How to Test If AI Agents Can Find Your Products

The prompts to ask, the answers to look for, and what it means when agents get it wrong.

Right now, millions of people are asking AI agents to help them shop. They ask things like "what's the best moisturizer for dry skin under $40?" or "recommend a running shoe for flat feet." The agent responds with product recommendations. Sometimes it recommends your product. Most of the time it doesn't.

The question most brands have never asked themselves: what happens when someone asks an AI agent about my product category?

We tested this across 250+ brands and 5 product categories for our agent blind spot research. The results were bad. Agents recommended discontinued products, got prices wrong by 30-50%, and ignored top brands entirely.

Here's how you can run the same tests yourself, right now, for free.

Step 1: The Brand Recall Test

Open ChatGPT, Claude, or Perplexity and ask:

"What do you know about [your brand name]?"

What to look for:

• Does the agent know you exist? If it says "I'm not familiar with that brand," you have a fundamental visibility problem.
• Is the description accurate? Does it correctly describe what you sell, your price range, and your value proposition?
• Is the information current? AI agents often have training data that's months old. If they describe products you discontinued last year, that's a staleness problem structured data can fix.

What it means: If the agent can't describe your brand accurately, nothing else matters. This is your baseline. The fix starts with Organization and Product Schema markup.

Step 2: The Category Search Test

Ask the agent to recommend products in your category:

"What are the best [your product category] under $[your price point]?"

For example:

• "What are the best wireless earbuds under $150?"
• "Recommend a good protein powder for post-workout recovery."
• "What's the best organic baby formula?"

What to look for:

• Are you in the list? If the agent recommends 5 products and you're not one of them, agents don't consider you a top option in your category.
• Are your competitors there? If competitors show up and you don't, figure out what they're doing differently (hint: check their Schema.org markup).
• Is the pricing right? We found agents getting prices wrong by 30-50% in our testing. If the agent quotes a price that doesn't match your site, you need priceCurrency and price in your Product Schema.

What it means: This test reveals your competitive position in AI-mediated discovery. If you're not showing up, the agent's training data doesn't include enough structured information about your products.

Step 3: The Product Detail Test

Ask about a specific product:

"Tell me about [exact product name] by [your brand]. What does it cost and is it in stock?"

What to look for:

• Does it know the product? If the agent confuses it with a competitor's product or makes up features, your product data isn't structured enough.
• Is the price correct? This is the biggest failure point we see. Agents pull prices from training data, which can be months old. If you've changed prices since then, the agent will quote the old price.
• Does it link to your site? Perplexity typically includes source links. ChatGPT and Claude may not. Check if the agent knows where to send the customer.
• Does it know availability? An agent recommending an out-of-stock product creates a terrible customer experience.

What it means: Product-level accuracy requires complete Product Schema with current pricing, availability, and identifiers (SKU, GTIN). An MCP server solves the staleness problem by giving agents live data.

Step 4: The Comparison Test

Ask the agent to compare you against a competitor:

"Compare [your product] vs [competitor product]. Which is better value?"

What to look for:

• Does the agent have enough data to make a fair comparison? If it has detailed specs for your competitor but vague descriptions for you, you're losing on information asymmetry.
• Are the ratings accurate? Agents often cite review scores. If yours aren't in your Schema markup, the agent may undercount or ignore them.
• What does it recommend? The agent's recommendation reveals how it weighs different factors. If price is wrong, you might lose comparisons you should win.

What it means: AI-mediated comparisons are the new product reviews. Brands with better structured data get more accurate (and often more favorable) comparisons. This is where aggregateRating, complete product specs, and FAQ Schema give you an edge.

Step 5: The Purchase Intent Test

Ask the agent to help you buy:

"I want to buy [your product]. Where should I get it and what will it cost?"

What to look for:

• Does it send customers to your site? Or does it send them to Amazon, a reseller, or a competitor?
• Is the price it quotes your current price? Stale prices create friction. If the agent says $59.99 and your site says $79.99, the customer feels deceived, even though neither you nor the agent intended it.
• Can the agent complete the purchase? Today, most agents can't. But Google's UCP is changing this. Brands with UCP profiles will be the first to enable AI-completed purchases.

What it means: This is the bottom of the funnel. If the agent can't accurately point customers to your store with the right price, you're losing conversions you never see in your analytics.

Score Your Results

After running all five tests, grade yourself on each:

0-1 passes: You're invisible to AI agents. Start with Schema.org basics.

2-3 passes: Agents know you exist but can't sell you effectively. Focus on Product Schema completeness and FAQ markup.

4-5 passes: You're ahead of most brands. Consider MCP and UCP for real-time data and AI-enabled checkout.

The Automated Version

Running these prompts manually works, but it's time-consuming and hard to repeat consistently. The Pacestack scanner automates the technical side: it checks your Schema.org markup, robots.txt, API endpoints, MCP configuration, and more. It gives you a score out of 100 and tells you exactly what to fix.

The manual tests above tell you how agents perceive your brand. The scanner tells you why. Both are useful. Together, they give you the complete picture.