SEO/GEO Training Manual

What You're Walking Into

Kaizen has built — over the last 60 days — a complete SEO/GEO/AEO operating system. The pieces:

• SEO/GEO Template v1.1 — our productized client deliverable framework (the what we ship)
• Dr. Strange's 4 skills — seo-audit, geo-analysis, seo-content-writer, schema-markup (the how we execute)
• GEO Score — proprietary 0-100 measurement system that tracks AI citation rates (the proof to clients)
• 900+ SME library files — pre-built industry knowledge that seeds Citation Magnet content (the competitive moat)
• ahrefs-intel skill — colony-shared Ahrefs data layer (full keyword, backlink, traffic, SERP, and broken-link intel via API). You never have to open ahrefs.com to do your job. Dr. Strange calls this skill for SEO work; Black Widow and Jarvis use it too. The measurement layer is autonomous.
• Discord-based workflow — every Dr. Strange job for an active client posts to that client's dedicated #client-[slug] channel, where Jen reviews and tags Don for sign-off (the quality gate). #seo-geo is reserved for miscellaneous inquiries and pre-conversion prospect work only — no formal client deliverables land there

Why GEO Matters Right Now

The market has fundamentally shifted. From a16z's 2025 analysis:

• ChatGPT queries are 23 words on average vs. Google's 4-word average
• Session length in AI search: ~6 minutes vs. Google's 60-90 seconds
• ChatGPT is already driving referral traffic to tens of thousands of distinct domains
• Apple is building Perplexity/Claude into Safari — Google's distribution chokehold is cracking

From Andrew Warner's interview with Zapier (March 2026):

• Zapier is mentioned millions of times per month by LLMs in product recommendations
• The play is no longer "rank on Google" — it's "be the brand the model recommends"

From @denohawari (March 2026):

• His team has driven $30.52M in client revenue using LLM SEO over the past year
• A B2B SaaS scaled from $20k to $100k MRR in 4 months — 760%+ non-branded traffic growth
• Method: "decision pages" — [competitor] vs [your brand], alternatives to [X], best [tool] for [specific use case]

Translation: Traditional SEO still drives ~60% of website discovery, but it's the floor, not the ceiling. The growth vector is GEO/AEO — being cited when someone asks ChatGPT, Perplexity, Claude, or Google AI Overviews for a recommendation in your client's industry.

The ⅓ / ⅔ Heuristic (Calibratable, Not Fixed)

Our current allocation for client work: ⅓ of effort on SEO foundation, ⅔ on GEO/AEO optimization.

This is a starting baseline, not a law. After 3-5 client builds produce real data, we calibrate. Some clients (local service businesses with strong existing SEO) may need 20/80. Brand-new businesses with no web presence may need 50/50 until their SEO floor exists.

The Three Disciplines Defined

His Five Skills

Dr. Strange — run seo-audit:
  client_name: "Rideout Law"
  target_url: "https://rideoutlawgroup.com"
  audit_scope: "full"
  keyword_targets:
    - "California foreclosure attorney"
    - "wrongful foreclosure lawyer"
    - "loan modification attorney California"
  competitor_urls:
    - "https://competitor1.com"
    - "https://competitor2.com"

Dr. Strange — run geo-analysis:
  client_name: "Rideout Law"
  client_url: "https://rideoutlawgroup.com"
  ai_models: ["openai", "grok"]
  test_queries:
    - "best foreclosure attorney in California"
    - "what should I do if I receive a notice of default in California"
    - "Rideout Law Group reviews"
    - "California foreclosure defense lawyer near me"
    - "can I sue my lender for wrongful foreclosure in California"
    - "Rideout Law vs [competitor]"
    - "I just got a foreclosure notice in Orange County, who do I call"
    - "best lawyer for loan modification in California"
    - "California homeowner facing foreclosure what are my options"
    - "specialized foreclosure attorneys Sacramento California"

Dr. Strange — run seo-content-writer:
  client_name: "Rideout Law"
  content_type: "practice_area"
  target_keywords: ["wrongful foreclosure", "foreclosure defense attorney California", "predatory lending lawsuit"]
  word_count_target: 1500
  client_voice:
    description: "Professional, authoritative, warm but not casual. Trust-building. Avoid legal jargon — clients are scared homeowners, not lawyers."
    existing_copy_urls:
      - "https://rideoutlawgroup.com/about/"
      - "https://rideoutlawgroup.com/foreclosure-defense/"
      - "https://rideoutlawgroup.com/attorneys/"
    existing_blog_urls:
      - "https://rideoutlawgroup.com/blog/notice-of-default-california/"
      - "https://rideoutlawgroup.com/blog/loan-modification-guide/"
  sme_context:
    - "/data/workspace/sme-cowork-library/coworkhive-sme-library-900/foreclosure-defense-sme.md"
  geo_optimize: true

Dr. Strange — run schema-markup:
  action: "audit_existing"
  client_name: "Rideout Law"
  page_url: "https://rideoutlawgroup.com"

Dr. Strange — run schema-markup:
  action: "generate"
  schema_type: "LegalService"
  client_name: "Rideout Law"
  entity_data:
    name: "Rideout Law Group"
    description: "California foreclosure defense and wrongful foreclosure attorneys"
    url: "https://rideoutlawgroup.com"
    phone: "+19165550101"
    email: "info@rideoutlawgroup.com"
    address:
      street: "..."
      city: "Sacramento"
      state: "CA"
      zip: "95814"
    geo:
      lat: 38.5816
      lng: -121.4944
    services: ["Foreclosure Defense", "Loan Modification", "Wrongful Foreclosure Lawsuits"]
    area_served: ["California", "Sacramento County", "Los Angeles County", "Orange County"]

Dr. Strange — run ahrefs-intel baseline:
  client_name: "Rideout Law"
  target: "rideoutlawgroup.com"
  competitors: ["comp1.com", "comp2.com", "comp3.com"]
  country: "us"
  date_from: "2025-11-17"
  date_to: "2026-05-17"

Dr. Strange — run ahrefs-intel:
  action: "rank_tracker"
  client_name: "Rideout Law"
  target: "rideoutlawgroup.com"
  keywords: [<the 50-100 keyword priority list>]
  country: "us"
  date: "2026-05-17"

Phase 1: Discovery + Baseline (Day 1-3)

Goal: Establish where the client currently stands. Build the proof points you'll use to demonstrate ROI later.

Step 1: Client intake via per-client Google Form → Google Sheet (Sebastian builds)

For each new Kaizen client, Sebastian generates a dedicated Google Form pre-filled with the right business + service + market questions for that vertical (legal, F&B, real estate, SaaS, healthcare, etc.). The form lives in the client's onboarding email and the link gets dropped in their dedicated Discord channel. Answers flow into a per-client Google Sheet that Dr. Strange reads at audit time.

You don't have to chase the client for context fields by hand — Sebastian provisions the form when the client channel is created, the client fills it in async, and Dr. Strange consumes it via Sheets API on first audit.

The canonical Form specification lives at kaizen/intake/client-intake-form-template.md. That doc has the full 30-question schema, vertical add-ons (legal, F&B, real-estate, SaaS, professional services), and the provisioning workflow.

Status as of 2026-05-17: The Form spec is finalized and ready to provision. The provisioning script needs one more OAuth scope grant from Don (forms.body, drive.file, spreadsheets) before it can run autonomously. Until that happens, the manual fallback is documented in the intake template — Don creates the first few Forms by hand from the spec, ~10 minutes each, while we wait for the scope grant. See kaizen/intake/client-intake-form-template.md → "Activation."

The canonical intake fields below describe what the Form collects (summary view — the full spec is in the intake template).

CLIENT CONTEXT SCHEMA — [Client Name]

BUSINESS BASICS:
- Business name
- Primary URL
- Address (if local business)
- Phone
- GBP URL
- Years in business
- Team size

SERVICES + MARKET:
- Primary service
- Secondary services
- Service areas
- Target customer
- Average customer value

SEO GOALS:
- Top 5 keywords client wants to rank for
- Keywords currently ranking for (if known)
- Keywords client should rank for but doesn't

CURRENT STANDINGS:
- Reviews: total, star rating, monthly volume
- GBP monthly views (if known)
- Monthly website traffic (if known)
- Map pack status
- Biggest SEO problem (one sentence)

COMPETITORS (minimum 3):
1. name — URL — GBP if local — why they're beating us
2. name — URL — GBP if local — why they're beating us
3. name — URL — GBP if local — why they're beating us

VOICE + EXISTING CONTENT (for client_voice calibration — see Pillar B):
- Brand voice description (how do you want to sound?)
- 2-5 existing web page URLs that best represent your current voice
- 2-5 existing blog/resource URLs that best represent your current voice (if any)
- Anything about your voice you want to deliberately change

WHAT'S ALREADY BEEN TRIED:
- List prior SEO work, agencies, tools, results

This context gets referenced by Dr. Strange in every subsequent skill invocation. Do not skip this step. This is the #1 reason agencies produce generic-feeling work — they never load the business context before starting. The Google Form workflow exists to make it impossible to skip.

Why Google Forms (not Typeform): we stay on existing tools (Google Workspace). One less SaaS subscription, native integration with Sheets, and Dr. Strange reads Sheets natively.

Step 2: Run seo-audit (full scope)

This gives you the technical + on-page + local picture.

Step 3: Run geo-analysis (10-20 queries)

This gives you the AI visibility baseline. Save the date — every monthly re-run compares against this.

Step 4: Run schema-markup action audit_existing

This shows you what structured data exists, what's broken, and what's missing.

Step 5: Run ahrefs-intel for the full Ahrefs baseline sweep

This is a shared colony skill — Dr. Strange calls it automatically as part of his SEO audit workflow, but you can also invoke it directly when you need ad-hoc intel.

A full new-client baseline sweep includes:

• domain_overview — DR, traffic, refdomains snapshot
• traffic_history — 6-month organic traffic + refdomains trendline
• top_pages — what's currently driving traffic (don't break these)
• keyword_universe — every keyword the client ranks for, with quick-wins (position 4-20, volume ≥ 50, KD ≤ 50) auto-flagged
• content_gap — keywords competitors rank for that the client doesn't, leverage-scored
• backlink_intel — referring domains, anchor distribution, recent backlinks
• broken_backlinks — broken inbound links with outreach priority — the highest-ROI recovery play
• anchor_intel — over-optimization warnings (>25% exact-match = risk)

You invoke it in the client's #client-[slug] channel:

Dr. Strange — run ahrefs-intel baseline:
  client_name: "Rideout Law"
  target: "rideoutlawgroup.com"
  competitors: ["comp1.com", "comp2.com", "comp3.com"]
  country: "us"
  date_from: "2025-11-17"
  date_to: "2026-05-17"

Unit cost: ~500-800 Ahrefs API units per full client baseline (we have 400K/month — capacity for ~500-800 new client baselines monthly, or ~2,500 active retainers).

What you get back:

• domain_overview.json + domain_overview-summary.md
• traffic_history.json + chart-ready time series
• top_pages.json ranked by traffic
• keyword_universe.json with quick-wins flagged in summary
• content_gap.json sorted by leverage score (run, don't walk, on these)
• backlink_intel.json with anchor distribution
• broken_backlinks.json with outreach priority per link
• anchor_intel.json with over-optimization warnings

Brand Radar (Ahrefs's 2026 AI-mention tracker) has no public API as of 2026-05-17 — confirmed. For now, our internal geo-analysis skill is the source of truth for AI citation tracking. Revisit if Ahrefs ships a /brand-radar/* v3 route.

You should not need to open ahrefs.com to do your job. If you ever find yourself wanting to (an ad-hoc visualization, a feature we haven't wrapped, exploratory click-through), tell me — we'll either add it to ahrefs-intel or get you a seat.

Deliverable for Phase 1 (two formats):

1. Internal: client-baseline-report.md posted in the client's dedicated Discord channel for Don's review.
2. Client-facing: an HTML version of the same report, published to a private Cloudflare Pages link (or equivalent) that Jen/Don sends to the client. Dr. Strange generates both formats from the same source data — the MD is the working draft; the HTML is what the client actually clicks. Never send the raw MD to the client.

Both formats summarize:

• Overall SEO Score from seo-audit (0-100)
• Overall GEO Score from geo-analysis (0-100)
• Top 5 critical technical issues
• Top 5 keyword gaps
• Top 5 schema gaps
• Competitor position summary
• Recommended deployment model (see Part 4 below)

Phase 2: The Three Re-Architecture Pillars (Week 1-4)

Once you have the baseline, Don approves a deployment model and budget, and you start the actual work. Three pillars run in parallel:

Dr. Strange — run seo-content-writer:
  client_name: "Rideout Law"
  content_type: "service_page"
  target_keywords: ["foreclosure defense California", "stop foreclosure California"]
  word_count_target: 1500
  client_voice:
    description: "Professional, authoritative, warm. Trust-building tone for scared homeowners."
    existing_copy_urls:
      - "https://rideoutlawgroup.com/about/"
      - "https://rideoutlawgroup.com/foreclosure-defense/"
    existing_blog_urls:
      - "https://rideoutlawgroup.com/blog/notice-of-default-california/"
  sme_context:
    - "/data/workspace/sme-cowork-library/coworkhive-sme-library-900/foreclosure-defense-sme.md"
    - "/data/workspace/clients/rideout-law/discovery-notes.md"
  geo_optimize: true

Why It Matters

This is the productized offering that separates Kaizen from every other SEO agency. We have 900+ SME library files (industry knowledge bases at 13,000+ words each). No competitor has that raw material. The Citation Magnet is how we turn that moat into client value.

The 4-Part Architecture

1. Industry Knowledge Graph

• Top 50-100 questions in the client's industry
• Each entry: declarative 1-2 sentence answer + statistic + dated citation + primary source link
• Uses DefinedTerm + FAQPage + Speakable schema
• Auto-generatable from the SME library files

2. "AI Audit" Public Page

• Format: "Here's what AI currently says about [topic], and here's what the data actually shows"
• Positions the client as the authority correcting AI misinformation
• Generates backlinks from journalists + industry pros
• Highly citable (AI prefers correction content)

3. Structured Data Feed

• /llms.txt — plain-text overview of the site's authoritative topics
• /knowledge-base/index.json — structured JSON feed of all Q&A pairs
• XML sitemap with <lastmod> dates signaling freshness

4. Monthly "State of [Industry]" Report

• Auto-generated from SME library (human review only, not authoring)
• Published as webpage + downloadable PDF
• Recurring citation target (AI re-crawls fresh content frequently)
• Builds email list (gated PDF)
• Generates social shares + backlinks

Validation Gate (Important)

Before we sell Citation Magnet as a standard deliverable, we need ONE proof-of-concept build that generates a Citation Magnet page from an existing SME library file using Dr. Strange's automated pipeline. If this requires significant human authoring rather than automated generation with human review, we don't ship it as a productized offering yet.

Pilot client: BDH Consultants (decided 2026-05-17). Dr. Strange + Jen run the first Citation Magnet POC against the tea/F&B SME library files for BDH. Until that pilot is shipped and validated, do not promise Citation Magnet as a deliverable to any other client.

Active client targets (immediate scope):

• rideoutlaw.com
• BDH Consultants (Citation Magnet pilot)
• Platform
• Fleet Intelligence
• Chaiagra
• additional clients TBD as bandwidth permits

These are the sites we work on first. New prospects continue through the normal sales cycle but don't get prioritized over this active list.

Model A: Greenfield Build (Full SEO/GEO)

When: New business, no existing site, or existing site is beyond salvaging.

• Full Astro 6 hybrid architecture on Cloudflare Pages
• Complete dual-audience page design
• Citation Magnet architecture included
• 30-day launch sequence
• Pricing: routed through Brandwyn

Model B: Migration (Existing → Astro)

When: Client has WordPress/Squarespace/Wix with valuable content but a platform that limits SEO/GEO.

• Phased content migration to Astro
• URL mapping + 301 redirect plan
• Preserve existing SEO equity during transition
• Citation Magnet added in Phase 2
• Pricing: routed through Brandwyn

Model C: SEO/GEO Overlay (Keep Existing Platform)

When: Budget-conscious SMB. Most common entry point. Don's preferred starter.

• Add JSON-LD to existing pages (works on any platform)
• Publish /llms.txt and knowledge-base/index.json
• Build Citation Magnet pages as subdirectory/subdomain
• Add FAQ schema to existing service pages
• Implement Speakable markup
• Monthly GEO audits + optimization
• Pricing: routed through Brandwyn

The SEO/GEO Audit as Sales Tool

Standalone GEO Audit (geo-analysis run) — Brandwyn sets the price.

• Show prospect their current AI visibility score (likely low)
• Show where competitors are being cited instead
• Propose Overlay or Build to close the gap

Cost to Kaizen per audit: ~$0.30 in API calls. High-margin, demo-ready, urgency-creating. This is the lead-gen weapon — Brandwyn and Don decide what to charge for it.

Tactic 1: Reddit (Highest ROI for GEO)

Autonomy: HYBRID. Quill can draft answers and identify threads autonomously; Jen must approve every post before it goes live (Reddit accounts get banned fast for AI-detection; voice calibration is non-negotiable). House account kaizen_strategist (or per-client account where appropriate) — never use Don's personal handle.

• LLMs heavily weight Reddit content because moderators vet answers over time
• Older posts are more valuable — LLMs trust them more
• Use a house account
• Answer questions that customers are likely to type into LLMs
• Ask questions that customers are likely to type into LLMs (then answer your own questions)
• Don't obsess over upvotes — Zapier found little correlation between vote counts and GEO utility
• It's a volume play — spread across hundreds of threads, not viral on one
• Agent-run pieces: thread discovery (Quill scrapes target subreddits + scores threads by LLM-query likelihood), draft answers, schedule posts
• Jen-run pieces: final voice review + post-publish moderation engagement

Tactic 2: YouTube

Autonomy: MOSTLY MANUAL. Video production, talent management, and creator partnerships require human judgment. Dr. Strange can support (script outlines from SME library, target-query mapping) but is not autonomous here.

• Gemini uses YouTube heavily; other LLMs are influenced by it indirectly
• Create your own videos, especially for B2B (less competition in B2B video)
• Work with both big creators (polished, high-reach) AND small creators (outsized LLM influence per view)
• Example from Warner's interview: a 835-view video earned 5.9% of the question citation share in its niche
• Agent-run pieces: script outlines from SME library, target-query research, post-publish citation tracking
• Jen-run pieces: filming/production coordination, creator outreach, contract negotiation

Tactic 3: Correct Outdated Articles

Autonomy: MOSTLY AUTONOMOUS. Black Widow can identify, draft, and send publisher outreach end-to-end. Jen approves the publisher list and reviews any high-stakes outreach (legacy press, sensitive verticals).

• Older articles about the client continue to influence LLM responses long after they're outdated
• Message publishers with outdated info; ask for updates
• Publishers often comply because they want credible content
• Track which 3rd-party articles AI is citing → systematically refresh them
• Agent-run pieces: Black Widow identifies outdated articles via geo-analysis citation tracking, drafts publisher outreach, sends from a shared Kaizen outreach inbox, follows up on a cadence
• Jen-run pieces: approve target list weekly, review any responses that need human judgment (corrections, disputes)

Tactic 4: Tools for Measuring LLM Citations

We stay 100% on internal tools for now (confirmed 2026-05-17). No third-party AI citation trackers needed.

• geo-analysis (Dr. Strange's skill) — our internal source of truth. Multi-model testing across GPT-4o + Grok, citation classification, GEO Score, append-only history.
• Ahrefs Brand Radar — has no public API as of 2026-05-17. Not wired up. Revisit if Ahrefs ships a /brand-radar/* v3 route.
• Profound / Petra Labs / Amplitude — not in scope. Existing tools cover the need.

Success Metrics (Day 30 build target + 90-day compound target)

The 30-day build delivers a measurable Day 30 lift, but compound results land in the 60-90 day window as Google re-indexes and AI crawlers re-cite. Targets below are split accordingly — set client expectations against the right column.

Monthly Report Template

MONTHLY SEO/GEO REPORT — [Client Name]
Month: [Month Year]

YOUR AI VISIBILITY SCORE: [X]/100 (↑/↓ from last month)

WHEN CUSTOMERS ASK AI ABOUT YOUR INDUSTRY:
- [X] of [Y] questions: AI recommends YOUR business ✅
- [X] of [Y] questions: AI mentions you but not first 🔶
- [X] of [Y] questions: AI doesn't mention you yet 🔴

TOP WINS THIS MONTH:
- "[Query]" — ChatGPT now cites [Client] (was absent last month)
- "[Query]" — moved from mention to direct recommendation

PRIORITY TARGETS NEXT MONTH:
- "[Query]" — competitor [X] is being cited; here's our plan to earn that spot

TRADITIONAL SEO:
- Google rankings: [X] keywords in top 10 (was Y last month)
- Website traffic: [X] visits (+Y% vs last month)
- Leads from website: [X] ([Y] mentioned finding you through AI)

THIS MONTH'S WORK:
- [List of pages updated, schemas added, content shipped]

NEXT MONTH'S PRIORITIES:
- [List of upcoming work, tied to specific GEO Score gaps]

The Revenue Attribution Reality Check

Honest caveat to know going in: the chain from "AI cited the client" → "customer walked in the door" is imperfect. Most AI-driven visits don't carry UTM parameters. Your three best proxies:

1. The "How did you find us?" intake form field (include "AI assistant: ChatGPT/Perplexity/Claude/etc.")
2. GEO Score trends (rising = more AI mentions = more downstream brand recall)
3. Anecdotal feedback ("a customer told me ChatGPT recommended you")

Don't oversell attribution precision to clients. The honest pitch: "AI is increasingly how your customers research. Our job is to make sure when they ask, you're the answer. Here's how we measure that."

Technical: The 2MB Rule (Charles Floate, March 2026)

• Googlebot fetches only the first 2MB of HTML. Everything after that = invisible to Google.
• Place these AS HIGH IN THE DOCUMENT AS POSSIBLE: <title>, <meta> tags, canonicals, all structured data (JSON-LD).
• External CSS/JS files get their own 2MB limit per file.
• PDFs get 64MB. Use PDFs for resource downloads that need to be indexed.
• WRS (Web Rendering Service) is stateless — it clears localStorage + session data between requests. If your content depends on session state to render, Google can't see it.

Technical: Templated Website Builders Quietly Block AI Crawlers

This isn't a Shopify-only problem. All the major templated site builders — Shopify, Wix, Squarespace, GoDaddy, Webflow's default templates, and most "AI website builder" platforms — ship with robots.txt configs that block GPTBot, ChatGPT-User, ClaudeBot, anthropic-ai, PerplexityBot, and CCBot by default. The platforms made this change quietly across 2025-2026 in response to publisher pressure on AI training data.

The diagnostic: before any other work on a new client, fetch https://[client-domain]/robots.txt and grep for GPTBot, ClaudeBot, PerplexityBot, anthropic-ai, CCBot, ChatGPT-User. If any are Disallow'd, the client's GEO ceiling is artificially capped — AI engines literally cannot read their site. Fix this first or the rest of the work compounds at zero.

Platform-specific fixes:

• Shopify: Online Store → Themes → Edit code → Add robots.txt.liquid to templates folder → Remove AI bot disallow blocks OR explicitly Allow: / for AI user agents → Keep /admin, /cart, /checkout, /account disallowed
• Wix: SEO Tools → robots.txt Editor → Override default with explicit Allow rules for AI bots. Wix only exposes this on Premium plans — if the client is on a free plan, upgrade is part of the engagement.
• Squarespace: Settings → Advanced → URL Mappings won't do it; Squarespace doesn't expose robots.txt directly. Workaround: add <meta name="robots" content="all"> and explicit AI-bot meta directives (<meta name="GPTBot" content="index, follow"> etc.) to every page via Code Injection in Site-Wide Header.
• GoDaddy Website Builder: Settings → SEO → Search Engine Visibility. The setting is binary (allow/block all crawlers) — confirm "allow." For more granular control, GoDaddy users typically need to migrate off the builder.
• Webflow: Project Settings → SEO → robots.txt → manually override the default block list. Webflow gives full control here, so it's a 2-minute fix.
• WordPress (any host): Edit robots.txt directly or via Yoast / Rank Math plugin. Most managed WordPress hosts (WP Engine, Kinsta, Flywheel) ship clean defaults but verify per client.

The escalation rule: if the client is on a platform where unblocking requires migration (most aggressive: GoDaddy Website Builder, some Wix free plans), that becomes a deployment-model decision (Model B migration to Astro). Don't try to optimize AI visibility on a platform that's structurally blocking it.

Google's Own Guidance on AI Optimization (Source: Google Search Central, updated 2026-05-15)

Google published an official "Optimizing for Generative AI Features on Google Search" guide (developers.google.com/search/docs/fundamentals/ai-optimization-guide). Read it directly when in doubt; it is the canonical Google position. Highlights:

Agentic Experiences + Universal Commerce Protocol (UCP) — Forward-Looking

New emerging frontier flagged by Google in May 2026: AI agents that act on users' behalf — booking reservations, comparing product specs, completing transactions. These browser agents access your site by:

• Analyzing visual renderings (screenshots)
• Inspecting the DOM structure
• Interpreting the accessibility tree

Implications for Kaizen client work:

• Accessibility tree matters more than ever — agents parse it. Don't skip ARIA labels, semantic HTML, alt text.
• DOM structure quality affects agent comprehension. Bloated component trees and aria-hidden chaos hurts.
• Screenshots — visual hierarchy, contrast, color-coding — agents see what users see. Don't rely on hover states or invisible text.

Universal Commerce Protocol (UCP) at ucp.dev is the emerging open standard for AI agents transacting on websites. Watch this; it'll be a Pillar A schema-equivalent for transactional sites within 12 months.

For now: Google's pointer is to "agent-friendly website best practices" at web.dev. Add this to the monthly retainer scope for any client with transactional pages (ecommerce, booking, lead-gen) starting in 2026 Q3.

Brandwyn sales angle ([For Brandwyn]): When a prospect asks "what about agents?" or "where is this heading?", UCP is the answer. Position Kaizen as the firm that's already preparing client sites for the agent era — schema, accessibility, DOM hygiene, all set up to make agents able to transact on the client's behalf, not just read. This is a 12-month-out moat that Google itself is telling us to build.

Voice: The Most Common Mistake

When writing client copy, the failure mode is writing in Don's voice or Sebastian's voice instead of the client's voice. Always pressure-test:

• Does this sound like the client? Or like a consultant?
• Is the formality calibrated? (Law firm ≠ SaaS startup ≠ medical practice ≠ tea shop)
• Are industry-specific terms used correctly?
• Would the client's existing site feel consistent with this?

If client_voice is vague when you brief Dr. Strange, stop and clarify with Don before generating. Generic voice is worse than no voice.

The Kalicube "First-Failing Gate" Rule (Jason Barnard, April 2026)

The AI engine pipeline has 10 gates between "discovered" and "won." Confidence passes multiplicatively — 90% at each of 10 gates = 35% at the end. A single weak gate destroys everything downstream.

The rule: Locate the earliest gate where confidence drops below threshold and fix THAT one first. Investing in citation quality when pages don't render properly = waste. Investing in third-party corroboration when entity isn't disambiguated = waste.

Common first-failing gates:

1. Rendering — pages don't load / JS fails / 2MB cutoff hits
2. Indexability — robots.txt or meta noindex blocks
3. Annotation — schema is missing or broken
4. Entity disambiguation — Google/LLMs can't tell which "Smith Law" we're talking about (fix with sameAs links to Wikipedia, Wikidata, LinkedIn, GBP)
5. Extraction — content is written discursively, not in Claim-Frame-Prove blocks
6. Citation — no statistics, no dated sources, no authoritative anchors

Always diagnose before you act. Your first GEO Analysis run tells you the symptom; your seo-audit + schema audit tells you which gate is failing.

The "Hidden Salesforce" Frame (For Client Conversations)

When selling GEO to skeptical clients, the most effective frame (per Kalicube):

"Seven AI platforms — Google, ChatGPT, Perplexity, Claude, Copilot, Siri, Alexa — are working 24/7 either for you or for your competitors. Untrained, they default to whichever competitor has the best-corroborated content. Trained, they become the most scalable sales channel a business has ever had. The question isn't whether AI is recommending businesses in your space. It's whose business it's recommending."

CEOs understand the salesforce argument in ~8 seconds. Use it.

Content Drift vs. Corroboration Decay (Governance)

Most agencies monitor content drift (your own content going stale). They miss corroboration decay — 3rd-party references that supported your credibility get rewritten, drop off, or get superseded by competitor-favorable articles. Your site hasn't changed; the evidence base under you has.

Add to monthly retainer scope:

• Track full brand footprint, not just owned site
• Refresh third-party content on a deliberate cadence (quarterly minimum)
• Set Google Alerts for client name + key terms; flag any new article that contradicts or supersedes our positioning
• Reach out to publishers proactively (correction requests, refresh offers)

Decision Pages > Top-of-Funnel Content

From @denohawari's $30M case study:

"Don't write 'what is HR software?' — that's broad keyword content where giants always win. Write '[competitor] vs [your brand]', 'alternatives to [competitor]', 'best [tool] for [specific use case]'. These pages directly match the questions buyers ask AI. AI rewards specificity, not authority age."

For every client, build at minimum:

• 1 [client] vs [top competitor] page
• 2-3 best [service] for [specific customer profile] pages
• 1 alternatives to [competitor] page (if the competitor is gettable)

The Karpathy "Untrained Salesforce" Insight (Operating Note)

Some of the most important content on a well-built website is rarely visited by humans, because it exists to teach the AI what the brand is. The seven AI platforms read that content to form their understanding of the client even when no human ever clicks the page.

Don't prune low-traffic pages purely on analytics. If the page exists to feed the entity understanding (about page, technical service definitions, methodology pages), it earns its place even with zero human visits. AI eats it.

Where to Post What — The Per-Client Channel Model

Every Kaizen client gets their own dedicated Discord channel. That channel is where all of that client's audits, GEO analyses, schemas, copy, roadmaps, and deliverables live. Don is in every client channel. Jen and Dr. Strange (and whichever colony bots are needed) are in every client channel. The client is not in the channel — it's internal-only.

The old shared #seo-geo, #website-builds, and #projects channels still exist, but they are not where formal client work happens. #seo-geo specifically is for miscellaneous SEO/GEO inquiries and potential client/prospect inquiries that surface before someone is an official Kaizen client — sales-cycle stuff, exploratory chatter, lightweight demos. Don is not going to review Dr. Strange's deliverables there.

The moment a prospect signs and becomes a client, Sebastian provisions a dedicated #client-[slug] channel and every Dr. Strange invocation, audit, copy draft, schema, and roadmap for that client lands there. Don reviews and signs off in the client channel. Nowhere else.

Channel provisioning: when a prospect converts to a client, Sebastian creates the #client-[slug] channel, invites Don + Jen + the colony bots, and drops the client's intake form link as a pinned message. The old prospect-channel threads stay where they are (historical record); new work happens in the dedicated channel.

Tagging Don: any question or request for Don about a specific client goes in that client's channel, period. Don does not want one-off questions sent to #war-room or any shared channel. The only thing that goes in #war-room is a cross-client decision — a strategy call that touches multiple clients at once, a pricing-policy question, a process change. If it's about one client, it goes in that client's channel.

The Quality Gate (Non-Negotiable)

Every Dr. Strange output is reviewed before client delivery. No exceptions.

The review happens in the client's dedicated channel — not in a shared review channel. Workflow:

1. Dr. Strange (or whichever bot owns the deliverable) posts the output to the client's channel
2. Jen reviews first — voice, vertical accuracy, completeness
3. Jen tags Don in-channel for final sign-off (@Don ready for review — [deliverable type])
4. Don signs off (or red-lines) in-channel
5. Sebastian publishes the client-facing version (HTML link, email, etc.) only after Don's sign-off

This includes:

• Audit reports
• GEO Analysis reports
• Generated copy
• Generated schemas
• Proposed roadmaps

Why: this is how we catch voice misfires, factual errors, and over-promising before they hit the client. Dr. Strange is fast but not perfect. The review gate is what makes the system trustworthy. Tagging Don in the client channel (vs. a separate review channel) keeps the full audit trail in one place — Don can scroll the channel and see every step.

Daily / Weekly / Monthly Rhythm

Daily:

• Sweep each active client's dedicated channel for any Dr. Strange runs awaiting your review (this is where formal client work lives)
• Check #seo-geo for any miscellaneous SEO/GEO inquiries or prospect questions worth surfacing to Brandwyn/Don
• Track active client tasks in shared task queue

Weekly:

• Run ahrefs-intel baseline refresh on each active retainer client (Dr. Strange does it; you just review issues)
• Review rank_tracker movement vs. prior run (Dr. Strange auto-diffs and surfaces gainers/losers)
• Brief Dr. Strange on the next batch of pages/schemas/audits in each client channel

Monthly:

• Re-run geo-analysis for every retainer client
• Generate monthly SEO/GEO report per client (template in Part 6)
• Update historical trend data
• Review 3rd-party corroboration decay (Google Alerts queue)

v1.1 — 2026-05-17

• Audience expanded to Jen + Brandwyn. Brandwyn needs the workflow vocabulary to sell the product confidently; the manual now serves both readers with [For Jen] and [For Brandwyn] tagging where relevant.
• "SAGEO" terminology retired. It was shorthand, not a real product name. Replaced everywhere with "SEO/GEO" — the actual product reference. Same framework; cleaner naming.
• Part 2 extracted to a standalone checklist. The standard engagement flow now lives as its own document (kaizen/training-manuals/seo-geo-client-checklist-v1.md) so Jen can manually check off each step per client. The Part 2 section in this manual now points to that checklist.
• ahrefs-intel reframed as Dr. Strange's Skill 5 (lives in the shared colony skills dir so other bots can call it, but Dr. Strange is the primary owner).
• 90-day Pillar C condensed to 30-day standard with parallel agent streams (Dr. Strange / Quill / Black Widow / Pepper running concurrently).
• Schema re-architecture: Dr. Strange does the work, Jen reviews (was previously framed as Jen doing it manually).
• Part 5 Zapier playbook: every tactic now tagged AUTONOMOUS / HYBRID / MOSTLY MANUAL so we know what runs end-to-end vs. what needs Jen's manual judgment.
• Part 7: AI crawler blocks generalized from Shopify-only to all templated builders (Shopify, Wix, Squarespace, GoDaddy, Webflow, WordPress) with platform-specific fix steps.
• Part 9: per-client channel model — every Kaizen client gets their own Discord channel where all formal Dr. Strange work + review + sign-off happens. #seo-geo is for miscellaneous + prospect inquiries only (no formal client deliverables land there). #website-builds / #projects are pre-conversion only. Quality gate = tag Don in the client channel, not in a shared review channel.
• Channel routing simplified: #the-cabinet row removed from the Part 9 table (it's Sebastian-only bot-coordination, not Jen's workflow). Tagging Don for client-specific questions = that client's channel, always. Cross-client decisions = #war-room (1487891856791703724).
• Phase 1 deliverable: internal MD + client-facing HTML link. Sebastian builds a per-client Google Form for intake; answers flow into a Google Sheet that Dr. Strange reads on audit. Stays on Google Workspace (no Typeform).
• client_voice variable expanded to include the client's existing web copy and blog articles as a calibration anchor. Dr. Strange now reads the existing site as input to voice matching, not just the verbal description.
• Citation Magnet pilot client: BDH Consultants (decided 2026-05-17).
• Pricing routed through Brandwyn. No dollar figures in the manual — Brandwyn owns pricing decisions working with Don.
• Active client targets named: rideoutlaw.com, BDH Consultants, Platform, Fleet Intelligence, Chaiagra (plus others as bandwidth allows).
• Part 10 removed. All open questions resolved inline; the section was no longer useful.
• rank_tracker action shipped in ahrefs-intel v1.1; Brand Radar confirmed to have no public API as of 2026-05-17.
• Co-author added: Don Ho (alongside Sebastian).
• Part 7 expanded with Google's official AI-search guidance (May 2026 update): RAG + Query Fan-Out mechanisms named explicitly. Acknowledged that Google specifically pushes back on /llms.txt, chunking, and schema-as-AI-lever — and reconciled by reframing those as multi-platform AI visibility levers (justified by geo-analysis lift across non-Google platforms), not Google-AI-Overview levers. Added new subsection on Agentic Experiences + Universal Commerce Protocol (UCP) as the 12-month frontier.

v1.0 — 2026-05-17 (superseded)

Initial draft. Single-author (Sebastian). Status DRAFT. Replaced same-day by v1.1 after Don's red-line pass.