Articles
40 articles
General capabilities
A 1,000-Node Content Tree for $0.35: Diverge-Then-Converge Topic Architecture
Diverge-then-converge topic architecture generated 1,221 candidate topics in one pass and converged them into a 1,324-node content tree for about $0.35 — heavyweight content planning you can now afford to rerun.
8 Facet Dimensions That Make Keyword Clustering Actually Accurate
Word-surface clustering mixes competing intents into one bucket. How an 8-dimension facet taxonomy — attribute, device, scenario, region — slices keywords before merging, with 1,167 attribute words vs 120 scenario words and 122 corrections baked into rules.
Writing Content Briefs That Prevent Cannibalization: The 5-Step Method
Two pages targeting the same keyword steal rankings from each other. How a Sibling Title Check on every brief — coverage from 0.3 down to 0.05, one unique anchor keyword, pillar-gap scans — stops cannibalization before writing.
Clicks Collapsed 99% While Rankings Rose: A GSC Forensics Story
Brand clicks fell 99% (6,323 to 61) while rankings rose. GSC forensics traced the decoupling to Google re-indexing replaced content.
GA4 → Google Ads Conversion Import: A One-Click Bridge
Google Ads shows no conversions while GA4 records real leads? How to audit GA4 key events, exclude ghost events like a page-load contact_us, and import generate_lead as a conversion action.
Killing the Global Token: Per-Org OAuth for Ads, GSC, and GA4
All clients reading Google data through one global token in .env — a cross-tenant leak waiting to happen. We moved to per-org OAuth: per-org refresh tokens in the DB, real-time resolution, org-scoped caches, self-service re-auth.
Server Logs Reveal Who's Really Crawling You: 20+ Bot Signatures
GA4 and Search Console hide bot-level detail. How to parse nginx access logs, recognize 20+ crawler signatures from Googlebot to GPTBot, and see which paths AI crawlers actually read.
GA4 Funnel Audits: What Actually Counts as a Lead
Generic form events inflate your funnel and mislead your ads. How to audit which GA4 events genuinely represent a lead — and why the answer changes your optimization.
When Ad Blockers Silently Killed Your Lead Tracking (and How We Fixed It Server-Side)
A real GA4 tracking failure: US users visited, filled the form, but conversions showed zero. Here is how we diagnosed ad-blocker interference and fixed it with Google's Measurement Protocol sent server-side.
Daily State Snapshots: A Time-Series Model for Account Health
A single spend number can't tell you if an account is improving. How to build a daily state snapshot time-series model with idempotent snapshots, delta computation and 14-day trends.
A Growth Timeline: Every Budget Change Auto-Logged as an Event
Budget, bid, keyword and negative changes were scattered and unverifiable. How we auto-log every optimization as a timestamped event — entity, field, old and new value — and what it took to make the record reliable.
LLM JSON Output Kept Failing? Disable Thinking Mode
LLM JSON parsing kept failing with empty strings and prose-wrapped arrays. Twelve logged failure patterns traced to one root cause: thinking mode. Turn it off, batch, and parse in layers.
2σ Anomaly Detection at 2 AM: Catching Campaign Drops Before You Wake Up
A 2σ scan catches campaign drops like a 67% impression fall. Statistics detect, an LLM explains, evidence persists — snapshot first, dedup, review.
When the AI Got Causation Backwards: A 70%-Confidence False Conclusion
An LLM blamed 43 correct negative keywords for a 67% impression drop with 70% confidence. Five limitations, four lessons on AI causal reasoning.
One Year of Zero Google Ads Conversions: A Consent Mode Defaults Story
A B2B account spent a year with Ads conversions frozen since 2024-08-23. The culprit was the consent defaults, not the update calls.
Safety-Gating Google Ads Writes: AI Recommends, You Decide
AI optimization touches real budgets: one wrong write burns cash. We gated every entity write behind ADS_WRITE_ENABLED (default false): AI recommends, a dedicated admin endpoint executes, humans confirm, and every change lands in the event timeline.
Search Term Mining at Scale: AI Review + One-Click Negative Keyword Push
Search term reports produce 200+ new terms daily. AI classifies each one by intent with rules fixed first — negatives only for mismatched terms — and humans approve one-click pushes to Google Ads, with low-signal terms cut to save tokens.
Wasted Spend Diagnosis: From GAQL Aggregation to P0/P1/P2 Action Lists
GAQL keyword-level aggregation flags zero-conversion, high-spend and high-CPC keywords; an LLM explains the root cause; and the output is a P0/P1/P2 action list where every fix becomes a tracked experiment with a baseline snapshot.
Why Your RSA Can't Be Edited (and What We Did Instead)
RSA assets can't be edited after creation — so push a fully-loaded new ad (15 headlines / 4 descriptions) and auto-pause the old one. AI completion added +12 headlines and +2 descriptions in one pass.
AI Campaign Creation: From a Marketing Goal to a Live Google Ads Structure
How AI campaign creation turns a plain-language goal into a complete Google Ads structure — ad groups, keywords, negatives and fully-loaded RSA copy — with one-click push and human review at every gate.
Lost Impressions: Budget or Ad Rank? Diagnosing the Real Reason Your Ads Don't Show
A real Google Ads case where a campaign was losing 85% of its impressions to ad rank, not budget. How impression-share diagnosis separates the two causes and what to fix for each.
How We Built a 27-Contact PR List with AI-Assisted Contact Collection
PR outreach starts with a list, and manual research is slow and leaky. How we used AI-assisted collection to build a 27-contact PR list — 18 media and 9 PR vendor contacts — and sent in batches with reply tracking.
PR Distribution Is Mostly Spam. Here's the 20% That Still Works (GEO Edition)
An audit of a $419.99 PR package found 150+ content farms among 300+ claimed sites. Nofollow links and devalued networks mean classic SEO value is gone — the real win is GEO visibility.
Building a PR Outreach Command Center with M365, SendGrid, and Resend
PR outreach from a personal mailbox collapses under scale. How we built a command center with tagged contacts, a template library, real connected mailboxes (M365, SendGrid, Resend) and a human-reviewed reply ledger.
Measuring Brand Visibility in AI Answers: A 10-Level GEO Ladder Test
A reusable 10-level GEO ladder test measures how often ChatGPT, Claude, Gemini and Perplexity mention your brand and where. One round of ~120 calls costs about $0.23 — cheap enough to run as a standing monitor.
Monitoring AI Traffic: 15 Preset AI Source Sites in GA4
AI platforms send traffic that GA4 does not group. How a preset list of 15 AI source domains aggregated with inListFilter turns AI sessions into daily trends, source mix, landing pages and regional drill-down.
PR for AI: Why Yahoo Finance Links Still Matter in the ChatGPT Era
nofollow blocks link equity, not AI absorption. Why releases on high-crawl sources like Yahoo Finance shape what AI answers about your brand — and how to buy PR for citation potential, not just authority.
Shopify
Bundle Rate Mining: Why Large Power Stations Bundle 3x Better
Bundle rate per SKU split our catalog: large-capacity models bundle at 27.6% vs 9.8% for small — and BP3000's 24 sales with zero reviews exposed a social-proof leak. Four steps to mine bundles.
SKU-Level Price Benchmarking: The €/kWh Comparison That Found an 18% Edge
List prices lie when capacity differs. How normalizing to €/kWh matched P2001 Pro at €439 vs €530 against its Bluetti equivalent — an 18% unit-economics edge that becomes value-for-money positioning.
Why 92% of Carts Were Lost: A Site-Level Checkout Diagnosis
Abandoned value (€970) beating AOV (€749) exposed a site-level checkout leak. Scoring funnel domains found it before touching any page.
Cross-Source Analysis: Finding the Funnel Step That Kills Your Orders
GA4 shows you where visitors drop; Shopify shows you what they buy. Crossing the two reveals funnel bottlenecks neither platform can show alone — with a real example of a 90%+ loss step.
The 7-Domain Shopify Store Audit: Where Your Money Is Actually Leaking
Abandoned carts, payment failures, retention, catalog issues, trends and marketing — a seven-domain health audit that turns a Shopify store's problems into a prioritized fix list.
Fact-Checking AI Articles: From Whack-a-Mole to a Convergent Pipeline
AI articles kept errors — one had 2 critical issues after 4 patch rounds. Whitelist sources and traceable numbers make fact-checking converge.
Not Every Article Needs a Fact-Check Loop: A YMYL Classifier That Cuts Cost 5x
We ran the full fact-check pipeline on every article at about $0.05 each. A rule-based YMYL classifier cut per-article cost to roughly $0.01 — a 5x saving — without touching the checks that actually matter.
How to Stop AI Content From Making Up Facts
AI writers hallucinate numbers and citations — a fatal flaw for SEO. Research-first writing plus mechanical number verification: how to keep AI content grounded in truth.
From Keyword to Published Article: How an AI SEO Content Pipeline Works
Keyword cleaning, multi-dimensional facets, topic trees, content briefs and fact-checked AI articles — the stages of a content pipeline that publishes dozens of SEO articles a week.
Measuring Translation Completeness by What Users Actually See: A/B/C Buckets
Our translation system said 100% — customers still saw English. Measuring by visible rendered text with A/B/C source buckets found the real hole: one unknown source was 63.9% of a page's 29,718 visible characters.
Translating 126 Blog Articles Into 6 Languages for $12
126 blog articles × 6 languages — the bill most stores fear. We ran it through an LLM translation pipeline in two batches for about $12: chunked, scored, audited, and rollback-ready.
Multilingual SEO That Google Can Actually See
Browser-side translation swaps text for humans but not for Google. Why server-side translations, SEO-safe URLs and quality-controlled copy make translated stores actually rank.
Translating the Text Other Shopify Translation Apps Can't Reach
Most Shopify stores have content locked in theme templates and hardcoded strings that standard translation apps never touch. How to find and translate it safely.