Amazon's shopping assistant will now watch a product and buy it the moment it reaches a price the shopper named: a Prime member sets the threshold, the assistant checks prices on their behalf and will "complete the purchase agentically using your default payment method and shipping address", with a free 24-hour cancellation and the request staying armed for six months (Amazon's own announcement, checked August 2026). For your own store the significant change isn't that software can buy — it's that a published price stops being a message and starts being a trigger.
Price is the field we handle most carefully in Contexta, and threshold buying is why. A number that used to be a persuasion problem is becoming an execution problem, where the wrong copy of it — the one in a feed nobody re-uploaded — can close a sale on terms you didn't intend to offer.
What does buying at a target price actually do?
It converts a shopper's intent into a standing instruction that executes without either party present. The shopper says something like "buy these when they drop to $25", the assistant monitors the price, and when the condition is met it places the order using stored payment and shipping details. Nobody re-reads the product page at the moment of purchase, which is the part that matters for merchants.
Two details make the mechanic more consequential than a price alert. It persists — six months is long enough that the person who armed it has forgotten the product exists. And it's paired with visible price history: Amazon shows 30- and 90-day price trackers, and its Alexa for Shopping assistant surfaces up to a full year, so the shopper arming the threshold is looking at your discount pattern, not just today's number.
One naming note, since half the coverage is already dated: Amazon renamed Rufus to Alexa for Shopping on 13 May 2026, appending an update to that effect to its own original announcement — the same rename that quietly ages every "Rufus guide" published since, and one we mapped alongside the other assistants in what actually separates the AI shopping agents.
Can an agent auto-buy from an independent WooCommerce store?
Not today. As of August 2026 threshold buying is a first-party feature running against Amazon's own inventory and checkout, so unless you sell on Amazon nobody is arming a trigger on your product page. Anyone telling a WooCommerce owner that agents are already auto-buying from independent stores is selling urgency rather than describing a live system.
What has already arrived for you is the enforcement half, and it's the half that costs money right now. Google's Merchant Center automations are enabled by default and update price, sale price, availability and condition from your landing page — Google's own worked example is a product submitted at $4 whose landing page says $3, which it updates to $3 in your listings (Merchant Center Help, checked August 2026). Google also states plainly that automations aren't a substitute for keeping your product data current; they exist to patch temporary inaccuracy on a small share of products.
That's the same discipline threshold buying will eventually require, arriving early and with penalties attached. When feed and landing page disagree too often, the account gets warned or products get preemptively disapproved, and the manual review that clears it takes seven business days. Your price is already being read as data by a system that acts on it — the structured record an agent assembles about your product simply doesn't include a footnote saying which copy you meant.
Which copy of your price would an agent actually watch?
Whichever one the surface it uses happens to read — and a WooCommerce store publishes at least four, refreshing on four different clocks. Changing a price in the admin updates the page immediately and the rest whenever their own schedules come round, which means for a window measured in hours your store is quoting two prices at once.
Source of truth
You change the price
admin, one field, instant
Rendered product page
immediate, and cached by your CDN for as long as the page TTL says
JSON-LD in the page source
immediate if the plugin regenerates it, stale if the block is cached separately
Merchant Center feed
changes at your next scheduled upload — hours, unless you push via the API
Copy held by an assistant
refreshed on its own crawl or feed cycle, on a schedule you can't see
The last row is the uncomfortable one. You control the first three and you can measure the third; you have no visibility into when an assistant last refreshed its copy, which is why "we fixed it an hour ago" is not an answer to a shopper quoting a price you no longer offer. This is the practical reason feed hygiene decides what the Shopping Graph knows about you: the graph isn't reading your intentions, it's reading the last snapshot it took.
Sale periods are where this bites hardest, and in a specific direction. A discount that goes live on the page before the feed catches up leaves the higher price advertised on the surfaces where discovery happens — you run the promotion and pay for none of its visibility. Run it the other way, with the feed dropping first, and you've published a price your checkout won't honour, which is the version that generates support tickets and, on Google, disapprovals.
What pricing discipline does this actually require?
One price, changed in one place, propagated everywhere before it goes live — and never a published price you can't honour for the length of your slowest refresh cycle. That last clause is the whole rule in practice: if your feed uploads every six hours, then any price you publish is an offer you've made for at least six hours, whatever your admin says now.
Four habits follow from it, and none are exotic:
- Schedule promotions to the slowest clock, not the fastest. Set the sale start far enough ahead that the feed upload lands first, so the discount is live everywhere or nowhere.
- Push price changes through the API rather than waiting for the next file. A scheduled upload is fine for a catalog that changes weekly and wrong for one that reprices daily.
- Use the sale price attribute rather than overwriting the regular price. Google asks for the lowest price to be the most prominent and submitted as sale price, and it's the only way the discount reads as a discount rather than as a new baseline.
- Never let editorial content hard-code a number. A buying guide quoting last quarter's price is one more copy of your price, disagreeing with the other four.
That last one is where our own product has an opinion. Contexta's AI editor is price-safe by rule: when it writes or rewrites product and category copy it pulls prices and specs from the live catalog rather than letting the model produce a number, because a generated price is a plausible-looking figure with no source. It governs the content Contexta writes and nothing else — your feed export and your theme are still yours to keep honest — but it removes the copy of your price that's easiest to forget about, which is the argument for binding editorial content to the real catalog generally.
Is discounting into agent-watched demand a good idea?
Sometimes, and less often than the feature makes it look. Amazon reports customers saving on average 20% per auto-buy purchase (its own figure, checked August 2026), which is the same sentence read from the merchant side: a fifth off, given to buyers who had already decided to buy and were willing to wait. Whether that's efficient depends entirely on whether they would have paid full price eventually.
- Armed thresholds convert the instant you cross them, with no ad spend and no campaign
- Demand that was invisible — people waiting for a number — becomes addressable
- A genuine, well-timed discount reads cleanly against a visible price history
- The trigger fires on whichever copy of your price the assistant holds, including a stale one
- There's no moment to reconsider: no cart, no checkout page, no human on either side
- Visible 30-, 90-day and year-long price history makes an inflated "was" price checkable
- A predictable discount rhythm teaches buyers, and their assistants, to wait for it
The honest limit on all of this is timing. Threshold buying is real, well-documented and confined to one retailer's own inventory as of August 2026, and nobody outside those companies can tell you when — or whether — the same mechanic reaches independent stores through Google's or OpenAI's checkout paths. That argues against re-planning your pricing strategy around it and firmly for the unglamorous half, which pays today: one source of truth for price, propagation you can time, and no number published anywhere you couldn't honour for the next six hours.
FAQ
Can an AI agent automatically buy from my WooCommerce store?
Not as of August 2026 — threshold buying, where an assistant purchases when a price hits a target, runs against Amazon's own inventory and checkout rather than independent stores. Agentic checkout paths on other surfaces exist but do not currently let a shopper arm a price trigger on your product page. What already applies to you is the data side: Google reads your price as machine data and acts on it, including disapproving products when your feed and landing page disagree.
What happens if my feed price and my product page price disagree?
Google's Merchant Center automations are on by default and will update the listing from your landing page — its own example takes a product submitted at $4 whose page shows $3 and updates the listing to $3. If the mismatches are frequent, the account gets warned or products are preemptively disapproved, and the manual review to clear that takes seven business days. Google is explicit that automations patch temporary inaccuracy and are not a substitute for keeping your product data current.
How far ahead should I schedule a WooCommerce sale price?
Far enough ahead that your slowest feed upload lands before the discount goes live on the page, so every surface shows the same number at the same moment. A discount that hits the page first advertises the old, higher price everywhere discovery happens; a feed that drops first publishes a price your checkout won't honour. If you reprice more often than your upload schedule runs, push changes through the Content API instead of waiting for the next scheduled file.
Does visible price history change how I should run discounts?
Yes — Amazon shows 30- and 90-day price trackers and up to a full year in Alexa for Shopping, so an inflated "was" price is now checkable by the shopper before they buy. A discount that reads as genuine against your own price history carries more weight than a larger one that clearly follows a monthly rhythm. The practical risk of a predictable discount cycle is that buyers, and the assistants watching on their behalf, simply wait for it.