AI visibility for US ecommerce: the price an assistant repeats is not the price at checkout

For a US online store the number an assistant repeats is the shelf price, and the shelf price is not what the shopper pays. American retail quotes before sales tax, usually before shipping, and often before the condition that makes shipping free, so a product page can be entirely accurate and still produce an answer the buyer finds wrong at checkout. The retailer controls that gap by publishing the parts of the total as text: the price, the tax sentence, the shipping cost and days, the free shipping threshold, and the returns window with who pays for the label. In our benchmark run on 15 October 2026, ChatGPT and Gemini each answered 102 buying questions in the United States, which is 204 US readings, and cited our own domain zero times on either engine. Across the full run of 597 readings, 85.6 per cent of 10,652 citations went to somebody's own website, which for a shopping answer means a retailer page, a marketplace listing or a brand site.

How far that measurement goes

It is one dated run: ChatGPT and Gemini, 15 October 2026, in the United States, the United Kingdom and India, 597 usable readings of 612 attempted. What the run did not do is measure any one industry on its own. The questions went out market by market, so there is no retail rate and no category rate in it, and what transfers is the source mix an American answer is assembled from. The method and the full citation table are in what AI assistants actually cite across 597 readings of ChatGPT and Gemini in three markets, and the American slice in the US cut of those 204 readings. One more boundary worth keeping straight. Five surfaces are what our product covers: ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews. This run is two of the five, and a shopping answer on the other three is not something these numbers describe.

The four numbers behind one American shopping question

What the shopper means	What has to be in text on the page
What does it cost me	Price, plus one sentence on how sales tax is applied at checkout
When will it arrive	Shipping options with a cost in dollars and a delivery window in business days
Is shipping free	The threshold as a dollar figure, and what it excludes
What if it is wrong	The returns window in days, who pays the label, and the restocking position

Those four facts sit in four different places on most American stores: the product page, a shipping page, a banner and a policy page written by somebody in 2022. An assistant assembling an answer takes whichever it finds, which is how a store ends up quoted at a price and a delivery promise it does not offer.

Write the total, next to the product

The change that moves this is small and nobody does it. Under the price, in text, put the shipping options with their cost and their delivery window, the free shipping threshold, the sales tax sentence, and the returns window. Keep it to five lines and date it. Then do the same on the category page, because a comparison question is answered from category pages far more often than from a single product page.

Make sure those numbers survive the page rendering. If the shipping cost is injected by an app after load, treat it as a number the fetcher may never see. The product data work underneath all of this, from markup that matches the visible page to one canonical URL per product, is set out in AI visibility for ecommerce. This page is the American layer: the total, the claims, and the surfaces that carry them.

Claims an American product page has to get right

US retail claims are regulated in ways that also make them quotable, which is a convenient overlap.

• Country of origin. The Federal Trade Commission's Made in USA standard requires that a product advertised that way is all or virtually all made in the United States. Where the truth is assembled in the United States from imported parts, write that sentence instead. It is more specific, so it is more useful to a model and to a buyer.
• Reviews and endorsements. The FTC's endorsement rules require a material connection to be disclosed and treat fabricated or incentivised reviews as deceptive. Say how reviews are collected and whether anything was given in exchange.
• Comparison and savings claims. A was price has to be a price you actually charged, and the period matters. Dating the comparison is what makes it defensible and what makes it worth quoting.
• California warnings. Where a Proposition 65 warning applies, put it in text on the product page rather than in an image, because a warning nobody can read is a warning nobody can act on.

The surfaces you share with everybody else

An American shopping answer is rarely built from one store. It is built from a marketplace listing, a retailer page, an editorial roundup and a review platform, and most of those are outside your control. Three practical moves: keep the marketplace listing carrying the same title, specification and current price as your own page, pitch the roundups that actually get cited in your category rather than all of them, and answer questions under your own name in the communities where your product gets discussed.

What is not worth a quarter of work is a directory push. Review directories took 0.8 per cent of the 10,652 citations in the run, which we counted individually in our count of every directory citation. Keep the profiles current and spend the time on the pages carrying your totals.

Checking what an assistant currently says about your prices

1. Pick five products where you know the real total, including tax and shipping to a named state.
2. Ask both ChatGPT and Gemini what each product costs and where to buy it, twice each, memory off, without naming your store.
3. Write down the price quoted, the retailer named, and every source shown.
4. For each wrong number, find the page it came from. It is usually a marketplace listing, an affiliate page or your own cached price, and it is fixed at the source rather than by arguing with the output.

Where our interest lies: we sell the work. For a store that means the total block, the claim wording and the listing clean up, done by AI Knows Us rather than by your team. 64 of the 102 build spec conversations asked for the commercial interest to be declared where a reader can see it, so here is ours alongside our own result from the same run, which is 3 citations in 597 readings and none in the United States.

Questions US retailers ask about being named by an assistant

Should we publish a price that includes estimated tax?

Publish the price you charge and one sentence describing how tax is applied, since the rate depends on the shipping address. The sentence is what makes the quoted price understandable, and it is cheaper than a calculator.

Why does an assistant quote a sale price that ended last month?

Because a page somewhere still carries it: a cached version, an affiliate post, a marketplace listing or a stale feed. Correct it at the source, then publish the current price in text with the date you checked it, and re test in two weeks.

Do our own on site reviews help?

They give a model text to summarise about the product and they belong on the page, marked up to match what a visitor sees. Independent reviews carry further because they are a second source saying the same thing, so both are worth having.

Our catalog is thousands of products. Where do we start?

With the top category pages and the twenty products that carry the revenue. Add the total block and the returns line to those first. A constrained shopping question is usually answered from a category page, so that is where the first week pays.

What to change on one product page this week

Take your best selling product, open its page in view source, and look for four numbers: the price, the shipping cost, the free shipping threshold and the returns window. On most American stores you will find one of the four. Put the other three into text under the price, with a review date, and do the same on the category page above it. That is the whole change, and it is the change that decides whether an assistant quotes a total your shopper will recognise at checkout or a number that makes your store look wrong.