How to Test If AI Shopping Agents Can Actually Find Your Store: A Step-by-Step Diagnostic Guide

92% of brands are invisible in AI search results, according to a May 2026 analysis of over 23,000 LLM citations across ChatGPT, Perplexity, Gemini, and Claude. For ecommerce stores, this means products that rank on page one of Google might as well not exist when a shopper asks ChatGPT to recommend them.

The gap is not always technical. Many stores have decent schema markup and even a Google Merchant Center feed, but fail at one of the other layers AI agents need: crawl access, structured content, or a clear llms.txt instruction file. The problem is that most store owners have no idea which layer is broken because they never test the full chain end to end.

This guide gives you six diagnostic tests you can run in under an hour to find out exactly where your store’s AI discoverability breaks down. Each test targets a specific layer: crawler access, content structure, citation readiness, and agent actionability. Run them in order. Fix what fails. Re-test.

The Six-Layer AI Discoverability Stack

Before running diagnostics, you need to understand what you are testing. AI shopping agents do not work like Googlebot. They combine web crawling, content extraction, structured data parsing, and real-time inference. A store can pass one layer and fail the next.

Each diagnostic below maps to one or more of these layers.

Test 1: Verify AI Crawler Access (Layer 1)

Your store’s robots.txt file is the gatekeeper. If it blocks AI crawlers, nothing else matters. ChatGPT uses GPTBot, Google uses Googlebot and Google-Extended, and Perplexity uses PerplexityBot. Each has a distinct user-agent string.

Step-by-step:

1. Open your browser and go to https://yourstore.com/robots.txt
2. Search for these user-agent strings: GPTBot, Google-Extended, PerplexityBot, ClaudeBot, Bytespider
3. Check if any Disallow rules block access to product pages, collections, or blog content
4. If you see Disallow: / under any of these bots, that crawler cannot index your store

Common mistake: Many Shopify stores use third-party SEO apps that inject broad Disallow rules. These sometimes block AI crawlers unintentionally. Check for patterns like Disallow: /collections/all* or Disallow: /blogs/* that restrict content AI agents need.

For a full audit walkthrough, see our robots.txt AI crawler access guide for ecommerce stores.

What a pass looks like: All major AI crawler user-agents are either not mentioned (which defaults to allowed) or explicitly allowed. No product or content paths are disallowed.

Data point: A May 2026 study by Seer Interactive analyzing 25.1 million Google AI Mode impressions found that 93% of queries ended without a click. When AI agents cannot crawl your store, you are not just losing clicks. You are losing the citation itself.

Test 2: Check Content Parseability (Layer 2)

AI agents do not execute JavaScript the same way browsers do. If your product content loads only after JavaScript execution, crawlers see an empty page. This is especially common with headless commerce setups, React-based themes, and Shopify Hydrogen stores.

Step-by-step:

1. Open a product page in your browser
2. Right-click and select “View Page Source” (not Inspect Element)
3. Search for your product name, price, and description in the raw HTML
4. If you cannot find them, your content is JavaScript-rendered and invisible to most AI crawlers

The text-mode test: Install a text-mode browser extension or use curl from your terminal:

curl -s https://yourstore.com/products/your-product | grep -i "product name"

If the product name does not appear in the raw HTML response, your content is not parseable.

Fix for JS-rendered content: If your store relies on client-side rendering, you need either server-side rendering (SSR), static HTML generation, or a pre-rendering service. Google has gotten better at processing JS, but AI crawlers like GPTBot and PerplexityBot are far less capable at JavaScript execution as of mid-2026.

What a pass looks like: Product name, price, description, and availability are all present in the raw HTML source without JavaScript execution.

Test 3: Validate Product Schema Markup (Layer 3)

Structured data is the vocabulary AI agents use to understand your products. Without valid Product schema, an AI crawler sees text but cannot reliably extract price, availability, SKU, or reviews.

Step-by-step:

1. Open Google’s Rich Results Test: https://search.google.com/test/rich-results
2. Enter your product page URL
3. Check for valid Product schema with these required properties:
   • name
   • image
   • offers.price
   • offers.priceCurrency
   • offers.availability
4. Also test with Schema.org’s validator: https://validator.schema.org/

The AI-specific check: Google’s Rich Results Test validates for Google’s search features, which are not identical to what AI agents need. AI agents also benefit from description, brand, sku, review, and aggregateRating. A page can pass Google’s test and still lack the depth AI models want for generating recommendations.

For a deeper comparison of what schema validators catch and what they miss for AI discoverability, see our schema validators analysis for ecommerce AI discoverability.

Data point: According to the 2026 AI Search Visibility Report by Omniscient Digital, only 8% of evaluated brands appear consistently in AI search recommendations. The report analyzed over 23,000 LLM citations and found that structured data quality was the single strongest predictor of visibility among technical factors.

What a pass looks like: Valid Product schema with all required properties plus at least 3 optional properties (description, brand, aggregateRating).

Test 4: Run Citation Tests Across AI Platforms (Layer 4)

This is the most important diagnostic. It tells you whether AI models actually surface your store when shoppers ask for product recommendations. You need to test on at least three platforms because each model has different training data, different citation behavior, and different indexing.

ChatGPT Test:

Ask ChatGPT (GPT-4 with web search enabled):

• “What are the best [product category] stores online?”
• “Where can I buy [specific product type]?”
• “Recommend online stores for [your niche]”

Do not mention your brand name. You are testing whether the model surfaces you organically.

Gemini Test:

Ask Google Gemini the same questions. Gemini has the advantage of Google’s index, so stores with strong Google SEO sometimes appear here even if they fail on ChatGPT.

Perplexity Test:

Perplexity is the most transparent. It shows its sources with citations, so you can see exactly which pages it references. Ask:

• “Best [product category] online stores 2026”
• “Where to buy [product type] online”

Scoring your results:

What to do if you fail: If ChatGPT does not mention your store, the issue is usually one of three things: (a) your store is not in the training data, (b) web search cannot find relevant pages, or (c) your content does not match the query pattern. Fix (a) by getting mentioned on external sites that AI models crawl. Fix (b) by improving crawl access and content structure (Tests 1-3). Fix (c) by writing answer-first content that directly addresses comparison and recommendation queries.

For more on why products fail to appear in AI recommendations, read our breakdown of why your products don’t show up when ChatGPT recommends them.

Data point: ChatGPT holds 64.5% of the AI search market as of March 2026, followed by Gemini at 21.5%, according to Datos and Stackmatix analysis. Together they account for approximately 45 billion AI search sessions monthly. Testing only Google means ignoring two-thirds of AI search traffic.

Test 5: Audit Your Product Feed Freshness (Layer 5)

AI agents that comparison-shop need current pricing, availability, and product details. If your product feed is stale, agents may surface outdated information or skip your store entirely.

Step-by-step:

1. If you use Google Merchant Center, check the “Diagnostics” tab for warnings
2. Export your product feed (XML or CSV) and check:
   • Are prices current?
   • Are out-of-stock items marked as unavailable?
   • Do image URLs resolve?
   • Are product descriptions present (not empty)?
3. Use a feed validator to check structural integrity

The freshness problem: A study of AI citation data found that 76% of ChatGPT’s top cited results were content published within the last 30 days. Stale product data does not just create a bad customer experience. It actively reduces your chances of being cited by AI models that weight recency.

For a complete walkthrough of feed testing tools, see our product feed validator guide for AI shopping agents.

What a pass looks like: Feed is updated at least weekly, all products have current prices and images, out-of-stock items are correctly flagged.

Test 6: Check for llms.txt and Agent Instructions (Layer 6)

The llms.txt file is a plain-text file at your domain root that tells AI agents what your store sells, how to navigate it, and what products to prioritize. It is becoming the robots.txt equivalent for AI model context. As of mid-2026, major crawlers including GPTBot and ClaudeBot check for this file.

Step-by-step:

1. Check if your store has an llms.txt file: https://yourstore.com/llms.txt
2. If it exists, verify it contains:
   • Your store name and category
   • Product categories with links
   • Key product pages
   • Contact and support information
3. If it does not exist, create one

Example structure:

# [Your Store Name]

> [One-line description of what you sell]

## Categories
- [Category 1](https://yourstore.com/category-1)
- [Category 2](https://yourstore.com/category-2)

## Featured Products
- [Product Name](https://yourstore.com/products/product-slug) - [Short description]

## About
- [Shipping info, return policy, contact]

For a full setup guide, read our llms.txt for ecommerce walkthrough.

What a pass looks like: A valid llms.txt file exists at your domain root with accurate category links, product links, and store information.

Putting It All Together: The Diagnostic Scorecard

Run all six tests and record your results:

Priority order: Fix crawler access first (Test 1). If bots cannot reach you, nothing else matters. Then fix content parseability (Test 2). Then schema (Test 3). Then work on citation optimization (Test 4). Feeds and llms.txt are important but secondary.

How Often to Run These Tests

AI agent behavior changes faster than traditional search rankings. New model updates can shift citation patterns overnight. A store that appears in ChatGPT recommendations today might disappear after a model update if its content does not match the new weighting.

What “Share of Model” Means for Your Store

A new metric is emerging in the GEO space called “Share of Model.” It measures the probability that a specific AI model will cite your brand when asked a relevant query. Unlike domain authority or keyword rankings, Share of Model is probabilistic and platform-specific.

You can approximate your own Share of Model by running Test 4 across 10 different query variations on each platform and tracking how often your store appears. A store with a 30% Share of Model for “best [niche] stores” on ChatGPT appears in roughly 3 out of 10 relevant queries. That is strong visibility. Most stores are at 0%.

Common Diagnostic Failures and Fixes

Failure: Crawler access is fine, but no citations.

This usually means the AI model has your pages in its index but does not consider them authoritative enough to recommend. Fix: get mentioned on external sites (blogs, directories, reviews) that AI models already cite. AI agents tend to recommend stores that appear across multiple trusted sources, not just on their own domain.

Failure: Schema is valid, but products show wrong data.

Schema validation checks structure, not accuracy. If your schema says a product is in stock when it is not, validators pass but AI agents serve wrong information. Fix: add automated checks between your inventory system and schema output.

Failure: Store appears on Perplexity but not ChatGPT.

Different models, different training data, different web search backends. Perplexity uses its own search index while ChatGPT uses Bing. If you are strong on Google SEO but weak on Bing, ChatGPT may struggle to find you. Fix: submit your sitemap to Bing Webmaster Tools and verify Bing indexing.

FAQ

How long does it take for AI agents to start showing my store after I fix discoverability issues?

It depends on the platform. Perplexity can surface new citations within days because it does real-time web search. ChatGPT with web search enabled can find new content within 1-2 weeks. Gemini relies on Google’s index, so changes follow Google’s typical crawl cycle of 1-4 weeks. Training data updates for base model knowledge happen on a longer cycle, typically months.

Do I need to test on every AI platform, or is checking ChatGPT enough?

ChatGPT has 64.5% AI search market share, so it is the most important. But Gemini at 21.5% represents a different user base and indexing system. At minimum, test on ChatGPT and Gemini. Perplexity is valuable because it shows source citations, making it useful as a diagnostic tool even if your customers are not using it heavily.

What if my store is very small with low traffic? Can AI agents still find it?

Yes. AI agents do not rank stores by traffic volume. They rank by content quality, structured data completeness, and citation authority. A small store with excellent Product schema, a well-written llms.txt, and mentions on a few authoritative niche sites can outrank a large competitor with poor structured data. The 2026 AI Search Visibility Report found that several smaller brands outperformed major retailers in AI citations within their specific niches.

Is there an automated tool that runs all six tests at once?

The diagnostic stack described here uses a combination of manual checks, free tools, and direct AI queries. Some GEO platforms (including shopti.ai) offer automated discoverability audits that cover crawler access, schema validation, and citation testing in one report. But the citation test (Test 4) still benefits from manual verification because automated tools cannot fully replicate how a real user phrases queries.

What is the single highest-impact fix for most stores?

Unblocking AI crawler access in robots.txt. It is the most common single point of failure. If GPTBot and PerplexityBot cannot crawl your store, you are invisible to two of the three major AI search platforms. Check your robots.txt first. It takes 5 minutes and fixes everything downstream.

Sources

1. Omniscient Digital / IssueWire. “2026 AI Search Visibility Report: Which Service Actually Gets Brands Recommended by ChatGPT, Perplexity, and Google AI.” May 7, 2026. https://www.issuewire.com/2026-ai-search-visibility-report-which-service-actually-gets-brands-recommended-by-chatgpt-perplexity-and-google-ai-1861410847710076

2. Stackmatix. “AI Search Market Share 2026: ChatGPT, Gemini, Perplexity & More.” March 2026. https://www.stackmatix.com/blog/ai-search-market-share-2026

3. Seer Interactive. “Google AI Mode: 25.1M Impression Study.” 2026. Reported 93% zero-click rate for AI Mode queries.

4. Bain & Company. “Goodbye Clicks, Hello AI: Zero-Click Search Redefines Marketing.” 2026. Reported 80% of consumers rely on zero-click AI results at least 40% of the time.

5. Datos / Datos Insights. “Q4-2025 AI Tool Traffic Report.” AI tool visits stabilized at 1.31-1.34% of US web traffic.

Check your store’s agent discoverability score free at shopti.ai.