Black Friday 2026: AI isn’t buying for your customers. It’s steering them with what it finds.

In this article

1. 1. AI is already sending shoppers to online stores. But how many?
2. 2. Buying directly inside an AI: still the exception
3. 3. ChatGPT, Google, Amazon: not every assistant plays the same role
4. 4. What AI assistants can actually read from your catalog
5. 5. For French and French-founded brands in the U.S., the real risk is often elsewhere
6. 6. Are you ready? 11 checks to run before Nov. 27
7. 7. In January, we’ll be able to measure what actually happened

In the U.S., Black Friday 2026 is shaping up to be the first one where AI assistants carry real weight in how shoppers discover products, though not necessarily in the purchase itself.

For any brand selling in the U.S., the priority isn’t getting ChatGPT, or any other AI assistant, to take orders. It’s checking what that AI is telling shoppers about your products, your prices and your delivery times. And with less than a month to go before Black Friday, you may still have a few things to fix.

This year, Thanksgiving falls on Nov. 26, Black Friday on Nov. 27 and Cyber Monday on Nov. 30. Between now and then, millions of American consumers will hunt for deals and gift ideas, compare products, check prices and wonder whether their order will arrive before the holidays. A growing share of that research now happens in ChatGPT, Gemini, Copilot and other assistants.

But three behaviors need to be kept apart: asking an AI for advice, landing on a store because of it, and letting it handle the entire purchase. They aren’t growing at the same pace or on the same scale. For brands getting ready for the holiday season, that distinction changes the priorities quite a bit. Selling stays, tickets or reservations rather than products? The same logic applies.

1Asking an AI for adviceCommonComparing options, finding a gift idea, checking a price.

2Landing on a store because of itGrowing fastStill low volume, but visitors often further along in their decision.

3Letting it handle the whole purchaseMarginalMost shoppers still prefer to check out on the seller’s site.

A reading guide, not a quantitative scale. Each behavior is measured by different studies, across different populations.

1. AI is already sending shoppers to online stores. But how many?

Adobe expects traffic from generative AI tools to U.S. retail sites to grow 130% in November and December 2026 compared with the same period in 2025. An important caveat: this means visits to retail sites after an AI interaction, not orders placed inside an assistant.

And that growth starts from a very small base. According to L.E.K. Consulting, in an analysis published in spring 2026, traffic clearly identifiable as coming from AI accounted for less than 0.5% of inbound traffic to e-commerce sites at the time. That’s an observation, not a year-end forecast: even if it doubles by the holidays, it will remain marginal. A channel can grow 130% without moving the needle on your acquisition. We’ve seen it before with some social platforms in their early days: spectacular growth doesn’t always translate into meaningful sales volume.

+130%Projected AI traffic growth

Visits referred by AI to U.S. retail sites, 2026 holiday season vs. 2025. Adobe, forecast.

<0.5%Share of inbound traffic

Share of total inbound traffic identifiable as coming from AI. L.E.K., observed.

Two different measures: a year-over-year growth rate and a share of traffic. They don’t add up.

The quality of those visits, however, deserves attention. A shopper who asks an AI to compare three robot vacuums under $300 that handle pet hair and can arrive before the weekend lands on your site with a very specific idea of what they want. The AI has already done part of the qualifying. It’s a reasonable commercial hypothesis, not yet a certainty: nothing guarantees that every AI-referred visitor will convert better.

What we do know today is that AI doesn’t just qualify shoppers. It changes their minds. According to Attentive (August 2026, 3,054 U.S. adults), 62% of shoppers who used AI ended up choosing a different product or brand than they usually would. But 93% also looked for information or proof outside the AI before buying. The survey doesn’t say where, but the options are limited: the brand’s website, a retailer, a marketplace, reviews. AI makes its case with what it finds, and your customers go check the same places. If the two don’t match, you lose on both counts.

In other words: this season, AI will send you few visits. But it can sway a lot of buying decisions, bringing you new customers or sending yours to a competitor.

2. Buying directly inside an AI: still the exception

Agentic commerce rests on a simple idea: instead of searching, comparing and ordering on their own, consumers hand off some of those steps to an assistant. In its most advanced form, the agent picks the product and pays for it.

Agentic commerce already exists in the U.S., at Google, Microsoft, Meta and Walmart. It remains marginal in actual use, and it isn’t open to consumers in France yet.

In practice, American consumers still want to complete their purchase on the seller’s site. According to L.E.K. (April 2026), 8% of AI users have completed at least one purchase entirely through an agent, without visiting a brand’s site. The study doesn’t say what share of their purchases they hand off this way: a single order is enough to be counted. According to Attentive (August 2026), 60% of AI-assisted shoppers prefer to finish their purchase on the seller’s website or app. The two studies survey different populations, so there’s no single adoption rate to draw from them. But they point in the same direction.

Even OpenAI has adjusted course: since March 2026, ChatGPT has focused on product discovery and comparison, and usually sends shoppers to the merchant’s site to pay.

So what’s holding agentic commerce back is no longer really technology. It’s people, first and foremost. Consumers aren’t ready to hand over the final decision, and many merchants haven’t yet prepared their data to be read by a machine.

For a growing e-commerce brand, investing in agentic checkout infrastructure right now isn’t the priority. Making sure assistants have reliable information about your catalog is.

3. ChatGPT, Google, Amazon: not every assistant plays the same role

There’s no single model for agentic commerce in the U.S. Here’s where the main players stand as of late October 2026, all launching in the U.S. first.

U.S. status as of late October 2026. These features change fast: double-check before making any investment decision.

Two takeaways. First, each assistant draws on its own sources: the same product can be well presented in Google and poorly, or not at all, in ChatGPT. Second, if you sell mostly on Amazon, agents from other platforms won’t be able to buy your products there.

This ties back to our analysis of DTC, Amazon and retail models: the channel that influences the purchase isn’t always the one that closes the sale. With AI, that distinction matters even more.

4. What AI assistants can actually read from your catalog

Take a product sold for $129 on a U.S. e-commerce site, in stock, with delivery in four business days. But the feed sent to Google still shows the old $149 price, a retailer’s listing shows a different SKU, and the return policy is buried on a generic page.

Your product page

$129

In stock · delivery in 4 business days.

Google Merchant Center feed

$149

Old price, never updated.

Retailer listing

Different SKU

A variant, or an older version?

Return policy

Hard to find

Buried on a generic page.

Which version will the assistant repeat? There’s no way to know for sure.

These systems don’t all check the same sources, or refresh them at the same pace. What we do know is which kinds of data feed product discovery and comparison:

• Product feeds: SKUs, descriptions, prices, availability, images, sometimes shipping details. Shopify Catalog, Google Merchant Center and direct merchant feeds play a central role here.
• Product pages: features, use cases, variants, selling points.
• Structured data: Product, Offer, OfferShippingDetails and MerchantReturnPolicy markup help Google identify prices, stock, shipping costs and return terms.
• Third-party sources: marketplaces, retailers, reviews and other public content.

Two shortcuts to avoid. Adding structured data doesn’t guarantee more recommendations: Google doesn’t present it as a requirement for appearing in its AI answers; it mainly makes certain commerce information more explicit. And the llms.txt file, sometimes pitched as the new must-have, has no proven commercial impact to date.

Fixing a wrong price in Merchant Center is worth more than days spent on a file whose effect is still unproven.

5. For French and French-founded brands in the U.S., the real risk is often elsewhere

A French company selling in the U.S. is often juggling several realities at once: a European catalog, U.S. pricing, different logistics, sometimes Amazon and a network of local distributors. Those differences are normal. The trouble starts when they become hard to make sense of for an American shopper, or for the assistant helping them compare. We see three areas to watch.

Delivery times that are clear, product by product

Every e-commerce operator knows delivery time weighs on the decision. What’s new is that an assistant can show your delivery time right next to a competitor’s, at the exact moment the shopper is comparing. If some of your products ship from a U.S. warehouse and others from France, say so product by product. Otherwise, in the shopper’s eyes, your longest international delivery time may end up applying to your whole catalog.

Prices that are consistent within each market, not necessarily identical across markets

The same product can legitimately cost more in the U.S. than in France: taxes, logistics, distribution margins. What causes problems is conflicting information: a promotion announced on the U.S. site but missing from the feed, an old price still showing at a retailer, a French product description copied onto a U.S. marketplace that is no longer accurate.

In our article on consistency between French and U.S. websites, we showed how a single product can end up existing in several versions across the official site, Amazon and distributors, without anyone deciding it should. Agentic commerce doesn’t create those gaps. It makes them more visible at the moment offers are compared.

Return policies designed for American shoppers

An American shopper needs to understand within seconds how long they have to return a product, who pays for the return and where to send it. A policy can be legally airtight and still fail to reassure if it’s hard to find or written for the European market.

Here again, this is a working hypothesis, not a proven ranking factor. But it’s concrete enough to warrant a check before the season.

6. Are you ready? 11 checks to run before Nov. 27

With less than a month to go before Black Friday, this isn’t about rebuilding your site or launching an AI search overhaul. It’s about locking down the information platforms and shoppers will rely on during the holidays. Aim to have your product pages and feeds up to date by Nov. 15: Google and AI assistants don’t re-read an updated page or feed right away, and it can take a few days for changes to be picked up. That’s not an official deadline, just a safety margin before Thanksgiving.

Your discounts don’t need to go live early. What needs to be accurate by Nov. 15 is your core information (regular prices, availability, delivery times, returns) and its consistency across channels: your site, feeds, marketplaces and retailers. Sale prices can be scheduled in your feed so they only show during the promotion (the sale_price_effective_date attribute in Google Merchant Center).

Here are 11 checks you can run without a technical project. The goal: make sure AI finds accurate information about your products when it recommends them, and that your customers find exactly the same information when they go check.

Check items off as you go: your progress is saved in this browser.

1. Check 1The mobile checkout flow, starting from an AITest on a real phone, not a resized desktop window, by opening your store from a link suggested by ChatGPT. On mobile, shoppers land on your site without leaving the app, in a built-in browser, much like tapping a link in Instagram: your site has to work flawlessly there. Common pitfalls: a cookie banner or sign-up pop-up covering the buy button, Apple Pay or Google Pay missing, mandatory account creation, shipping costs revealed only at the last step, a promo code field nobody can find.
2. Check 2Your top products, identical everywhere they’re soldSite, feeds, Amazon, retailers: same name, same variants, same GTIN, same availability. Start with your best sellers.
3. Check 3What your platform sends to AIShopify, WooCommerce, PrestaShop, BigCommerce: check which feeds your platform powers (Google Merchant Center, Meta, marketplaces). Since March 2026, Shopify has shared its merchants’ catalogs with several AI assistants by default. On WooCommerce, PrestaShop or BigCommerce, this generally requires an extension or a feed you configure yourself: check what’s actually in place.
4. Check 4Google Merchant CenterFix disapproved products and critical warnings.
5. Check 5U.S. delivery timesA realistic delivery time per shipping method, in days, making it clear what ships from the U.S. and what ships from Europe.
6. Check 6Your returns, through an American shopper’s eyesIs the policy one click away from the product page, written in English, in days and dollars? Does it say who pays for the return label, and whether the package goes back to a U.S. address or to France? Is it identical on your site, in Merchant Center and on Amazon? If you extend your return window for the holidays, have you said so everywhere?
7. Check 7Prices across channelsSame regular price on your U.S. site, in your feeds, in the cart, on Amazon and at retailers. Look for unexplained gaps, forgotten promotions and currency errors.
8. Check 8Your Black Friday–Cyber Monday promotionStart and end dates, terms and eligible products: identical on your site, in your feed (scheduled sale prices) and on marketplaces.
9. Check 9AI traffic in GA4Set up a segment now for visits coming from chatgpt.com, perplexity.ai, gemini.google.com or copilot.microsoft.com. ChatGPT usually adds utm_source=chatgpt.com to its links. Without that segment, you’ll have nothing to compare against in January. Part of AI’s influence will stay invisible no matter what.
10. Check 10What AI says about youAsk the main assistants yourself. See the method below.
11. Check 11Whether your discounts are realFor every product on sale, compare your Black Friday price with the lowest price charged in recent months, by you and by your retailers (Google Shopping price history, Keepa for Amazon). Your “compare at” price must be one you actually charged, not one raised right before the sale: the FTC regulates fake reference pricing, and an AI can quote an older, lower price.

Check 10: the method

The goal isn’t to find out whether an AI “knows” your brand. It’s to measure two things: are you recommended when a shopper looks for what you sell, and is what’s said about you accurate?

How to do it:

• Write 10 to 15 questions in three groups: buying questions (“which product for this need, this budget”), comparisons (“your brand vs. a competitor”) and factual questions about you (price, shipping, returns).
• Ask them in ChatGPT, Gemini, Perplexity, Copilot and Google’s AI Mode, in a fresh session, logged out of your account.
• For each answer, note whether you’re mentioned, in what position, which competitors appear, which sources are cited, and every inaccuracy (price, delivery time, feature).
• Repeat every week through the end of November: answers vary, and a single one measures nothing.

Then, along the same lines, a comparison (“[your brand] vs. [competitor], which one is worth it?”) and a factual question (“Does [your brand] ship to Texas, and what is its return policy?”).

An error tells you what to fix. A cited source tells you where to fix it: your site, your feed, a retailer listing, an old article. And if you never show up, look at which sources are cited for your competitors before drawing conclusions: that’s often where the gap is.

Check 11: a real case

We saw this firsthand with a brand we work with. Asked about one of its products, an AI assistant cited a retailer and pointed out two things: that retailer’s current price was higher, but it had sold the same product for significantly less in the past, during a discount on its own website. That discount was no longer live. The assistant didn’t present it as today’s price: it reported it as a price charged previously.

AI doesn’t just read today’s price. It can also report what you, or your retailers, charged before. For shoppers, that past price becomes a reference point.

7. In January, we’ll be able to measure what actually happened

Salesforce predicts that AI agents will drive 20% of e-commerce traffic during the 2026 holiday season, with Black Friday and Cyber Monday as its peak moments. It’s a striking number, but it measures visits, not purchases, and it lumps together very different things.

Assistants advising a shopper in real timeIncluded

Agents buying on the customer’s behalf (agentic commerce in the strict sense)Included

Technical agents and automated price-comparison botsIncluded

What the number doesn’t say: one purchase in five made by an AIFalse

Agentic commerce in the strict sense is counted, but it gets lost in a total dominated by visits and by bots that never buy anything.

After Cyber Monday, two kinds of measurement will be possible. Adobe and Salesforce will publish their recaps of the U.S. market, and we’ll break them down in January. And every brand will be able to measure its own results, provided it set up its GA4 segment before the peak (check 9): the number of visits from AI, conversion rate and average order value, compared with the rest of its traffic.

What is AI saying about your products?

If an American shopper asked an AI today to recommend a product like yours, do you know what it would say? We can run the test together, in 20 minutes, on your actual products: your shoppers’ questions, asked to the main assistants, and what they say about you.

Run the 20-minute test

Sources

1. Adobe — 2026 holiday forecast
   Projected 130% increase in AI-referred traffic to U.S. retail sites.
2. L.E.K. — AI discovery and traffic
   Less than 0.5% of total inbound traffic.
3. L.E.K. — consumer survey
   8% of AI users have bought entirely through an agent; April 2026 survey of 2,650 U.S. consumers.
4. Attentive — State of AI in Retail 2026
   Survey of 3,054 U.S. adults, Aug. 12–14, 2026: 62% of AI-assisted shoppers chose a different product or brand, 93% looked for information or proof outside the AI, 60% prefer to finish on the seller’s website or app.
5. Salesforce — 2026 Holiday Predictions
   20% of traffic forecast; includes several categories of agents and bots.
6. OpenAI — Powering product discovery in ChatGPT, March 24, 2026
   Refocus on discovery, with checkout on the merchant’s side.
7. OpenAI — Shopping with ChatGPT Search
   In-chat checkout for some eligible products and merchants.
8. Google — Universal Cart, Oct. 15, 2026
   U.S. rollout; checkout via Google Pay or the merchant’s cart.
9. Microsoft — Copilot Checkout, Jan. 8, 2026
   U.S. rollout with PayPal, Shopify and Stripe.
10. Meta — Introducing Muse, Sept. 8, 2026
    Purchase after the user approves the total.
11. Perplexity — Instant Buy with PayPal
    Launched in 2025; current availability unconfirmed.
12. Amazon — statement on Comet (Perplexity)
13. TechCrunch — Meta’s AI agent has been blocked from using Amazon.com, Sept. 21, 2026
14. Walmart — announcement, Oct. 21, 2026
    Gateway for third-party agents, launched in beta.
15. Google — Booking travel in AI Mode, Aug. 27, 2026
    Hotel booking in AI Mode, in the U.S., in English.
16. Google — Product structured data
    Price, availability, shipping and returns.
17. Google — AI features and your website
    No special file or markup required to appear in AI answers.