SUBSCRIBE
Notes from the agentic commerce frontline — in your inbox.
Weekly cornerstone writing on Rufus, ChatGPT, Gemini, Perplexity, Copilot, Claude — plus DACT measurement, protocol updates, and case studies you can actually use. Numbers-first. No fluff.
Subscribe
Get the issue Monday morning.
No spam. No sponsored injections. Unsubscribe in one click.
What you get
Weekly cornerstone
One deep-methodology piece each week — 2,000-3,000 words with a named case study, real numbers, and a 90-day plan. No hot takes.
Protocol updates
Every ACP, UCP, AP2, MCP, A2A, TAP change that matters for merchants — with implementation notes, not just announcements.
DACT + Citation Rank benchmarks
Quarterly reads on where the multipliers are moving, which surfaces are gaining share, and where the arbitrage windows sit.
Recent
Sample the corpus
Aug 7, 2026
Query Fan-Out: One Question Becomes 5-7 Searches (And ChatGPT Rewrites All of Them)
A user typed 'ALTRR Portable Spice Mill (200W)' into ChatGPT. The shopping backend searched 'portable spice grinder travel' — brand stripped, spec dropped, intent added. That rewrite is stamped into every card's payload as generated_product_query, and it is only one branch of a fan-out that turns a single buyer question into 5-7 sub-queries across 8 distinct axes. Classical SEO optimizes one keyword per page. AI shopping ranks you across the whole fan-out — which is exactly why Amazon holds 63-79% presence on every sub-intent type we measure.
Aug 7, 2026
The 86% Rule — In AI Shopping, Eligibility Beats Ranking
When a retailer product page gets cited in a ChatGPT shopping answer, it wins the top recommendation slot 86.03% of the time — the highest hit rate of any content type in an 18,942-citation dataset. But retailer pages are only 8.28% of what ChatGPT cites. That inversion rewrites the whole playbook: the scarce, high-conversion battle in AI shopping is getting cited at all, not ranking once cited. Here are the three gates that decide citation eligibility — reviews, product-type fit, sub-intent presence — and the two beloved PDP levers that tested statistically dead.
Aug 7, 2026
AI Visibility Has Two Layers: The Placement vs Citation Dashboard That Actually Works
AI shopping answers have two independent layers — the products the model recommends (placement) and the sources it attributes the answer to (citation) — and every brand's AI-visibility dashboard is quietly blending them into one number that lies. On Amazon's India mixer-grinder data, one layer is 61.48% and the other is 0.78% — a 79× gap. On sunscreens the same structural pattern holds at 25×. And both layers moved in opposite directions twice in eight months. Here is the four-KPI-group model that fixes it.
Aug 7, 2026
The Amazon Shelf Story: What Really Happened to Amazon in ChatGPT Shopping Between May and July 2026
Everyone screenshotted the May chart: Amazon's ChatGPT presence rose to 96% while its citation rate collapsed to 1.91%. Two months later, on the exact same 425 buyer prompts, both metrics reversed — and the popular narrative got the story completely wrong. The 96% was on a half-empty shelf (56% zero-card responses). The July shelf is about three times healthier. And Amazon still wins the top-2 slot plus cheapest price on 210 of those 425 prompts. Here is the reverse-engineered version of what actually moved.
Aug 7, 2026
Inside chatgpt.com/shopping: A 200-Query Teardown of OpenAI's Real Commerce Stack
We ran 200 buyer queries against ChatGPT's dedicated shopping surface, captured every response payload, and reverse-engineered the plumbing. A hidden GPT-5.3 handles 56% of shopping queries. Fifteen direct feed partners power the carousel — Amazon isn't one, and still owns the single badge OpenAI ships. Ads run through Criteo. Every image is face-detected. Here's what a teardown of the payload actually reveals.
Jul 26, 2026
AI Commerce Conversion Rate: The 15.9% vs 1.76% Stat, Explained
ChatGPT-referred shoppers convert at 15.9% versus 1.76% for Google (Adobe, 2025). Here's what's actually driving that gap — and why it doesn't mean what a lot of decks imply.
Also available