Data & Research
We Audited 151 Merchant Catalogs for AI Agent Readiness. Average Score: 4.4 out of 10.
These are merchants doing $20M to $150M in annual revenue — automotive, supplements, marine. Not small shops. Most are invisible to the fastest-growing discovery channel in commerce.
Published March 31, 2026 · 12 min read
TL;DR
- 1. 151 mid-market catalogs audited: 39% Magento, 24% Miva, 18% NetSuite
- 2. Average AI visibility score: 4.4 out of 10. More than half scored below 5.
- 3. 47% are actively blocking ChatGPT's crawler — most don't know it
- 4. 59% have no JSON-LD Product schema. 62% have product descriptions under 80 words.
- 5. Merchants with complete schema appear 5.25x more often in AI shopping results
Over the past 90 days, we ran AI readiness audits on 151 mid-market ecommerce catalogs. Merchants doing $20M to $150M in annual revenue, running on Magento, Miva, NetSuite, and BigCommerce. The average AI visibility score across all 151: 4.4 out of 10.
AI shopping agents are already routing purchase decisions. ChatGPT has 700M+ weekly active users. These agents don't browse your website the way a human does. They read structured data, parse schemas, and pull from product feeds. If your catalog isn't formatted for that, you're invisible.
Why We Ran 151 Audits
We started running AI visibility audits in late 2025 as part of our sales process. The original intent was narrow: show a merchant how their catalog appears to AI shopping agents, then offer to fix it.
After the first 40 audits, the pattern was already clear. By 80, we had enough data to segment by industry and platform. At 151, we decided to publish.
Dataset breakdown:
- Automotive and powersports: 61 merchants
- Supplements and health: 48 merchants
- Marine and boating: 42 merchants
Platform split: 39% Magento/Adobe Commerce, 24% Miva, 18% NetSuite/SuiteCommerce, 12% BigCommerce, 7% other.
Every merchant in the dataset does at least $15M in annual online revenue. Most are between $25M and $80M. They're not failing at ecommerce. They're failing at AI commerce specifically.
The Scoring Methodology
Each audit scores a merchant catalog on seven dimensions, weighted by actual impact on AI agent discoverability. Maximum score: 10.
JSON-LD Product schema with name, description, price, availability, brand, SKU, GTIN. Partial schema scores 0.5. Complete with offer markup scores 2.
Under 80 words scores 0. 80–200 words with specs scores 0.75. Over 200 words with use cases, compatibility, and specs scores 1.5.
GTINs, UPCs, MPNs. No identifiers: agents skip. Full GTIN coverage scores 1.5. Under 50% SKUs scores 0.
GPTBot, ClaudeBot, PerplexityBot all allowed in robots.txt, returning 200 status codes — not blocked by WAFs or CAPTCHAs.
The emerging standard that tells AI agents what your site sells and how your catalog is organized. Think robots.txt for AI commerce.
Sub-second product API with real-time inventory scores 1.5. Daily XML feed scores 0.75. No API scores 0.
Real-time inventory in structured data scores 1. Daily updates score 0.5. No inventory data in markup scores 0.
Score Distribution: What 4.4/10 Looks Like
The distribution is not a bell curve. It's bimodal — a large cluster between 3 and 5, a small cluster between 7 and 9.
| Score range | Merchants | Share |
|---|---|---|
| 0–2 (Critical) | 18 | 12% |
| 3–4 (Poor) | 67 | 44% |
| 5–6 (Partial) | 38 | 25% |
| 7–8 (Good) | 22 | 15% |
| 9–10 (Excellent) | 6 | 4% |
More than half scored below 5. Only 28 scored above 6.
What a 4/10 store looks like:
A $35M powersports retailer on Miva. Product pages have titles and prices in HTML. No JSON-LD schema. Descriptions average 45 words pulled from manufacturer spec sheets. No GTINs. GPTBot is blocked in robots.txt — a security plugin added it six months ago and nobody noticed. No llms.txt. Weekly CSV feed, no inventory counts. When we queried ChatGPT and Perplexity for products this retailer carries: zero results. Not buried. Absent.
What an 8/10 store looks like:
A $50M supplements brand on Magento 2. Complete JSON-LD Product schema on every page. Descriptions average 280 words with ingredient lists, dosage, and comparisons. All three major AI crawlers allowed. Well-structured llms.txt. Product API responds in 340ms with real-time inventory. This merchant appeared in 6 out of 10 relevant AI shopping queries we tested. No exotic setup. Just doing the basics completely.
The 5 Most Common Failures
These five issues accounted for 83% of all points lost across 151 audits.
GPTBot blocked in robots.txt
47% of merchantsAlmost half the merchants we audited are actively blocking ChatGPT's crawler. In most cases, this wasn't a deliberate decision. Security plugins like Wordfence and Sucuri have been adding GPTBot to block lists by default since mid-2025.
The fix is a single line in robots.txt. But if nobody checks robots.txt after plugin updates, the block persists. We found merchants who had been blocking GPTBot for over eight months without knowing it. When blocked, agents skip you entirely — there is no fallback.
No JSON-LD Product schema
59% of merchantsJSON-LD Product schema is how you tell AI agents the structured facts about your products. Without it, an AI agent has to parse raw HTML and guess. Agents don't guess. They look for schema. If it's not there, they move to a merchant that has it. Among the 89 merchants with no schema, 23 had partial implementations — schema on some pages but never configured for the full catalog.
Product descriptions under 80 words
62% of merchantsThe most common failure by count. AI agents build their understanding of a product from the description. "High-performance brake pad. Fits 2018–2024 models. OEM quality." is 11 words. The agent doesn't know which vehicle makes, what the material is, or how it compares to alternatives. In our testing, descriptions under 80 words result in agents either skipping the product or generating inaccurate summaries that lead to returns.
Missing GTINs and MPNs
54% of merchantsUniversal product identifiers connect your listing to a known product in the agent's knowledge graph. Without them, your "Premium Fish Oil 1000mg" is one of 4,000 indistinguishable fish oil listings. In automotive and marine, MPNs are critical for fitment. Without them, an agent can't verify a part fits — and won't recommend it.
No llms.txt file
91% of merchantsOnly 13 of 151 merchants had any form of llms.txt. Of those, only 6 were well-structured. An llms.txt file serves as the front door for AI agents: here's what we sell, here's how the catalog is organized, here's how to access product data. Without it, an agent has to discover everything by crawling — slow, incomplete, and often blocked by the same security tools that caused failure #1.
Industry Breakdown
| Industry | Count | Avg score | Highest |
|---|---|---|---|
| Automotive/Powersports | 61 | 3.8 | 8.5 |
| Supplements/Health | 48 | 5.1 | 9.0 |
| Marine/Boating | 42 | 4.2 | 7.5 |
Supplements score highest because Amazon and Google Shopping compliance forced structured data investment years ago. That foundation transfers directly to AI readiness.
Automotive scores lowest because fitment complexity makes structured data harder. A single brake pad SKU might cover 200 year/make/model combinations. Most merchants store this in proprietary databases that don't map to JSON-LD. Product descriptions average 38 words — the shortest in the dataset.
The 5.25x Visibility Multiplier
Among the 62 merchants with complete JSON-LD Product schema, the median AI visibility rate was 4.2 appearances per 10 relevant queries. Among the 89 with no schema: 0.8 appearances. That's a 5.25x difference.
Schema markup is the single highest-impact fix. Most ecommerce platforms have schema plugins or built-in support. The problem is that 59% of mid-market merchants haven't configured it correctly. And 81% have no way to track whether AI agents are surfacing their products at all.
Check Your Own Score in 15 Minutes
Step 1: Crawler access (2 min)
Go to yourdomain.com/robots.txt. Search for GPTBot, ClaudeBot, PerplexityBot. If any are followed by Disallow: /, that agent is blocked.
Step 2: Schema markup (5 min)
View source on any product page. Search for "@type":"Product". Check whether gtin, availability, and aggregateRating are populated. Or paste your URL into Google's Rich Results Test.
Step 3: AI visibility test (8 min)
Open ChatGPT or Perplexity. Ask 5 purchase-intent questions for products you sell. Count appearances. Zero means you have a visibility problem. Three or more puts you in the top quartile of what we've seen.
Get your full score
FlowBlinq runs free AI visibility audits that score your catalog across all seven dimensions. Takes 48 hours. No commitment, no sales pitch — just a score and a specific list of what to fix.
Get your free audit →