Zyppy — What Wins Google in the Zero-Click / AI Era (Cyrus Shepard)

Source: raw/newsletter-zyppy-signal-12ab477b85.md (5 Data-Backed Features of Websites Winning Google in 2026), raw/newsletter-zyppy-signal-76c9b7236c.md (17 Content Types to Survive Google’s Zero-Click Future), raw/newsletter-zyppy-signal-5ede0ab978.md (Content Type Frameworks) — Cyrus Shepard, Signal by Zyppy. Published Apr 2026.

Two linked Shepard pieces on the strategic question above the AI-citation tactics: as Google zero-click and AI answers eat commodity traffic, which sites and content types actually keep winning? The throughline: Google has shifted from ranking “good content” to rewarding what AI can’t replicate — proprietary, experience-based, niche, task-completing content.

Evidence class (be precise): the “5 features” piece is Shepard’s own correlational study — ~400 sites, manual binary classification of features, Spearman correlations of 0.21-0.39. Shepard states it plainly: “correlational and rely on manual classification, so they indicate association and cannot establish causation.” It also measures general Google traffic change, not AI citations specifically. The “17 content types” piece is an expert-judgment catalog (“in our judgment”), not a study. This sits below the AI-citation cluster’s correlational tier in rigor (single practitioner, manual coding, small N) — read it as informed content strategy, not measured effect sizes.

Key Takeaways

• 5 site-level features predicted winning vs losing Google over 12 months (Spearman correlation; % of winners vs losers showing the feature):
  • Offers a Product or Service — r=0.391 (winners 70.2% / losers 34.6%). Highest correlate. Owned products beat third-party platforms; services/subscriptions/digital goods count.
  • Allows Task Completion — r=0.381 (83.7% / 50.2%). Users can do the thing on-site (buy, book, calculate, practice), not just read about it.
  • Proprietary Assets — r=0.357 (92.9% / 57.1%). Something others can’t replicate: databases, UGC, software, first-party reviews/photos. “The AI moat.”
  • Tight Topical Focus — r=0.250 (75.9% / 61.3%). Note: plain “topical focus” showed no winner/loser difference; only tight focus did. Niche beats broad.
  • Strong Brand — r=0.206 (32.6% / 16.1%). Scored via branded-navigational-term volume + share of traffic (Ahrefs top-20 keywords). Recognizable ≠ strong brand; it’s about traffic people seek out by name.
• The features are additive — you usually need several to win. Win rate by feature count: 0 → 13.5%, 1 → 15.4%, 2 → 22.0%, 3 → 30.7%, 4 → 68.1%, 5 → 69.7%. The cliff between 3 and 4 features is the headline.
• Three features Shepard expected to matter but didn’t correlate: demonstrating first-hand experience, hosting UGC/community, and uniqueness of information. His read: likely “baked into the algorithm long ago” rather than un-rewarded — a larger study might surface them. (An honest null worth preserving.)
• 17 content types most likely to survive zero-click (16 are enumerated in the recovered body; the source’s numbering skips #8), rated by strength + effort — see the catalog below. They share the same DNA: proprietary, experience-based, niche, task-completing. None are “get out of jail free” — done at average effort they still lose.
• Zero-click is already here for many. Shepard cites Datos/SparkToro: ~60% of Google searches end without a click to a third-party site. But traffic “isn’t going away” — it’s concentrating toward sites with these features.
• Pro business-model rubrics exist (gated). The companion “Content Type Frameworks” piece maps these content types to 6 business models — SaaS, Affiliates, E-commerce, Local Services, Media/Editorial, Marketplace/Directory — as a Pro-subscriber publishing rubric. Content is behind the paywall; only the model list is public.

The 17 Content Types (by strength tier)

Very Strong

• Owned Audience (Highest Effort) — email/SMS/in-app distribution you control (Morning Brew, Stratechery, Lenny’s). More strategy than format; the gold standard.
• Transaction Pages (Highest Effort) — pages that complete an action: book, subscribe, buy (Hipcamp, Chewy, Sweetwater). Among the strongest zero-click pages even under agentic commerce.

Strong

• Original Research (High) — proprietary data you originated (ChartMogul, Carta Data, Orbit Media). AI cites it and users want the source; journalists link it.
• UGC Communities (High) — forums/communities (FlyerTalk, Mumsnet, Rick Steves). Dual moat: retention + experience-based content Google favors.
• Creator Video & Podcast (High) — host/brand/format that people seek by name (Practical Engineering, MKBHD, Fireship).
• In-Depth Reviews & Tests (High) — firsthand evidence, repeatable protocols, teardowns (RTINGS, BabyGearLab, Tom’s Hardware). AI can summarize but not reproduce firsthand testing.
• Brand Pages (Low) — About/Trust/Social-Proof/Contact (Buffer, Bankrate editorial policy). Foundational for entity trust, low traffic.

Moderate

• Directories & Databases (High) — must be supplemented with proprietary/first-party data + freshness (AlternativeTo, Clutch, Nomads).
• Expert Perspective (Medium) — opinion anchored in insider experience/data (a16z, Not Boring, Tomasz Tunguz).
• Templates (Medium) — downloadable frameworks tied to a product (Canva, Miroverse, Airtable). Harder-to-replicate = more valuable; aids lead-gen + retention.
• Case Studies (Medium) — before/after with exposed metrics + process (Vanta, Webflow, Shopify). Most under-deliver by being promotional not instructional.
• Original Reporting (High) — fast/in-depth/exclusive news (The Information, The Verge, Reuters). Commodity news loses to AI summaries.
• Support & Document Pages (Low) — canonical docs/help (Plausible, Mux, Supabase). Teaches LLMs your product.

Weak

• Guides & Explainers (Medium) — at high risk of AI-summary displacement; survive only with proprietary data/differentiation (Nerd Fitness, Examine, HubSpot).
• FAQs & Glossaries (Low) — easiest to replicate with AI; work only as canonical industry definitions (Moz, Cloudflare DNS, Agile Alliance).
• Lists & Roundups (Low) — under scrutiny for fake self-serving lists; survive only with original testing + published scoring criteria + niche focus.

What Publishers Should Do

A winning site tends to: offer a unique product/service; help users do something (buy/download/search/book); stay in a niche of deep expertise; own hard-to-replicate assets (datasets, products, images, UGC); and become a destination via its own brand. Commodity content is dying — but not equally; execution quality still decides within each type.

Methodology & Caveats

• Dataset: seeded from Lily Ray’s Dec 2025 core-update winners/losers analysis (built on Sistrix visibility). 12-month YoY traffic change per site via the Ahrefs API. Every site hand-classified against a fixed feature set (binary flags). Brand strength from Ahrefs top-20 keywords (branded-navigational count + traffic share).
• Stat used: Spearman rank correlation (tolerant of traffic-data outliers).
• Limits (Shepard’s own): correlational + manual classification → association not causation. Measures general Google traffic, not AI citations. Vendor-adjacent data (Ahrefs/Sistrix). Correlations are weak-to-moderate (0.21-0.39).

Open Questions

• Do these general-Google features transfer to AI-citation outcomes specifically? The study measures organic traffic change, not citations — the AI-era framing is Shepard’s interpretation, not a measured link.
• Would the three null features (first-hand experience, UGC, uniqueness) correlate at larger N? Shepard suspects yes; unresolved.
• What’s in the gated business-model rubrics? Only the 6-model list is public.

Related

• Zyppy AI Citation Playbook — the tactical layer (how to earn citations) beneath this strategic layer (what to build at all).
• Zyppy AI Citation Ranking Factors Meta-Analysis — Brand/Entity Trust and Topical Authority appear in both; here as site strategy, there as citation factors.
• How Google Click Signals Drive Rankings & AI Answers — the engagement mechanism behind “Task Completion” and “Strong Brand” (last-longest-clicks, brand search).
• Datos + SparkToro State of Search — the zero-click market data Shepard leans on (~60% no-click).
• Zyppy — Preparing for AI Shopping & Agentic Commerce — the e-commerce cut of “Transaction Pages” and “Strong Brand.”
• SEO Content Ops Fundamentals — Google’s production-method-agnostic quality/E-E-A-T stance that underpins “commodity content is dying.”

Try It

1. Score your site on the 5 features (binary). If you hit 0-3, you’re in the 13.5-30.7% win band; getting to 4 is the cliff to 68%. Pick the cheapest missing feature to add (often Task Completion via a tool/calculator, or Proprietary Assets from data you already have).
2. Classify your top content by the 17 types. Anything in the “Weak” tier (guides, FAQs, listicles, roundups) needs proprietary data or first-party testing to survive — or should be deprioritized.
3. Add one task-completion element to an informational page that currently sends users elsewhere (calculator, quiz, searchable database, booking) — “own the next step of the user journey.”
4. Audit branded search share in GSC (branded filter). If almost no traffic is navigational-by-name, brand is your weakest lever regardless of recognition.