The Synthetic Shopper: Auditing Your Diamond Feed for ChatGPT and Claude Readability

High-net-worth diamond buyers are no longer scrolling through ten pages of Google search results. They are delegating the discovery phase to a new entity: the Synthetic Shopper.

Whether it is a custom GPT built by a private concierge, or a user prompting Claude 3.5 Sonnet to “analyse the best value 2-carat emerald cut diamonds available in South Africa,” Large Language Models are now acting as direct procurement proxies. If your WooCommerce architecture is unreadable to the synthetic shopper, the human buyer will never even know your inventory exists.

Under the Coetzee Liquidity Protocol (CLP), we do not assume machine readability; we engineer for it and we audit it. You cannot rely on Google Search Console to tell you how OpenAI or Anthropic views your catalogue. Here is the technical protocol for auditing your diamond feed to ensure complete LLM readability.

The Render Blockade: How Modern Frameworks Fail

The first point of failure in a synthetic audit is the render phase. When OAI-SearchBot (OpenAI’s crawler) or ClaudeBot hits your product URL, it does not execute JavaScript with the same robust, patient rendering engine as Googlebot.

If your diamond store uses a headless architecture (React, Next.js, or Vue) with heavy client-side rendering, or if your WooCommerce theme relies on JavaScript to load pricing dynamically via an API after the initial page load, the LLM crawler simply sees a blank <div> or a loading spinner.

The Audit Action: You must test your raw HTML payload. Disable JavaScript in your browser or run a simple cURL command (curl https://yourdomain.com/product-url) in your terminal. If the exact carat weight, certification number, and live ZAR pricing are not visible in that raw text output, the synthetic shopper cannot see them. Your architecture must shift to Server-Side Rendering (SSR) or inject critical pricing and attribute data directly into the static HTML and JSON-LD payload.

The Token Dilution Test

As established in the CLP, LLMs operate within strict context windows. When a synthetic shopper extracts your page, it consumes tokens.

If your product page contains 150KB of inline CSS, bloated SVG icons, and tracking scripts before the actual product data begins, you are diluting the crawler’s token allocation. The LLM may truncate the extraction before it ever reaches the technical specifications of the diamond.

The Audit Action: Extract the text of your product page exactly as an LLM would. You can use a text-only browser like Lynx, or paste your URL into a tool that strips the DOM down to markdown. Count the words that appear before your core diamond attributes. If an LLM has to read through 800 words of generic site navigation, footer links, and promotional banners before hitting the phrase “Colour: D, Clarity: Flawless”, your signal-to-noise ratio is critical. You must elevate the JSON-LD bridge and strict HTML attribute tables higher in the document tree.

The Prompt-Based Interrogation

The most definitive way to audit your feed’s readability is to actively interrogate it using the models themselves. You must simulate the synthetic shopper.

The Audit Action: Open ChatGPT (using GPT-4) or Claude. Provide the model with your live product URL (or paste the raw HTML/JSON-LD payload if the crawler is blocked by a firewall) and execute a strict extraction prompt:

“Act as a diamond procurement system. Extract the following exact variables from this payload: 1. SKU, 2. Live Price, 3. Carat, 4. Cut, 5. Colour, 6. Clarity, 7. Certification Authority, 8. Setting Metal. Output only the variables in a JSON format. Do not infer or guess any missing data.”

Analyse the output.

• Did it hallucinate the price because it read a “related product” widget instead of the main product?
• Did it fail to find the certification authority because it was buried in an image alt-tag rather than explicit text?
• Did it confuse the metal of the ring setting with the metal of the prongs?

If the LLM cannot output a perfect JSON object of your diamond’s attributes, your semantic structure has failed the audit.

Structuring the Dedicated AI Feed

For enterprise-level jewellery operations, relying on on-page crawling is a baseline, not the ceiling. The ultimate application of the Coetzee Liquidity Protocol is bypassing the HTML crawl entirely by offering a dedicated, synthetic-only feed.

High-end concierge services and custom GPTs prefer to ingest data via structured files (XML, CSV, or dedicated API endpoints) rather than scraping DOMs.

The Audit Action: Review your WooCommerce XML feed generation. Ensure you have a feed specifically stripped of HTML descriptions, outputting only strict CLP key-value pairs (e.g., <g:custom_label_0>GIA-Certified</g:custom_label_0>, <g:material>18k Rose Gold</g:material>). This feed should be referenced in your robots.txt or made available to partner AI procurement agents, creating a frictionless, liquid data pipeline directly from your database to the synthetic shopper.

The New Baseline of E-Commerce

Auditing for synthetic shoppers is no longer a futuristic SEO experiment; it is the current baseline for high-ticket e-commerce. If an LLM cannot definitively parse your diamond inventory, it will route its high-net-worth human user to a competitor who has properly implemented their data liquidity.

Stop assuming your site is readable. Run the synthetic audit.