the no-API-key end of the same need. That two audiences this different buy the identical feed is the ai-search-shift showing up in the plumbing.

The display stack gets a remedy, and it is behavioural (2026-09-16)

google-ad-tech-remedies-2026 is the first source here on the other Google auction — the display exchange, not search — and it lands at the remedy stage. Judge Brinkema declined the DOJ's divestiture of AdX and ordered conduct instead: AdX and DFP wired to Prebid, exchange bids to rival ad servers on equal terms, the publisher ad server's auction logic published, AdWords kept out of DFP, and a court-appointed monitor inside Google for six years. Two readings sit side by side. One: a monopolist found liable keeps every asset and gains a compliance department. The other, the court's own: the publishers the case was brought to protect run on the infrastructure a forced sale would disrupt, and an appeal would take years. For this spoke the operative fact is the documentation order — DFP's pricing and bid-adjustment logic must be written down and published — which is the first time any Google auction's rules will exist as a document rather than as quality-score-auction-estimation's inference from outside. Watch whether the search auction is ever held to the same standard.

Distribution is the missing half of the AI-visibility advice (2026-10-05)

content-distribution-system (Simmonds, SEJ, T4) turns the off-site argument into an operating plan. great-content-no-longer-works says move to platforms where the audience is; brand-depth-ai-recommendations says AI recommends brands present across many surfaces. Neither says who does the work. Simmonds does: a named distribution owner, a 50/50 time split instead of 90/10, and a 90-day calendar. His supporting number, that only 10.15% of AI citations land on brand-owned domains (50 B2B brands, his own agency's study), is the strongest quantified version of the off-site claim in the spoke, and it is unreplicated. It sits against the on-site framing of generative-engine-optimization. Open: nobody here has tested whether a distribution calendar moves AI citations; the schedule is argued, not measured.

Contradictions / tensions

Why this spoke's floor is trade press

SEO has almost no peer-reviewed or standards literature, so practitioner trade press is the field's record rather than a shortcut around one. That is a structural fact about the subject, not a sourcing failure, and it sets what this spoke can claim: most pages here are practitioner opinion on a fast-moving target, and they are weighted accordingly.

Ruled on 2026-08-30 (curator's call), and it is now a reporting duty rather than a caveat. The tier ladder gained a decision this spoke's own backlog had been waiting on since 2026-08-28: a named practitioner arguing from experience with no data attached is T4, because T4 grades evidence and there is none. Eleven pages here — six Search Engine Land, five Search Engine Journal — sat unresolved because none of the three re-tier rules fitted them, and they are not being re-graded.

What changed is what the number is allowed to mean. QUALITY.md's T4 no longer reads "trade press / SEO blogspam" but "unevidenced", of which blogspam is one kind and the experience argument is another — and where a field publishes no data, the spoke must say so beside the histogram. So for this spoke, read the T4 count as a measurement of what search marketing publishes, not of what this wiki collected. The industry's own record is argument from practice; there is no body of evidence being ignored here, because there is no body of evidence.

That is the same finding the section above already made, promoted from an apology to a result.

The floor has risen a long way since the caveat was first written. A 2026-06-14 audit found T1 3 / T2 2 / T3 0 / T4 13; as of 2026-08-07 the spoke measures T1 13 / T2 3 / T3 12 / T4 9 across 37 tiered pages. The T1 base is the primary-source core the caveat asked for — Pew's click-through data (pew-ai-overviews-clicks), Google's own Search Essentials and e-e-a-t documentation, and the llms-txt spec — and it is now the largest single tier rather than a rounding error.

What has not changed is the discipline: raise the floor by preferring primary data (Pew-style studies, platform documentation, first-party announcements) whenever a gap can be closed with one, and say so plainly when only trade press exists. The T3/T4 half is still 21 of 37 pages, so the caveat stands even though the numbers behind it have moved.

(Earlier running tally consolidated 2026-08-07 — it had accreted a chain of dated +1 Tn notes and was reporting T1 4 / T4 14, well behind the corpus. The per-ingest history lives in log.md.)

Re-tiered on what pages rest on

The tier audit quoted above was done by masthead, and the rule added to ../QUALITY.md on 2026-08-04 grades a page by what it rests on instead. Re-run over all 36 tiered pages:

T1 12 / T2 5 / T3 3 / T4 16 → T1 13 / T2 3 / T3 11 / T4 9

Where the 16 T4s went:

The rule cut upward too: agentic-commerce-protocol T2 → T1 (an official published spec, where the ladder puts specs regardless of author, same as llmstxt-spec), while gsc-validate-fix-guidance and google-domain-migration-requirements went T2 → T3 — both are SEJ relaying Google, which is the same move as the six above and had been graded differently.

What this changes and what it doesn't. The T4 count nearly halves, and that is not the floor improving — it is the previous number having conflated "the trade press said it" with "the trade press made it up." Six of those pages were carrying Google's own words. The nine that remain are the honest measure of how much of this spoke is unevidenced opinion, and 9 of 36 is still the highest T4 share in the corpus.

The structural point in the caveat above is unchanged and is why: SEO has almost no peer-reviewed or standards literature, so practitioner trade press is the field's record. The floor rises by preferring primary data when a gap can be closed with one — and the clearest available instance is now named: hold MIT's AI Labor Exposure Map directly rather than SEJ's reading of it.

Cross-spoke adjacency

Index — Search Marketing Wiki

Catalog of every page, grouped by schema.org @type. Spine: synthesis (thesis), log.md (history), this file (catalog). Scope broadened 2026-06-03 to the whole search/SEO/PPC landscape (AI-era + traditional). Some wiki-links resolve to bridge nodes in sibling wikis (../agentic-tooling-wiki, ../research-wiki) — intentional cross-wiki links.

DefinedTerm (concepts / mechanisms)

Collection (curated catalogs)

Report (neutral research)

ScholarlyArticle (peer-reviewed)

TechArticle (first-party reference)

Service (paid products)

Organization / WebSite (institutions)

Person

PodcastSeries

SoftwareApplication (tools)

Article / BlogPosting / NewsArticle (sources)

Traditional search marketing (SEO / PPC / affiliate)

Person (added)

Synthesis

Bridge nodes (live in sibling wikis, linked cross-wiki)

agentic-seo-skill · claude-skills-ppc (tools, ../agentic-tooling-wiki) · retrieval-augmented-generation (../research-wiki) · llm-wiki (the LLM-maintained wiki as an artifact, ../research-wiki) · cloudflare (the operator — canonical node in ../cloud-wiki; this spoke cites it as the publisher of crawl-to-refer-ratio and the vendor behind pay-per-crawl)

Related pages

Links to

Linked from