Machine-First Architecture for Makers
Part 2 of the Agentic Commerce Series
In Part 1, I laid out what agentic commerce means for independent jewelry designers: the protocols (UCP, ACP), the platforms (ChatGPT, Google AI Mode, Perplexity, Gemini), and why this shift is happening right now.
This post goes one level deeper: why your site isn't ready for it yet, and what "ready" actually looks like.
The framework comes from Slobodan Manić, host of the No Hacks podcast, who recently laid out what he calls "machine-first architecture" in a piece for Search Engine Journal. His argument is the clearest I've seen for why structural changes, not just content changes, are what matters now.
But Manić speaks to a technical SEO audience. This post translates his framework for makers: independent jewelry designers, small-batch producers, studio-based brands. The people who already have incredible products and brand stories, but whose websites were built to persuade humans, not serve machines.
Key Takeaway
AI doesn't browse your photos or feel your brand energy. It extracts structured data to decide whether to cite you, recommend you, or skip you entirely. The fix isn't a redesign. It's adding a "warehouse" layer underneath the storefront you already have. I did this on 10 of my own product pages. Within three months, my DA 7 site was being cited above Deloitte and Mastercard across four AI platforms.
The Storefront vs. Warehouse Frame
Manić put it simply: for two decades, your website was a storefront. People came to you, browsed your pages, and made decisions there.
That era isn't ending. But it's no longer the whole equation. Your website needs to be both a storefront and a warehouse. A place where AI agents can pull structured, accurate product data without ever walking through the front door.
If you've ever done a craft show, you already understand this distinction. Your booth is the storefront: the display, the lighting, the way you've arranged your pieces to catch someone's eye. But behind the scenes, you also need an inventory system: what's in stock, what metal each piece uses, what the price is, what size it runs. The booth sells. The inventory system fulfills.
Right now, most maker websites are all booth and no inventory system, at least from an AI's perspective. Your photos are gorgeous. Your brand voice is compelling. Your about page tells a story that makes people feel something.
But when an AI agent asks "handmade aquamarine necklace, gold, under $500, independent designer," it doesn't feel things. It looks for structured fields: product type, material, price, availability, creator credentials. If those fields aren't machine-readable, your store is invisible to the query, no matter how beautiful it is.
“Merchants with the most structured, high quality data foundations will be positioned to win.”
Ginny Marvin, Google Ads Liaison
Four Pillars, Translated for Makers
Manić's machine-first architecture has four pillars. Here's what each one means if you make jewelry (or any handmade product) and sell it from your own website.
1. Identity: Does AI Know Who You Are?
This isn't about your about page. It's about whether an AI system can extract, without guessing, who makes the products, where they're made, what materials they use, and what makes them different from every other jewelry brand on the internet.
Before I optimized my own product pages on andreali.com, only 3 out of 10 products mentioned my name. Zero mentioned Denver. Zero had a scarcity signal explaining that each piece is one-of-a-kind.
The brand story existed on my about page and my homepage. But product pages, the pages that actually get evaluated when an AI agent is answering a shopping query, had almost none of it.
The maker translation: Your identity needs to live on every product page, not just your about page. Name, location, process, materials, uniqueness, all in the description text, in the schema markup, in the FAQ content. Every page needs to independently answer the question: who made this and why should I trust them?
2. Structure: Can AI Read You with JavaScript Off?
Manić's test is brutal and simple: disable JavaScript in your browser and visit your own site. If content disappears, AI crawlers can't see it either.
Most AI systems don't render JavaScript the way a browser does. They read raw HTML. If your product descriptions, pricing, or availability are loaded dynamically, those details may be invisible to the very systems you're trying to get cited by.
The maker translation: This is the Squarespace gotcha. Squarespace 7.0 does render most content server-side, but certain elements (interactive lookbooks, dynamically loaded product variants, AJAX-loaded reviews) can disappear when JavaScript is off. Code Blocks with custom HTML render fine. Dynamic product grids sometimes don't. Test it.
3. Content: Does Your First Paragraph State What the Page IS?
This is the one that stings for makers. Because we love story. We love starting a product description with the emotion, the inspiration, the narrative.
But Manić's point is precise: the first paragraph of every page needs to state what the page is about in plain, entity-dense language. Not because story doesn't matter, but because AI reads top-down and decides relevance fast. If your opening is "The mystical power of Orenda flows through this piece," an AI system has no idea whether it's looking at a necklace, a bracelet, a candle, or a yoga mat.
When I rewrote my product pages, I shifted from openings like that to: "One-of-a-kind tourmaline cuff bracelet, hand-fabricated in 14k gold by Andrea Li in her Denver studio." The story still lives on the page. It comes right after. But the first sentence tells a machine exactly what it's looking at.
The maker translation: Lead with what the piece IS. Material, gemstone, jewelry type, metal, designer, location. Then tell the story. You're not killing the narrative. You're giving it a machine-readable headline.
4. Interaction: Can an Agent Complete a Flow on Your Site?
This is where Part 1 connects directly. The protocols I covered there (Google's Universal Commerce Protocol, OpenAI's Agentic Commerce Protocol) are the emerging infrastructure for this pillar.
For most independent jewelry designers, this pillar is the furthest from implementation. You're not going to build a UCP endpoint this year. But you can make your shipping, returns, and availability data machine-readable right now. That's the foundation the protocols will build on.
The maker translation: Make your policies extractable. Clear shipping timelines on every product page (not buried in a footer link). Returns policy in plain language. Availability status (in stock vs. made to order vs. sold out) visible in both the page content and the schema. When an agent can read these without clicking around, you're already ahead of most indie brands.
What Happened When I Built the Warehouse Layer
This isn't theory. I ran this on my own store, andreali.com, across 10 product pages in April 2026.
The approach: a three-layer system applied to each product page.
Layer 1: Voice description (150 to 400 words). Entity-dense opening sentence, then the brand story. Written for humans, structured for machines.
Layer 2: FAQ accordion (3 to 4 questions per product). Collapsed by default so it doesn't clutter the mobile experience. Answers the questions a buyer would ask AND an AI agent would extract.
Layer 3: Enriched schema (invisible to buyers). Product + FAQPage + BreadcrumbList markup injected into the page code. This is the pure warehouse layer. Machines read it, humans never see it.
| Metric | Before (avg) | After (avg) | Change |
|---|---|---|---|
| Word count | 186 | 406 | +118% |
| FAQ questions | 0 | 3.8 per product | 0 to 38 total |
| Schema items | 6.9 (sitewide defaults) | 9 per product | +3 per product |
| Designer name | 3 of 10 | 10 of 10 | +233% |
| Location (Denver) | 0 of 10 | 10 of 10 | 0 to 100% |
| Scarcity signal (OOAK) | 0 of 10 | 10 of 10 | 0 to 100% |
| Sold-out redirect | 0 of 2 (dead ends) | 2 of 2 | Commission CTAs |
10 products. 6 collections. 5 product types. 2 sold-out sunset test cases. Every product validated at 0 errors, 0 warnings in Google's Rich Results Test. Zero fabricated content.
10
Products
38
FAQs Added
+118%
Word Count
0/0
Errors / Warnings
The Proof: A DA 7 Site Cited Above Deloitte
After implementing these changes across my product pages, pillar pages, and gemstone guides over the past three months, I tested what AI platforms actually say when someone asks about agentic commerce for the jewelry industry.
The query: "I want to find out more about agentic commerce for the jewelry industry."
| Platform | What Happened |
|---|---|
| RedPinGeek.com ranks #1 and #3. Above agenticcommerceindex.com. Above Mastercard. | |
| ChatGPT | Cites redpingeek.com in its research panel alongside blog.google, stripe.com, and alhena.ai. Uses my framing throughout its response. |
| Perplexity | Cites redpingeek.com repeatedly across a 14-source answer. My "back door vs. front door" framework and three-layer model structure the entire response. |
| Gemini | Lists RedPinGeek as source #1 in the sources panel. |
My site has a domain authority of 7. Deloitte is 91. Mastercard is 91. This is not an authority story. It's a legibility story.
The ChatGPT 6.5/10 That Started Everything
Before I built any of this, I asked ChatGPT to evaluate my jewelry store, andreali.com, against a specific shopping query: "handmade aquamarine necklace, independent designer, one-of-a-kind, gold, under $500."
It gave me a 6.5 out of 10.
Not because my products were bad. Not because my brand wasn't strong. ChatGPT explicitly said my brand positioning was already good and my trust signals were solid.
The problem was matchability. The aquamarine necklaces on my site were mostly over $500 and mostly sterling silver, not gold. The price, material, and availability signals were scattered across different pages instead of concentrated on one clearly matchable product page. An AI agent couldn't confidently recommend me for that specific query without guessing, so it wouldn't.
That 6.5 wasn't a quality score. It was a warehouse score. The storefront was beautiful. The warehouse was disorganized.
Persuasion Layer vs. Extractability Layer
Most makers only have the first column. You need both.
| Your Storefront (persuasion) | Your Warehouse (extractability) |
|---|---|
| Beautiful product photography | Descriptive alt text with material + gemstone + type |
| Compelling brand story on about page | Designer name, location, process on EVERY product page |
| Emotional product descriptions | Entity-dense opening sentence stating what the piece IS |
| Trust through visual design | Trust through schema: FAQPage, BreadcrumbList, enriched Product |
| Policies linked in footer | Shipping speed, returns, availability on each product page |
| Reviews displayed visually | AggregateRating schema so AI extracts sentiment as data |
| "One of a kind" vibe | Explicit scarcity: "one piece exists" in description + schema |
Try This Right Now (5 Minutes)
Manić's simplest diagnostic, adapted for makers:
1. Open Chrome. Go to Settings, Privacy and Security, Site Settings, JavaScript, toggle it OFF.
2. Visit your own site. Navigate to your best-selling product page.
3. Read what's left. Can you still see: product title, description, price, material, availability, shipping info?
4. Ask yourself: if this is ALL an AI agent can see, would it confidently recommend this product to someone?
5. Turn JavaScript back on.
If anything critical disappeared, that's your first warehouse gap. If everything stayed but the description is "Made with love, perfect for any occasion," that's your second warehouse gap. The content is there, but it doesn't say anything a machine can extract.
Want this done for your product pages?
The same three-layer system I used on my own store is now available as a service. Voice description, FAQ accordion, enriched schema, validated at 0/0 before delivery.
See Product Page OptimizationFAQ
Do I need to redesign my website for machine-first architecture?
No. Machine-first architecture isn't a redesign. It's an additional layer. Your storefront stays exactly as it is. You're adding structured data, entity-dense descriptions, and machine-readable policies underneath it. Most of this is invisible to your buyers.
Is this only relevant if I sell on Shopify?
No. The principles apply to any platform: Squarespace, Wix, WordPress, BigCommerce, even a custom site. The implementation details differ (Squarespace 7.0 handles schema through Code Injection, Shopify uses Liquid templates or apps), but the architecture is the same.
How long does it take to optimize one product page?
With a workflow in place, about 30 to 45 minutes per product. That includes writing the entity-dense description, creating the FAQ accordion, generating the schema, and validating at Google's Rich Results Test. The first one takes longer because you're building the system. Products 2 through 10 go faster.
I'm a one-person studio. Is this worth my time?
The 10-product batch I ran on my own store took about two working days. Within three months, my DA 7 site was being cited by AI platforms above companies with domain authorities 10 to 13 times higher than mine. The time investment was real, but so were the results.
Sources
Manić, Slobodan. "Machine-First Architecture: AI Agents Are Here and Your Website Isn't Ready." Search Engine Journal, April 2026. searchenginejournal.com
Manić, Slobodan. "What Google's UCP Tells Us About Agent-Ready Websites." No Hacks, April 2026. nohacks.co
Marvin, Ginny. Google Ads Liaison, quoted in Google Marketing Live coverage, 2026.
Related
Part 1: The 4-Layer Product Page: What 815K Queries Reveal for AI
Case Study: Andrea Li Designs AI Visibility Audit
Service: Product Page Optimization
Premium Tool: Agentic Commerce Triage Framework
Source Engineering, Not Prompt Engineering: What Three Studies Reveal About AI Visibility in 2026
About the Author
Andrea Li is the founder of Red Pin Geek, an SEO and AI visibility consultancy for independent jewelry designers. She's also the designer behind Andrea Li Designs, a Denver-based handcrafted gemstone jewelry studio, which means everything she teaches, she tests on her own business first. Andrea has 18+ years of jewelry design experience and was previously invited twice to Pinterest HQ for platform strategy work.
Get the next post in this series
I share what's actually changing in AI visibility for jewelry designers every week. No jargon, no doom.
Subscribe to Red Pin GeekWant Red Pin Geek in Your AI Answers?
Google now lets you choose preferred sources that get labeled in AI Overviews and AI Mode. If this content helps you navigate AI visibility for your jewelry business, add us and you'll spot our links faster next time Google answers your questions.
Add Red Pin Geek as a Preferred SourceTakes one click. Works in Google Search, AI Overviews, and AI Mode.
