AI Commerce WatchTracking the agents that shop

Oct 2, 2026 - AI Commerce Watch

What do AI shopping agents do when a product is out of stock?

An out-of-stock product is a dead end for a human shopper and a decision point for an AI agent. What the agent does next depends on signals most stores never designed for machines.

The agent does not see "out of stock" the way you do

When a human sees an out-of-stock notice, they understand the situation instantly: the thing is gone, come back later or pick something else. An AI shopping agent has to infer all of that from page structure. Is the product gone temporarily or discontinued? Is there a restock date? Is there a close variant available? Most product pages answer these questions with visual cues a human parses effortlessly and structured data an agent may miss entirely.

This matters because agents act on incomplete information. An agent told to buy a specific SKU will, upon finding it unavailable, do whatever its instructions and the page allow. That might mean buying the closest variant, waiting and retrying, or abandoning the purchase. The store does not get a vote unless it designed the page to cast one.

The three things agents actually do

First, substitution. Many agents are instructed to find the closest alternative when the exact item is unavailable. They will pick a different size, color, or a nearby product, sometimes with and sometimes without confirming with the user. If your variant structure is clean and your product relationships are explicit, the substitution is usually sensible. If your catalog is messy, the agent may buy something bizarre.

Second, waiting. Some agents retry the product page on a schedule, checking for restock. This creates a distinctive traffic pattern: regular, patient polling of a single out-of-stock URL for days or weeks. It looks like a bot because it is one, but it is a bot acting on a legitimate purchase intent.

Third, abandonment with a report. Better agents tell the user the item is unavailable and suggest alternatives or a restock alert. This is the outcome stores should want, and it depends on the page giving the agent something accurate to report.

Structured data is the vote the store gets

The difference between a good agent outcome and a bad one usually comes down to structured data. A product page with proper availability markup, a restock date where one exists, and explicit variant relationships gives the agent facts to work with. A page where availability is a CSS class on a button and the restock date is buried in prose gives the agent guesses.

The highest-value additions are specific: machine-readable availability status, an expected restock date when you have one, and clear discontinued flags when you do not. "Notify me" functionality that works without an account is a bonus. Every one of these turns an agent guess into an agent fact, and facts produce better purchase behavior.

The retry traffic is real and growing

Stores are starting to notice the polling pattern in their logs: the same product URL requested at regular intervals by automated clients, sometimes for weeks. Before treating it as hostile bot traffic, check what the client is doing. An agent polling an out-of-stock page every six hours is not scraping your prices or probing your infrastructure. It is a customer waiting in line.

Blocking that traffic has a cost. The agent will report the product as unavailable or the store as broken, and the purchase goes to a competitor whose pages the agent could read. Rate-limiting is reasonable; blanket blocking of restock polling is turning away buyers.

Design the out-of-stock page for two audiences

The out-of-stock page has always had two jobs: keep the human shopper engaged and preserve the SEO value of the URL. Now it has a third: inform the agent. The good news is that all three jobs want the same things: clear availability status, alternatives that are actually in stock, a restock date or notification option, and honest discontinued flags.

Treat the agent as a shopper with perfect patience and zero intuition. It will not notice the banner image, it will not infer the restock date from your Instagram, and it will not guess that "more on the way" means three weeks. Write it down, mark it up, and the agents will do the right thing most of the time.