Log — AI Search Wiki
Append-only history. Each entry starts with ## [YYYY-MM-DD] <op> | <title> where
<op> is ingest, query, lint, or split, so grep "^## \[" log.md | tail -5 works.
[2026-06-01] split | ai-search-wiki created from _inbox cluster ai-search-landscape (4 sources)
The cluster reached a clean 4 in one Telegram burst — all on the AI transformation of search: google-io-business-visibility (agentic commerce / business-visibility gap), duckduckgo-no-ai-search (consumer backlash / market fragmentation), ai-content-seo-visibility (AI-era SEO playbook), and brand-depth-ai-recommendations (the GEO mechanism). Too specific for any existing spoke; the SEO/PPC skill packs in agentic-tooling-wiki are tools, not this strategy/landscape topic.
Scaffolded from CLAUDE.template.md; domain = the AI transformation of search & discovery.
Ingested all four (URL-only) as source summaries + 3 concept pages: ai-search-shift (umbrella
event — Google’s May 2026 overhaul), generative-engine-optimization (the central concept),
agentic-commerce. Synthesis frames it as one shift with two reactions (consumers contest AI
search; brands adapt via GEO) and flags the measurement gap + vendor-incentive caveats. Cross-wiki
bridges: agentic-seo-skill/claude-skills-ppc (tools, agentic-tooling-wiki),
retrieval-augmented-generation (research-wiki). _inbox ai-search-landscape records cleared.
Spoke count 7 → 8.
[2026-06-02] ingest | Why Great Content No Longer Works (SEJ / MIT research)
Routed in by the hub (clear AI-era match; not the parked seo-ops traditional-SEO bucket). Created great-content-no-longer-works (Article): AI Overviews extract content → click-through collapse (“digital enclosure of publishing”, Fishkin); MIT AI Labor Exposure Map = 65% of marketing tasks automatable; response = ignore traffic, build inimitable products. Updated ai-search-shift (added the content-enclosure / click-collapse mechanism) and synthesis Contradictions/tensions with a new strategic fork: optimize-the-channel (GEO) vs. abandon-the-channel-and-differentiate. Caveat: SEO-trade op-ed; 65% is a labor-exposure estimate.
[2026-06-03] ingest | How SEO Turns Customer Success into AI-Readable Proof (SEL)
Routed in by the hub (AI-era; explicitly “not traditional SEO”). Created customer-success-ai-readable-proof (Article): Assistive Agent Optimization (AAO) — publish verifiable post-sale proof (quantified outcomes, testimonials, delivery data) in machine-readable form because “agents verify brand claims against the open web.” Framework OPIDC. Updated generative-engine-optimization with an operational-proof bullet (the retrieval-verification side of brand depth, sourced from CS/delivery not marketing). Notably also answers the contrarian great-content-no-longer-works from one side: real outcomes are inimitable proof AI can’t fake. Caveat: SEO-trade op-ed; AAO/OPIDC are coinages.
[2026-06-03] split/broaden | ai-search-wiki scope broadened to all search/SEO/PPC (human directive “B”)
Asked the human (search-marketing-ops cluster hit 3): (a) spin out a sibling vs (b) broaden this
spoke. Chose (b). Absorbed the 3 traditional sources from _inbox: seo-commissioning-workflow,
seo-affiliate-alignment, b2b-ppc-metrics. Added umbrella concept search-marketing
(organic SEO + paid PPC + affiliate; ops/strategy/measurement) framing the wiki as one continuum:
traditional search marketing → ai-search-shift → generative-engine-optimization. Broadened
the CLAUDE.md domain + synthesis (new “Scope: one continuum” section) + index (split AI-era vs
traditional). Updated wikis.md domain/keywords. Dir name kept (ai-search-wiki, historical). Deleted
the 3 _inbox records; the search-marketing-ops cluster is now resolved into this spoke.
[2026-06-03] rename | ai-search-wiki → search-marketing-wiki (human directive “Please rename”)
Renamed the spoke directory to match the broadened scope (now the whole search/SEO/PPC landscape,
not just AI-search). Wikilinks are slug-based so they were unaffected; updated the CLAUDE.md title +
header, index title, hub CLAUDE.md spoke list, and the wikis.md registry block header. Prior log
entries above retain the historical ai-search-wiki name (append-only). Clean site rebuild + verify.
[2026-06-04] ingest | Why So Much SEO Work No Longer Drives Growth (Search Engine Land)
Hub-routed (Telegram). New source summary seo-no-longer-drives-growth (Article, url-only). Thesis: the traditional SEO deliverables (packaged keyword research, high-volume content, standalone on-page) have commoditized and no longer drive growth; teams must reallocate to five non-commoditizable capabilities — entity/brand building, original research, distribution & PR, AI-search visibility, analytical depth. In-house: senior strategists over production headcount; agencies: capability offerings over retainer deliverables.
New concept page seo-operating-model-shift (DefinedTerm) — the labor/org layer (what-work-by- whom) beneath the existing what-to-optimize-for visibility pages. Updated synthesis (new “operating-model consequence” section + a partial reconciliation of the optimize-vs-abandon fork: the five capabilities draw from both camps, so the choice is “drop commoditized production, keep the non-commoditizable work,” not optimize-vs-abandon) and search-marketing (added an operating-model section + cross-links). Index: new DefinedTerm + Article rows. Maps the MIT “65% automatable” finding (great-content-no-longer-works) onto org design. Caveat: SEO-trade op-ed, qualitative only (“18 months across our client base”). No fact contradictions — narrows an existing tension.
[2026-06-07] ingest | Google’s new guidance claims authority over SEO, tools, and AEO/GEO
Hub-routed (Telegram). New source summary google-guidance-seo-authority (NewsArticle, url-only; SEJ, Roger Montti). Google’s new Search Central guidance makes Google’s own docs the benchmark for SEO advice, claims AEO/GEO fall under its purview, warns third-party tools lack Google’s internal ranking data (“predictions … may not happen”), and steers practitioners to Google Search Console as the “first-party tool.” Author questions the motive (imminent algorithm change vs. canonicalizing authority).
New concept page answer-engine-optimization (DefinedTerm) — AEO, the near-synonym sibling of GEO the wiki was missing (the article foregrounds “AEO/GEO”). Updated generative-engine-optimization (new “contested ownership — the platform claims authority” section + AEO cross-link) and synthesis (new “platform asserts authority” section — opacity-from-the-source + inverted vendor-incentive caveat; two new open questions on definitional ownership). Index: new DefinedTerm (AEO) + Article (source) rows. Caveat: SEO-trade reportage of a Google stance; Google’s own canonical-authority/GSC-adoption incentive flagged.
[2026-06-09] ingest | Google says hyphenated domain names are okay for SEO
Hub-routed (Telegram, SEJ, Roger Montti). New source hyphenated-domains-seo (NewsArticle, url): John Mueller (Google) confirms on Bluesky that dashes in domains carry no SEO penalty (Google has indexed domains with up to 61 hyphens). Nuance: the only real cost is user perception (spammy/ typing/trust), not the algorithm — brands like mercedes-benz.com / coca-cola.com / e-verify.gov rank fine. Filed under traditional/technical SEO: updated search-marketing (new “technical-SEO myths” bullet + Related/date). Cross-linked to google-guidance-seo-authority as another instance of the “Google as SEO authority” thread. Index updated. No new concept page; no synthesis change (minor tactical fact). Caveat: staff-writer relay of an off-the-cuff social reply.
[2026-06-09] ingest | +2 foundations (E-E-A-T, llms.txt) — all-spokes cron test
Rich spoke, so a tight, non-padding pass adding two genuinely-absent first-party/standard pieces: e-e-a-t (DefinedTerm, src — Google’s Experience/Expertise/Authoritativeness/Trust quality framework; not a direct ranking factor; the Google-authored root of the brand-depth/credibility thesis) and llms-txt (DefinedTerm, src — proposed /llms.txt curated-markdown map for AI at inference time; the GEO counterpart to robots.txt/sitemaps). Both sharpen “is GEO just SEO rebranded?” — old credibility signals re-aimed (E-E-A-T) + genuinely new plumbing (llms.txt). 18 → 20 pages.
[2026-06-10] ingest | AI Overviews + Google Search Console — all-spokes pass (the product + the meter)
Two pages grounding the thesis’s most abstract points in concrete dated objects (spoke recently grown, so kept to 2). ai-overviews (DefinedTerm, url, Wikipedia) — Google’s Gemini-powered summaries atop results: the ai-search-shift as a product, on >48% of searches by March 2026 (from ~6.5%), ~67.1% desktop screen with a snippet — hard supply-side data under both the click-collapse (great-content-no-longer-works) and GEO-necessity threads; publisher lawsuits + Jan-2026 health restriction sharpen the opacity/platform-power theme. google-search-console (SoftwareApplication, url, Wikipedia) — Google’s first-party Search-performance tool: the meter behind “the only authoritative signal is the one Google exposes” (google-guidance-seo-authority says use GSC over third-party tools); but it reports Search, not AI-Overview/GEO consideration — pinning the measurement-gap open question to a specific tool (AI recommendation stays unmeasurable → unfalsifiable advice risk). Folded into synthesis (new “product and the meter” section) + index (DefinedTerm + new SoftwareApplication group). No contradictions — both deepen existing threads. 20 → 22 pages.
[2026-06-12] ingest | Pew Research — AI Overviews suppress clicks
All-spokes daily expansion. Added pew-ai-overviews-clicks (@type Report) — the wiki’s first neutral, non-trade/non-vendor data point, directly addressing the standing vendor-incentive caveat. Pew field study (n≈900 US adults sharing browsing data; 58% hit an AI summary in March 2025): AI-summary pages halve link clicks (8% vs 15%), only 1% click a link inside the summary, and 26% vs 16% session-ending. Independently corroborates the ai-overviews click-collapse / great-content-no-longer-works thread without trade incentive; tips the optimize-vs-abandon fork’s shared “click channel is collapsing” premise toward confirmed. Does NOT close the GEO-measurability gap (measures clicks lost, not model consideration). Updated synthesis (vendor-caveat + fork notes), index (new Report group). 1 new page.
[2026-06-15] ingest | Google Search Essentials — T1 ruleset anchors the platform-authority thesis
Quality cycle, T1-floor raise (the spoke was the most trade-press-heavy: T1 3 / T4 13). Added the
canonical first-party ruleset as a source: google-search-essentials (TechArticle, T1) —
Google’s three pillars (technical requirements + 16 named spam policies + best practices). Grounds two
threads previously carried by SEO trade press: (a) the platform-authority thesis
(google-guidance-seo-authority) now points at Google’s actual published benchmark, not commentary
about it; (b) the AI-content thread — 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 per se — first-party grounding for ai-content-seo-visibility. Opacity theme reinforced
from the source (“meeting the requirements doesn’t mean Google will crawl/index/serve”). Threaded into
synthesis (Foundations section retitled; Source-base caveat T1 3→4). Quoted from the Search Essentials +
Spam Policies pages. Found via direct fetch. Linked from synthesis + index (new TechArticle section). 1 new page.
[2026-06-16] ingest | Google Index Checker use cases (hub-routed, Telegram) — the measurable floor
Ingested the OfficeChai how-to “8 Google Index Checker Use Cases Beyond New Blog Posts” (T3, vendor- adjacent — built around promoting Rapid Index Checker). Created source summary google-index-checker-use-cases (Article) + a tool-neutral concept node search-indexation (DefinedTerm): whether a URL is in Google’s index at all. Gap-relevance: it pins a measurable floor under the recurring measurement-gap question — indexation is checkable (google-search-console‘s index-coverage report or third-party checkers), unlike AI consideration which isn’t. Wired both into google-search-console (index-coverage bullet + Related) and folded a sentence into synthesis distinguishing the verifiable traditional layer from the unmeasurable AI-recommendation layer. Lands on the traditional/operational side of the spoke (technical-SEO hygiene). Entity discovery: no new nodes — publisher OfficeChai and the named tool are too thin/promotional to page (evidence-only); google already exists cross-wiki (llm-providers-wiki). +2 pages.
[2026-06-17] ingest | Why Publishing More Content Is Making Your SEO Worse (SEJ / Shelby)
Routed in by the hub (clear AI-era search/SEO match; runner-up none). Created publishing-volume-hurts-seo (Article, T4, url-only): the “more pages = more chances” coverage logic inverts under AI retrieval — volume now hurts via semantic dilution, internal vector competition (pages “competing for embeddings, not just rankings”), crawl waste, and entity-coherence erosion; remedy is “publish with intent” / consolidate. Added two DefinedTerm nodes: authority-density (the metric that replaces page-count — “clarity, not volume”; supply-side mirror of brand-depth-ai-recommendations) and keyword-cannibalization (classic SEO self-competition, generalized to semantic/vector overlap — a traditional→AI-era bridge). Updated brand-depth-ai-recommendations (Related + a supply-side-mirror line) and synthesis (new “volume inverts — density beats coverage” thread under the brand-depth mechanism; lands optimize-the-channel on the central fork; tier audit T4 13→14). Entity discovery: no new nodes — spoke convention pages no columnist authors (0 Person nodes across ~15 sources) and no publisher nodes; kept Carolyn Shelby / CSHEL inline. Caveat: conceptual op-ed, zero quantitative evidence; harm + remedy unmeasured. +3 pages.
[2026-06-18] ingest | Google Tightens Requirements For Domain Migrations (SEJ)
Telegram-dropped; hub-routed → search-marketing (runner-up: none). Source summary
google-domain-migration-requirements (NewsArticle, url, T2 — SEJ/Roger Montti relaying a
first-party Google Search Console doc/changelog change). New concept domain-migration (DefinedTerm —
site move / Change of Address). The change: a migration now needs a Change of Address for every
verified domain variant (each subdomain + www/non-www, even unused ones), since stray linking makes Google
crawl forgotten permutations. Gap-relevance: fills out the spoke’s thin traditional technical-SEO ops
half and reinforces the platform-authority / first-party-GSC thread (google-guidance-seo-authority,
google-search-console); a botched migration shows up as search-indexation loss. Integrated into
synthesis (new “traditional technical-SEO ops” subsection + tier-tally note) + index; cross-linked GSC.
Entities: none noded — author (Roger Montti) / publisher (SEJ) per house style for trade-press ingests
(no existing SEJ author/org nodes). freshness: fresh. Rebuild + commit follow.
[2026-06-18] ingest | DefinedTerm enrichment pass (subagent)
Deepened the five thinnest DefinedTerm pages with new first-party sourced facts (subagent pass):
- domain-migration — added Google’s first-party site-move mechanics from google-site-move-doc: 301/308 keep PageRank, ≤3-hop redirect chains, keep redirects ~1yr for signal transfer, HTTP→HTTPS exempt, weeks-to-longer timeline, change-one-thing-at-a-time.
- llms-txt — added the actual spec structure from llmstxt-spec (Jeremy Howard, 2024-09-03):
H1-only-required, blockquote summary, H2 link-lists, the
Optionalskip-section; nollms-full.txtin spec. - keyword-cannibalization + authority-density — grounded “consolidate” in Google’s canonicalization doc google-consolidate-duplicate-urls (split signals/wasted crawl; redirect > rel=canonical > sitemap), framing density as the embedding-space re-reading of classic canonicalization.
- agentic-commerce — added the shipping ACP standard from agentic-commerce-protocol: merchant keeps business logic + own PSP, OpenAI not merchant-of-record, discovery runs off product feeds (the GEO substrate); payments-rails contest noted as agentic-tooling-wiki territory.
New source pages: google-site-move-doc (T1), google-consolidate-duplicate-urls (T1),
llmstxt-spec (T1), agentic-commerce-protocol (T2). All four fetched live. Blocked: openai.com/index
(403) → used developers.openai.com ACP docs instead. Index updated (TechArticle group), updated bumped on
all five concept pages. No build/commit (orchestrator handles).
[2026-06-20] ingest | floor-raise — Google’s first-party AEO/GEO + third-party-tools guidance (T1)
Quality-cycle follow-up: the spoke’s weakest floor (T4×14, flagged 2026-06-19) is partly upgradeable —
several T4s only reported a Google/MIT primary. Replaced the trade-press grounding of the AEO/GEO
pages with the actual Google Search Central primaries (found via WebSearch, fetched live): new
google-ai-optimization-guide (T1 — AI features = core ranking + RAG/query-fan-out; SEO fundamentals
hold; mythbusts llms.txt/chunking/AI-rewrites/structured-data) and google-third-party-seo-guidance
(T1 — Google doesn’t endorse 3rd-party tools; scrutinize “AEO/GEO tools”; use Search Console). Upgraded
answer-engine-optimization T4→T1 (re-grounded its url: from the SEJ op-ed to Google’s primary),
deepened generative-engine-optimization with Google’s deflationary stance, and corroborated
llms-txt with Google’s explicit “Search doesn’t use llms.txt.” Net floor effect: search-marketing T4
14→13, +2 T1. The remaining T4s are genuine practitioner analysis (no higher primary) — left honest.
[2026-06-22] ingest | X-Frame-Options matters for SEO (SEJ / Mueller)
New source page x-frame-options-seo (T4, freshness volatile) — SEJ relaying a John Mueller (Google)
Reddit comment: of the HTTP security headers, the only one he could imagine affecting SEO is the
iframe-blocking header (X-Frame-Options / CSP frame-ancestors); the benefit is defensive (stop a
third party ranking your content framed under their domain), not a positive ranking lever — a
technical-SEO hygiene point. Same genre as the existing hyphenated-domains-seo
(SEJ-relays-Mueller-on-a-narrow-technical-SEO-myth): folded into synthesis under Traditional technical-SEO
ops and the “Google as SEO authority” thread (google-guidance-seo-authority). Fetched live via
WebFetch. Index updated (Article → Traditional search marketing); synthesis.md updated bumped. No new
Thing page — the mechanism is captured in the source summary and cross-linked to existing nodes. (No
build/commit here — orchestrator handles.)
[2026-07-08] ingest | AI-powered content-gap-analysis workflow (SEL / Sara Vicioso)
Routed from Telegram (hub). New source ai-content-gap-analysis-workflow (HowTo, T3, searchengineland.com) — Sara Vicioso’s 6-step workflow (choose competitors → gather Semrush/GSC/GA4 → Claude clusters by intent/funnel/relevance/authority → 5-factor score → page-level briefs → measure & re-run quarterly), fed via CSV or MCP, with a self-validation prompt (“if GSC/Analytics changes your recommendation, explain why”). New concept node content-gap-analysis (DefinedTerm) — the classic find-competitor-gaps-and-prioritize technique; captures the traditional workflow + what AI changes (the organizing step, not the data pull). Integrated into synthesis under The operating-model consequence as a task-level instance of seo-operating-model-shift (commoditized row-scrolling → the model; strategy stays human; validate against first-party GSC); also lands on the clarity-over-volume side. Cross-spoke: the Claude+MCP mechanism is agentic-tooling-wiki territory (runner-up) — noted on the source page, routed here because the SEO practice is the dominant substance. T3 (trade-press practitioner how-to, vendor- specific, no primary data). Ran avoid-ai-writing. +2 pages.
[2026-07-08] ingest | Search Console platform properties for social/video (Google Search Central)
Routed from Telegram (hub). Fetched via firecrawl (WebFetch returned only the archive shell — JS page). New source gsc-platform-properties (TechArticle, T1, developers.google.com; by Moshe Samet, PM Lead GSC, 2026-07-07) — a new GSC property type for social/video accounts (Instagram, TikTok, X, YouTube): Performance (clicks/impressions by post & query, exportable), Insights (trends/top posts/discovery), and Achievements (milestones), showing how off-site posts perform on Search & Discover even without a website; follows a Dec-2025 experiment; rolling out over the coming weeks. Updated the google-search-console Thing page (“What it exposes” + note that this extends GSC’s meter beyond your own domain). Synthesis: folded into the product-and-the-meter thread — “the platform owns the meter” now reaches off-site content, same bound (measures Search/Discover clicks, not AI-Overview/GEO consideration). Entities (Moshe Samet, Google) deferred per the spoke’s standing posture. No structural move → no verify. Ran avoid-ai-writing. +1 page.
[2026-07-10] ingest | 8 Semrush Alternatives (AnswerSocrates blog) (hub-routed, Telegram)
Ingested the AnswerSocrates “8 best Semrush alternatives” listicle — the wiki’s first map of the SEO-software landscape. New semrush-alternatives-answersocrates (BlogPosting, source, T4 — vendor listicle, self- ranked #1, conflict of interest flagged) + new concept seo-tools (all-in-one suites Semrush/Ahrefs/Moz vs budget Ubersuggest/Mangools vs niche AnswerSocrates/SEO Scout; trust-the-tool-not-the-ranking; question-mining feeds GEO). Did not create 8 tool pages (disproportionate for a T4 promo source) — named inline, earn pages on a substantive source. Synthesis: added the software-layer note to the scope section, tying it to the Google-third-party-guidance caveat and the traditional→AI-era (question-research → GEO) seam. Prices volatile. avoid-ai-writing applied (clean). +2 pages.
[2026-07-10] ingest | groas — autonomous AI Google Ads management (hub-routed, Telegram)
Ingested groas.com (vendor landing page; WebFetch 403 → Firecrawl). New groas (SoftwareApplication/ Organization, source, T4 — self-promotional, all figures unaudited). Product: fully-autonomous ML agents running a Google Ads account end-to-end (bids/keywords/budget/ad-copy + intent-matched dynamic landing pages) for ROAS; “trained on $500B+ profitable Google spend” (the “$500B engine”); self-reported $100m+/yr across 500+ clients; sold as a full PPC-manager/agency replacement. Wired into the paid-search half: search-marketing (new paid-automation bullet), b2b-ppc-metrics (backlink + “automation stress-test” — groas optimizes the very ROAS/conversion metrics that page calls unreliable), and the seo-operating-model-shift carried to PPC. Synthesis: new “Paid-search automation” thread + T4 tally bump. The Telegram title’s “ChatGPT Ads” is NOT on the homepage (100% Google Ads) — recorded as an unsubstantiated forward claim toward agentic-commerce. Entities: only groas itself substantive (testimonial names too thin) — no extra nodes. No structural move → no verify. avoid-ai-writing applied. +1 page.
[2026-07-14] ingest | Scaled AI Content Often Fails & Google’s Crawl Economics Explain Why (SEJ, hub-routed, Telegram)
Ingested SEJ (Dan Taylor, VIP Contributor): scaled AI content fails on crawl economics, not AI-dislike. Google’s finite compute is allocated by inventory-vs-utility / demand / authority; thin-page floods exceed the crawl budget Google will spend → burst-crawl → freshness boost → decay → de-index for pages not re-crawled within ~75–140 days, with low-value clusters getting reduced crawl frequency (compounding). Scaled-content-abuse manual actions (keyword-swap / mass auto-translation / summarize-without- gain) read as resource-economics enforcement. New pages: source scaled-ai-content-crawl-economics (Article, T3 — analysis of Google’s crawl-budget docs, no original data; the day-window is the author’s number); new mechanism node crawl-budget. Wired the supply-side half into the volume-inversion thread: updated publishing-volume-hurts-seo (crawl-inefficiency bullet now has its own node), search-indexation (indexation is not permanent — recedes via crawl decay), authority-density (crawl = what buys continued indexing, not only embeddings). Synthesis: extended “Volume inverts” with the crawl-economics supply-side gate (retrieval-side harm + supply-side gate = the full inversion); nudges the density family off pure conceptual reasoning toward documented crawl mechanics (still T3). Followed spoke convention: no author/publisher nodes (this spoke pages none). avoid-ai-writing self-pass (clean, house voice). +2 pages, 3 updated.
[2026-07-19] ingest | Google Search Central (YouTube) + “When to use Validate Fix” (SEJ)
Two interlocking sources routed here by the hub, ingested together. (1) google-search-central — the
YouTube channel URL landed via Telegram, but it’s really Google’s whole Search Relations program
(ex-”Webmasters”): docs + channel + search-off-the-record podcast + SEO Office Hours. Typed
[Organization, WebSite], source: true, url=youtube, T1. Fetch caveat: the channel index is JS/anti-bot
blocked — WebFetch and Firecrawl both returned only YouTube’s shell (only the “Welcome to Google Search
Central” featured video was readable), so the page is grounded in uncontested public record, not a scrape;
noted on the page per the hub fetch-fail rule. (2) gsc-validate-fix-guidance — SEJ relaying a
search-off-the-record episode (Mueller/Splitt): GSC’s Validate Fix only
samples URLs + accelerates a recrawl, doesn’t verify the fix; skip it unless you fixed a real site-wide
error. Typed [NewsArticle], T2 (first-party in substance, podcast-relay in form — cf.
google-domain-migration-requirements).
New pages: google-search-central, gsc-validate-fix-guidance, search-console-validate-fix (mechanism), plus
entities john-mueller (Search Advocate), martin-splitt (Developer Advocate), search-off-the-record
(PodcastSeries). Reused cross-wiki google (Organization, llm-providers-wiki) — not duplicated. Updated
google-search-console (Validate Fix bullet + links), hyphenated-domains-seo + x-frame-options-seo
(now link the john-mueller entity they already named). Synthesis: new subsection “The apparatus behind
‘Google as SEO authority’” — names the program + humans, adds the T1-doc → T2-podcast → T4-social tier
ladder on the same first-party authority, and folds Validate Fix into the meter (google-search-console) +
crawl-economics (crawl-budget) threads (the button’s only lever is recrawl timing). avoid-ai-writing
self-pass (house voice). +6 pages, 4 updated. Per-route verify deferred per hub policy.
[2026-08-04] lint | re-tiered on what pages rest on, not who published them
Second application of the rule added to ../QUALITY.md on 2026-08-04 (music-tech-wiki was the first),
run at the curator’s instruction over all 36 tiered pages here.
T1 12 / T2 5 / T3 3 / T4 16 → T1 13 / T2 3 / T3 11 / T4 9
The 16 T4s split three ways. Six → T3, because Google is the vendor in this domain and a trade piece carrying its guidance is the vendor speaking through a masthead: google-guidance-seo-authority (new Search Central guidance), hyphenated-domains-seo and x-frame-options-seo (John Mueller confirmations), google-io-business-visibility (I/O demos), duckduckgo-no-ai-search (a shipped product reported), and groas — a company’s own landing page, which was never T4 material.
One → T2: great-content-no-longer-works, the only page here resting on a named third-party measurement (MIT Work Analytics Lab, 65% of marketing-specialist tasks automatable). Tier note added to the page, including the caveat that what this wiki holds is SEJ’s argument built on MIT’s number rather than the map itself.
Nine stay T4 — practitioner argument with no data and no first-party source, plus one listicle (semrush-alternatives-answersocrates).
The rule cut upward as well: agentic-commerce-protocol T2 → T1 (an official published specification, where the ladder puts specs regardless of author — same grade as llmstxt-spec); gsc-validate-fix-guidance and google-domain-migration-requirements T2 → T3, both SEJ relaying Google, which is the same move as the six above and had been graded inconsistently with them.
Read the new number honestly. The T4 count nearly halving is not the floor improving. The old count conflated “trade press said it” with “trade press made it up”, and six of those pages were carrying Google’s own words. The nine that remain are the real measure, and 9 of 36 is still the highest T4 share in the corpus. The structural reason stands: SEO has almost no peer-reviewed or standards literature, so practitioner trade press is the field’s record.
[2026-08-05] ingest | awesome-ai-lead-generation — the outbound half arrives (curator-directed)
Routed from the hub. The routing history is part of this entry, because it should be. The router
parked this source in _inbox/ earlier today as out of scope: this wiki owns search, and cold email,
scraped contact lists and voice diallers are the other direction of the funnel. The curator overrode the
park and directed it here. The park record is deleted and the hub log.md carries both entries.
Pages: awesome-ai-lead-generation (Collection, source, T3), lead-generation (the practice — signal → resolution → contact → personalization), cold-email-deliverability (the gate on the contact step). synthesis gains “The outbound half arrives”, which states the case for the fit and the case against it rather than resolving one into the other, plus three open questions.
The connection that makes it more than a filed link. Someone posting “what tool does X?” in a subreddit ran a query nothing indexes. That is the demand search-marketing competes for, reaching sellers through a channel where monitoring replaces ranking. The shared input is concrete: seo-tools already records question-and-intent research feeding both keyword SEO and GEO, and content-gap-analysis mines those questions to decide what to publish — social listening mines them to decide whom to message.
The second one is a hypothesis and is labelled as one. cold-email-deliverability and google-search-essentials describe the same gatekeeper structure, and the search side of it is documented (publishing-volume-hurts-seo, scaled-ai-content-crawl-economics: density beat coverage once generation got cheap). Whether warming more inboxes ends the way publishing more pages did is unsourced on the email side — no source in this corpus documents mailbox-provider policy, SPF/DKIM/ DMARC, or anti-spam law. Recorded in open questions, kept out of the thesis.
Quality gate. T3, justified: single pseudonymous maintainer (toofast1), no inclusion criteria, no
outcome data on any of the 24 tools, and one entry (Leado) marked “Editor’s choice” with no disclosed
relationship — the semrush-alternatives-answersocrates problem again. Two entries contradict the
list’s own outbound thesis (Fireflies.ai is a meeting recorder by its own admission; Taplio generates
inbound LinkedIn content). Dedup: nothing in the corpus touched lead-gen, outbound or cold email —
grep found zero hits, so all three pages are new rather than refreshes. Gap-relevance is the weak leg:
this closes no open question the wiki had, it opens three.
Entity discovery: none created, deliberately. The 24 vendors are the source’s subject, and
../ENTITIES.md excludes the subject from discovery. The only agent is a pseudonymous GitHub handle with
no evidenced affiliation, which the relevance gate makes a plain mention. entity-index not run — no
candidate cleared the gate to look up.
Standing caveat: the entry roster is a volatile 2026-08 snapshot of a market that churns.
[2026-08-08] ingest | How to spot an emerging category in search data (Search Engine Land)
Routed from the hub (Telegram). Claire Taylor, 2026-08-06. T2 — trade-press masthead, but the author pulled and analysed SE Ranking US/UK keyword data rather than relaying a vendor, which is what the tier rule keys on.
New pages: emerging-category-search-signals (source), category-emergence (practice), plus entities claire-taylor and search-engine-land.
Why it earns a concept page. This spoke is almost entirely about competing for demand that exists. This works a step earlier — detecting a category while it forms — and the signals are shape, not size: authentic phrasing absent from the tools, formal vocabulary (standards, regulations, job titles) before plain language, several labels contesting the same idea, difficulty below demand, and SERPs where boutiques outrank the Big Four. Three anti-self-deception rules make it a method: fixed cohorts, 12-month window, commercial-intent terms.
Best evidence in the piece is the same category in two markets — AI governance at US difficulty 68–72 against UK 25–32 — the window closing in one and open in the other.
Folded into synthesis on two threads: it gives seo-operating-model-shift a concrete non-commoditizable task, and it sharpens rather than closes the measurement gap — everything here is counted in searches and rankings while the spoke’s #1 edge asks for share of AI recommendation. Recorded plainly: if ai-search-shift holds, this is a better instrument aimed at a shrinking surface, and the article does not say so.
New edge (5): a false-positive rate for the method. Every case is told forward from a category that emerged; nobody reports the clusters that showed all five signals and went nowhere.
Cross-spoke, noted not duplicated: the primary case study is AI governance, which is its own spoke here — the figures are demand-side evidence that field is in early-adopter growth. And the method is an empirical instrument for research-wiki’s diffusion/adoption cluster (Rogers, Moore, hype cycle, Bass), which holds the models and no way to observe one live.
Entities: 2 created.
[2026-08-08] ingest | GEO: Generative Engine Optimization (Aggarwal et al., KDD 2024)
Research pass (20-edge run), hub edge #4 — the spoke’s #1 growth edge and, as its own text said, “the edge the whole GEO thread rests on.” Closed. geo-paper new, T1.
The omission it fixes is embarrassing and worth logging plainly: generative-engine-optimization is this wiki’s central concept and the page was built entirely from trade coverage about the paper. The paper itself was never ingested.
What it supplies: a visibility metric defined over the generated answer — position-adjusted word count, plus a subjective-impression measure — which is the consideration metric the spoke kept saying nobody had. Plus GEO-bench (~10,000 queries, nine datasets) and results: up to 40% visibility gain, largest from adding verifiable statistics, credible quotations and source citations, varying by domain.
Tension recorded rather than smoothed: brand-depth-ai-recommendations holds that entity salience drives recommendations and citations are outcomes, not levers. The GEO paper’s measured interventions are passage-level content edits. Both can hold — different mechanisms — but the only controlled measurement in this spoke is of the passage-level kind, and the page now says so.
Not recorded: C-SEO Bench (arXiv 2506.11097) came back with an internally inconsistent extraction (it misnamed GEO and its stated conclusion did not hang together). Named as the next read, no finding taken from it. Also considered and not ingested: Semrush’s 2026 AI Visibility Index — 126M prompts, Jan–Apr 2026, real figures (ChatGPT ~15 sources/response vs Gemini ~3; only 36 brands held top-100 visibility on all four platforms) — but it publishes no methodology, and its press release doubles as a sales argument for Semrush’s own tooling. A measurement nobody can check does not close the edge this spoke opened because GEO claims were uncheckable.
[2026-08-08] ingest | C-SEO Bench — and a same-day correction to the GEO ingest above
c-seo-bench new (T1, Puerto et al., NeurIPS 2025). Read after the geo-paper ingest earlier today, once a local PDF extractor made the paper readable — the WebFetch summary of it had been internally inconsistent and, correctly, nothing had been recorded from it.
Reading it properly reverses this morning’s edge closure. geo-paper was logged as closing the spoke’s #1 edge — a consideration metric measuring recommendation share rather than clicks. C-SEO Bench shows the metric does not measure that:
“The differences are due to the metrics used. They report their main results as word count… this metric does not measure the LLM preference, contrary to the citation ranking that we use.”
Its results across two tasks, six domains and varying adopter counts: C-SEO methods “largely ineffective” and often harmful — Statistics lowers ranking in 19 of 24 settings; 26 of 30 product- recommendation cases on Haiku 3.5 show significant negative effects; 19 of 30 QA cases on o4-mini. Retrieval ranking dominates, and gains shrink as adoption rises (zero-sum).
Edge #1 reopened, with the lesson written into it: it was closed on a proxy metric nobody had checked, which is the exact failure the edge exists to prevent. Both papers kept — this is a single-actor result in a narrow domain failing to generalise, not a discredited paper.
generative-engine-optimization carries a warning at the top now: most advice on that page descends from interventions this benchmark found ineffective or harmful.
[2026-08-12] ingest | The auction under the paid half (quality cycle)
Coverage edge 6 closed with both halves it asked for. The spoke had b2b-ppc-metrics and groas — what an advertiser measures — and no page on the mechanism producing those numbers.
gsp-auction-paper (T1, Edelman, Ostrovsky & Schwarz, American Economic Review 97(1):242–259, read in full from the authors’ 25-page copy). Held: the commercial scale (over 98% of Google’s $6.14 bn 2005 revenue came from GSP auctions); the three generations (batch cost-per-impression → Overture’s 1997 generalized first-price → Google’s second-price AdWords Select, February 2002); the instability example (bids climbing $2.01 → $2.02 → $2.03 with no pure-strategy equilibrium, and revenue falling to 2.02 cents per click when one bidder automates against slow humans); and the paper’s main result — GSP is not VCG, has no dominant-strategy equilibrium, and truth-telling is not an equilibrium of it. In the worked example the top advertiser pays $800 under GSP against $600 under VCG. Also held: quality score enters as a rescaling (bid × γ), and Google’s estimated CTR is computed conditional on attaining first position.
google-ad-rank (T3, Google Ads Help). The first-party half, tiered T3 with the reason on the page: this is the seller describing its own auction, and it is promotional in places. Six factors, one composite Quality Score, no published weights.
search-ad-auction (concept) synthesizes the two and names what changed since 2007: quality became a gate rather than a multiplier (an Ad Rank threshold can exclude an ad at any bid), the auction became contextual (location, device, time in the rank, so a keyword has no single clearing price), and the formula stopped being published — a trend the 2006 authors already noted.
Folded into synthesis, not left as a note. The 1997 → 2002 arc is this spoke’s own thesis, dated twenty years earlier: mechanism rewards gaming, industry invests in gaming, platform changes the mechanism. And one asymmetry worth holding against the GEO argument — the ad auction has a peer-reviewed analysis from outside the platform, which generative-engine-optimization does not. That is a difference in available evidence, not only in maturity.
Freshness. pew-ai-overviews-clicks and hyphenated-domains-seo regraded stable — a dated
field study and a dated news article; re-reading either URL returns the same text, so the flag could not
be cleared by anyone. ai-overviews stays volatile and past window: it tracks a live product and a
real re-verification is owed. Carried, not hidden.