Skip to content
← Back to writing
Writing · ai visibility

Sold out on Black Friday, still in stock for AI agents

By Leo Nguyen · Nov 19, 2026 · 6 min read
Sold out on Black Friday, still in stock for AI agents
Jump to section›

Short version: Black Friday is on 27 November. On a normal week, a product that sells out and still reads "in stock" to an AI agent is a small leak. During BFCM it's a different problem. Stock moves in hours, more shoppers ask an assistant to do the first pass, and every recommendation of a sold-out item sends a buyer to a page they can't buy from, on your busiest day of the year.

#Two signals, one of them lags

When a variant hits zero, the storefront usually reacts fast. The button greys out, a "Sold out" badge appears, and a person on the page sees it right away.

An AI shopping agent doesn't look at the badge. It reads the availability value on the Offer in your JSON-LD, and on many surfaces it reads your product feed too. Those are separate outputs. They're generated by different code, sometimes by different apps, and they don't always update when the badge does.

So the shopper asks "find me this in a size M, in stock, arriving before Friday." The agent finds your product, sees InStock in data that's a few hours old, and recommends it. The shopper clicks through and gets a sold-out page. They don't blame the agent. They bounce, and the next time that agent has a choice, it has one more reason to trust somebody else's data over yours.

We wrote about the opposite failure before: agents saying you're out of stock when you're not. That one costs you sales you could have had. This one costs you trust at the exact moment traffic peaks.

#Where the lag usually comes from on Shopify

These are patterns we look for when reading Shopify storefronts, not measurements of any one store. On a given site it's usually one of them, sometimes two.

Product-level availability on a variant product. A lot of theme JSON-LD outputs one Offer built from product.available. That's true as long as any variant is in stock. Size M sells out, sizes S and L are left, and the schema keeps saying InStock for a URL that a shopper may have landed on with M already selected. Fix: one Offer per variant, each with its own availability from variant.available, and the selected variant's URL in the offer.

"Continue selling when out of stock" left on. If the variant allows overselling, Shopify treats it as available, so the schema says in stock while your warehouse says zero. Sometimes that's on purpose for pre-orders. Then the schema should say PreOrder or BackOrder, with a date, not InStock.

A second JSON-LD block from an app. Review and SEO apps often add their own Product node. Some build it from cached data. The theme's block flips to OutOfStock, the app's block doesn't, and the agent now has two answers on one page.

The feed runs on a schedule. The Google & YouTube channel and most feed apps don't push a change the second inventory moves. If your feed app syncs every few hours, that's how long a sold-out item can stay in stock to anything reading the feed. Merchant Center will flag a mismatch it finds on crawl ("mismatched value: availability"), but by then you're already in the leak.

#The shipping promise is the other half

"Ships before Friday" is part of the same question. If your cutoff only lives in an announcement bar, a countdown widget or an image, an agent has nothing to read. It either skips you or guesses.

Put it where it can be parsed:

  • The cutoff as plain text on the product page or a shipping page that's linked from it. "Order by 18 December for delivery before Christmas" in actual HTML.
  • OfferShippingDetails in the JSON-LD, with handlingTime and transitTime in days and a shippingDestination.
  • Merchant Center shipping settings that say the same thing. If the page promises 2 days and the feed says 5, you've given the agent two answers again.

#A 15-minute test before the 27th

You don't need a tool for this. Do it on a duplicate product or a preview theme if you don't want to touch a live bestseller.

  1. Pick a bestseller with variants. Set one variant to zero and make sure "continue selling when out of stock" is off.
  2. Open the product page with that variant selected. View source (not inspect), search for "availability", and read what the Offer says. You want OutOfStock on that variant.
  3. Check whether there's more than one Product block in the source. If there are two, check both.
  4. Open your feed app, or the product in Merchant Center, and note when it last synced and what availability it shows. Check again after the next sync.
  5. Search the page source for your shipping cutoff date. If you can't find it as text, an agent can't either.

If the badge says sold out and anything in steps 2 to 4 still says in stock, that's your BFCM leak. If the cutoff isn't in step 5, that's the second one.

#What to fix first, with 8 days left

Don't start a theme rebuild this week. In order of what usually pays off fastest:

  1. Make the theme's Offer per-variant and read from variant.available. On most themes that's one snippet.
  2. Remove or switch off the duplicate Product JSON-LD from whichever app is adding it.
  3. Shorten the feed sync interval for the sale window, if your feed app allows it.
  4. Write the shipping cutoff into the page as text and add OfferShippingDetails.

Then run the test again. Write down the date and what each source said, so you have a before and an after.


Want a second pair of eyes before the weekend? Send one store URL and we'll run a free async AI-correctness read: we check what your structured data, feed and shipping data tell AI agents about stock and delivery, and send back a list of where they disagree with your storefront: luma-e.com/ai-readiness. No call required.

Frequently asked
Why would an AI agent recommend a product that's already sold out?›
Because it doesn't read the sold-out badge. It reads the availability value in the page's JSON-LD and in your product feed. If inventory hits zero and the badge changes but the JSON-LD still says InStock, or the feed hasn't synced since morning, the agent still sees an in-stock product and recommends it. The shopper clicks through and lands on a page they can't buy from.
How do I test whether my Shopify structured data and feed agree on stock?›
Pick a bestseller, or one variant of it, and set its inventory to zero, ideally on a duplicate product or a staging theme. Open the product page, view source, and find the Offer availability value in the JSON-LD. It should read OutOfStock for that variant. Then check the same item in your feed app or in Google Merchant Center after the next sync. If the page says sold out and either of those still says in stock, that's the gap.
Where should my shipping cutoff live so AI agents can read it?›
In text and data, not only in an image banner or a popup. State the cutoff date in plain text on the product or shipping page, declare delivery times in OfferShippingDetails in your JSON-LD, and keep your Merchant Center shipping settings in line with the same numbers. An agent asked 'ships before Friday?' can only answer from something it can parse.
NEWSLETTER

More like this, only when it's worth it

Occasional email on AI x ecommerce - how agents read stores, what breaks, what we shipped. No weekly filler.

No spam, no sequence-drip. Unsubscribe anytime.