The conversation started happening in late 2024, and it kept repeating. A retailer would show us their Shopify or WooCommerce catalog: clean product titles, well-written descriptions, proper Google Shopping feed setup. Their organic search traffic was solid. Their Google Shopping placements looked good. Then we would ask them: "Have you tried asking ChatGPT or Perplexity to find one of your products?" The result was almost always the same. Nothing. Or worse, a recommendation pointing to a competitor whose product data was actually weaker by any traditional SEO measure.
The Gap We Kept Seeing
At some point you have to stop calling something a coincidence and start treating it as a pattern. We worked with enough catalog teams to see this play out across categories: outdoor gear, home goods, specialty food, consumer electronics accessories. The retailers who had invested in strong Google Shopping feeds and rich product content were still invisible to AI shopping agents. The ranking signals these agents used were simply not the same as what Google's crawlers had been rewarding for the past decade.
The root issue is structural. Google's product ranking rewarded keyword density in titles and descriptions, feed completeness in a small set of standard attributes, and price competitiveness. AI shopping agents like ChatGPT Shopping, Perplexity Shopping, and Google AI Mode work differently. They are trying to answer a question, not match a keyword. To do that, they need to understand what the product actually does, who it is for, what its physical or technical specifications are in comparable units, and whether it is genuinely available at the stated price.
A catalog optimized for keyword search looks like this: "Men's Waterproof Hiking Boots Outdoor Trail Shoe Anti-slip Sole." A catalog optimized for AI agent comprehension looks like this: a product title that names the item and its primary use case, a structured attribute section that lists waterproofing standard (Gore-Tex membrane vs. hydrostatic coating), sole compound (Vibram, rubber hardness), weight per boot in grams, last width category, and a description that places the product in a specific scenario ("designed for day hikers on rocky Pacific Northwest trails who need ankle support without the weight of a full mountaineering boot"). These are fundamentally different documents. Most existing catalog tools were designed to produce the first kind. We built ReFiBuy to produce the second.
Why Existing Tools Did Not Cover This
When we looked at the catalog optimization tools available, they solved for feed management: getting your products into Google Shopping, Amazon, and comparison sites without errors or attribute mismatches. That is genuinely valuable work. But feed management assumes the underlying product content is already good. It is mostly plumbing. What we needed was something that could evaluate whether each product listing contained the information an AI agent would need, and then rewrite the ones that did not.
The second option was to hire a content team to manually rewrite listings. For a catalog of 500 SKUs that is painful but possible. For 5,000 or 50,000 SKUs it is not a viable approach at the pace AI shopping agents are evolving. The requirements change as the agents change.
We are not saying that feed management tools are wrong or that good content writers are not valuable. They solve real problems. What we are saying is that neither tool was built for this specific question: does this listing contain what an AI agent needs to confidently recommend this product to someone asking a natural language query?
What Scoring Against AI Agent Criteria Actually Means
When we started building the scoring system at the core of ReFiBuy, we had to get specific about what we were actually measuring. "AI-ready" is not a useful target without a definition. We worked backward from how AI shopping agents actually behave: what information do they cite when they make a recommendation, and what causes them to skip a product or caveat it heavily?
The dimensions we settled on include: structured attribute completeness (are the key specifications for this product category present in machine-parseable format), description agent-readability (does the description answer evaluative questions rather than just listing features), price clarity (is the current price available in structured format with currency and any applicable restrictions), availability signal (is inventory status parseable without clicking through), and schema markup coverage (does the page-level markup expose these fields to non-Googlebot crawlers).
Each of those dimensions gets scored independently because they have independent remediation paths. A listing with poor attribute completeness needs different work than a listing with good attributes but an unreadable description. A prioritized list of which SKUs to fix first, and why, is more useful than a single aggregate score that tells you everything is wrong but not where to start.
The First Few Months
We opened early access to three retailers in October 2025. All three had what we would consider well-maintained catalogs by traditional standards. All three had observed their AI shopping referral traffic sitting near zero despite solid organic and Google Shopping performance.
The scoring runs confirmed what we had expected: an average of around 70 percent of their listings lacked the structured attribute coverage that AI agents need for confident recommendations. But the distribution was not uniform. In two of the three catalogs, about 20 percent of SKUs, typically the best-selling and most actively managed products, already scored reasonably well. The remaining 80 percent had not been touched in the same systematic way.
That finding shaped how we think about prioritization. You do not need to fix every listing at once. Fix the high-traffic, high-margin products first, get them visible to AI agents, and work down from there. The 60-day results from those pilots gave us enough signal to know we were solving a real problem. We share more detail on what we measured in our pilot results article.
Why Seattle, Why Now
The timing is not accidental. AI shopping is not a speculative future state: Perplexity Shopping launched its product recommendation features in 2024, ChatGPT added shopping integrations in 2024, and Google AI Mode now surfaces product recommendations directly in search results for a substantial share of queries. Retailers who had five years to adapt to Google Shopping have a much shorter window to adapt to AI agent shopping before the gap between visible and invisible catalogs becomes too wide to close incrementally.
We are a small team in Seattle, close to where a lot of the retail and catalog operations work happens in the Pacific Northwest. We are not claiming to be the final word on AI catalog optimization. The agents will keep evolving, and so will we. What we have built is a system that measures the right things, fixes them at scale, and updates as the criteria change.
Is your catalog ready for AI shopping agents?
ReFiBuy audits your product feed and rewrites the listings AI agents cannot evaluate. See how it works.
See plans
