From AI Answers to Holiday Orders: What CPG Brands Need to Know

AI search research

Blog Cover Image

How many orders are AI shopping agents actually placing?

The pitch for agentic commerce this holiday goes like this: a shopper says "restock my snacks," an AI shopping agent finds the product, and the order is placed without a single click. Q3 earnings calls just put a number on that pitch. It is small.

Platform

What it reported about third-party agents

For scale

Source

DoorDash

"Several hundred" orders placed via third-party agents

993M total orders last quarter

TechCrunch Oct 30; The Verge Nov 10

Uber

4 rides via third-party agents in Q3; about 500 Uber Eats orders in Halloween week

About 320M trips and deliveries that week

The Verge Nov 10; Forbes Nov 11

Airbnb

Agent bookings "very non-material"

Nearly 500,000 bookings a day

Forbes Nov 11

eBay

Appears to block Meta's Muse; "restricted agent access in certain cases"

Declined to confirm a new block

The Verge Nov 10

Amazon

Blocks outside agents entirely

Runs its own agent, Alexa for Shopping, with 400M+ users

The Verge; Amazon Q3

 

Read that table as a CPG brand and one thing stands out. The agents that are supposed to shop on a customer's behalf are either blocked at the door or barely used. Our Searchpanel team tracked a similar pattern on Nov 9, when one in seven US and UK retailers in a Blend Commerce test refused to serve an AI shopping agent at all.

So is agentic commerce just hype?

No. It is growing fast from a tiny base, and most of it is happening inside walled gardens.

The agents that work today mostly belong to the retailer. Amazon says Alexa for Shopping has more than 400 million users. Walmart keeps adding features to Sparky. And Forter found that 77% of AI-agent orders on its network came from agents driving the store's own checkout pages in the week of Oct 19, up from 1% in late July (Forter, Nov 2).

That last number matters. When an agent does buy, it usually buys through your product page and your checkout, using the same facts a human shopper reads. There is no separate "agent channel" to optimize for. There is your product data, and whether it holds up when a machine reads it.

Where is the sale actually decided this holiday?

In the answer. Shoppers ask ChatGPT, Gemini, Google AI Mode or Alexa for a shortlist, then decide for themselves. The survey data lines up:

  • BCG: 96% of consumers who use AI use it for discovery, 80% for consideration, 24% for purchase (Oct 30).

  • Circana: 34% say AI influences their decisions; only 16% would let an agent buy without approval (Nov 10).

  • Tinuiti: 57% of US shoppers plan to use AI for holiday shopping this year (Nov 5).

  •   Adobe: AI-referred visits to US retail sites converted 90% better than other visits in September, and Adobe forecasts AI traffic on Thanksgiving up 159% (Oct 29).


 

Put those together and the holiday math is simple. Millions of buying decisions will start with an AI answer this season. A few thousand will be finished by a third-party agent. If your product is missing from the answer, the agent question never comes up.

What happens when an AI answer leaves your brand out?

The shortlist is short. In a November test, ChatGPT named no brand at all in 28% of buying answers, and 3.1 brands when it did (Babel42). Three names is the whole shelf.

Agents do not fix this. In Visa's Nov 5 benchmark of 22 AI models, 54% of failed shopping runs still ended in a purchase. The assistant picks something whether or not your facts are there. If the deal-breakers a shopper never says out loud (allergens, pack size, dietary claims) are not on your page, the assistant fills the gap with whoever states them clearly.

The first step is knowing which questions your category gets and who wins them today. Prompt tracking on questions like "best protein bar for runners" or "gentle shampoo for curly hair under $20" shows which brands each engine names, every day, so you are not guessing from one screenshot.

Why do AI answers and AI agents want the same product data?

Circana's analysts put it plainly: many AI trust problems are really price, inventory and product-data problems. That is good news for a small team, because one set of fixes serves both the answer and the agent.

  •  Identity: product name, size, count and flavor in the first 40 words of the page, so a reorder by text (now live in Alexa for Shopping) can identify the item without an image.

  •  Deal-breakers in plain text: allergens, certifications, shelf life and use case, written where a model can quote them.

  • Price and availability in sync: ChatGPT is testing a "Monitoring prices daily" task under shopping answers (Search Engine Watch, Nov 10). A price that differs between your site, your feed and your retailers will get noticed.

  • Machine-readable structure: Product schema, GTINs and complete feed fields. A technical and schema check catches missing or broken fields across the catalog in one pass.

 

 

What should CPG brands do before Black Friday Week?

Amazon's Black Friday Week and Cyber Monday events run Nov 19 to Nov 30, and TikTok Shop's campaign is already live through Nov 30. Here is the order we would work in, starting this week.

  1.   Write 20 buyer questions per hero SKU. Mix category ("best electrolyte powder"), use case ("for long runs"), price cap ("under $25") and dietary needs ("no added sugar").

  2. Run them in ChatGPT, Gemini, Google AI Mode and Alexa for Shopping. Save every answer. That is your baseline.

  3. Log who gets named and who gets cited. If a rival is named off a retailer page or a review site, that page is now part of your shelf. Competitor and citation tracking shows which sources each engine leans on for your category.

  4. Rewrite the first 40 words of each hero product page so they state the product, size, flavor and main claim.

  5. Put deal-breakers in text, not only in images: allergens, count, certifications, shelf life.

  6. Match price and availability across your site, Google Merchant Center feed, Amazon and Walmart before Nov 19.

  7. Check GTINs and feed fields on every SKU that will carry a deal.

  8. Claim your ChatGPT Business Hub profile if you have not. It is free. OpenAI says claiming does not change ranking, but it is the only way to correct the profile and see mention data.

  9.  Re-run the baseline after Cyber Monday (Nov 30) and compare answer by answer.

 

FREE  ·  60 SECONDS

See how AI engines describe one of your products today.

Get a SKU-level read on where AI answers name you, skip you or get your facts wrong. Run a free audit →

 

Tracking is the easy part. The fixes are the work.

Most teams that start checking AI answers end up with the same problem: a long list of things that look wrong and no idea which one to fix first. A missing allergen line on a best seller is worth more than a perfect FAQ on a SKU nobody searches for.

That is why we built the Fix Queue to rank every content gap and technical error by the visibility it would recover against the effort it takes. Each fix opens to the evidence behind it and the exact change to make, then gets an owner and a status, and the score is timestamped against the edit so you can see what the work returned.

When the fix is copy, listing generation drafts the product title, bullets and description around the facts AI engines quote, so the rewrite in step 4 takes an afternoon instead of a sprint.

What we would skip this month

  • Building a custom shopping agent. The numbers above say shoppers are not using third-party agents at scale yet.

  • Paying to "restyle" copy without adding facts. A Politecnico di Milano audit found fact-preserving rewrites moved rankings by about 0.02 positions, while a real brand name moved them 0.44 (arXiv, Nov 5).

  • Judging AI visibility from one prompt on one day. Answers change between runs and between accounts.

 

 

Answer-ready checklist: 8 checks before Nov 19

Tick each one off for your top five SKUs. (Build as the interactive "0 of 8" checklist in Framer.)

☐  20 buyer questions written for each hero SKU

☐  Baseline answers saved from ChatGPT, Gemini, Google AI Mode and Alexa for Shopping

☐  Named and cited competitors logged for each question

☐  First 40 words state product, size, flavor and main claim

☐  Allergens, count and certifications written in text

☐  Price and availability match across site, feed, Amazon and Walmart

☐  GTINs and required feed fields complete on every deal SKU

☐  Re-check scheduled for Dec 1, after Cyber Monday

Want a single number to track across all of this? The AI Visibility Score rolls named, cited and described-correctly into one SKU-level score you can compare week to week.

Frequently asked questions

Are AI shopping agents worth investing in for Holiday 2026?

Not as a separate project. Third-party agents placed very few orders in Q3, and major marketplaces are restricting them. Invest in the product data that both AI answers and agents read: identity, deal-breakers, price and availability.

Why are eBay and Amazon blocking AI shopping agents?

eBay says it has restricted agent access in some cases, including unauthorized scraping and training on its data. Amazon blocks outside agents and runs its own, Alexa for Shopping. Both want to control the shopper relationship and the ad revenue that comes with it.

What is the difference between an AI shopping agent and an AI answer?

An AI answer is the shortlist a shopper reads in ChatGPT, Gemini, Google AI Mode or Alexa. An AI shopping agent goes further and completes the purchase. Today most shoppers stop at the answer and buy themselves.

How do I check whether ChatGPT recommends my product?

Write the questions your buyers ask, run them in ChatGPT and the other major engines, and record which brands get named and cited. Repeat on a fixed schedule, because answers change. A free SKU audit is a quick way to get a first read.

 

 

Find out what AI says about your products, and what to fix first.

Run a free SKU audit at searchpanel.ai/try-audit  or  book a demo

 

Turn AI visibility into

Turn AI visibility into

revenue.

Find out which of your products AI names, which it ignores, and what it would take to change that.

Run a free audit

takes 2 minutes

No card needed · Cancel anytime · Read only, always