AI Shopping Agents by the Numbers: Who Actually Uses Them

Last updated: November 5, 2026

For most of 2026, the story about AI agents has been told in valuations and demos. An assistant books your dinner, reorders your groceries and finds a flight while you sleep. Investors put $10 billion on one of these startups. Meta says its agent has passed 10 million users.

Then Uber counted. In the busiest week of its October, AI agents booked 4 rides and about 500 food orders, out of about 320 million trips and orders. This page puts the real numbers in one place: what the companies have reported, why so few orders come through agents, and our own check of how 50 large US shopping, delivery and travel sites treat AI agents today.

Key Statistics

• AI agents produced 4 Uber rides and about 500 Uber Eats orders in the week of Halloween 2026, Uber's busiest week of October. Uber Q3 2026 earnings call
• Uber and Uber Eats handled about 320 million trips and orders that same week, so agents placed roughly 1 in every 635,000. (Our math from TechCrunch and the call)
• DoorDash said it has seen "maybe several hundred orders in total" from third-party agents, against an average of 3 million deliveries a day. TechCrunch
• Meta's Muse agent has more than 10 million users, and about 25% of people who set it up use it daily, according to Meta. TechCrunch
• Of 42 large US shopping, delivery and travel sites whose robots.txt file we could read, 28 (67%) don't mention a single AI bot. (Our check, November 5, 2026)
• Only 3 of those 42 sites, Amazon, Yelp and Instacart, use that file to close all or nearly all pages to an AI agent that browses for a person. (Our check)

How Many People Use AI Shopping Agents?

Very few, if you count completed orders. Uber's finance chief said agents fulfilled "a grand total of 4 trips" and around 500 Uber Eats orders in Halloween week. DoorDash reported several hundred agent orders in total. Neither company has published a percentage, but both numbers round to zero against their normal volume.

The gap between attention and use is the story. Here is what the companies themselves have said this fall:

Sources: Uber call transcript, TechCrunch on DoorDash, TechCrunch on Muse, TechCrunch on Instinct.

Uber's CEO Dara Khosrowshahi summed it up on the call: "It is very, very early and very limited traction that we are seeing so far." He also said every major AI company is talking to Uber about terms, and that he expects the field to grow from about 8 assistants to more than 10.

What Did Uber Say About AI Agents?

Uber said agents are a small, new way in, not a threat to its app. The company handled 4.1 billion trips in the third quarter of 2026, up 16% from a year earlier, and has 215 million monthly active customers. Against that, 4 agent rides is a rounding error, and Uber said so plainly.

Two details from the call explain the number. First, speed: booking a ride in the Uber app takes about 50 seconds and reordering a favorite meal about 90 seconds, according to TechCrunch's report. An agent has to beat that to be worth asking. Second, Uber is building its own help inside the app. Khosrowshahi said early signals show baskets built with its new cart assistant are roughly twice the size of an average basket.

So Uber's bet is clear. Its own app stays the main door, and outside agents get a side door once the "terms of trade" are agreed.

Why Are So Few Orders Placed by AI Agents?

Four reasons show up across the reporting, and they stack on top of each other.

The apps are already fast. A 50 second ride booking leaves little room for a chatbot to save time, especially when the person still has to check the price and pickup spot.

Many big sites don't let agents in. Amazon began blocking Meta's Muse in September with an error that reads: "Continued access by an unauthorized AI agent violates Amazon's Conditions of Use" TechCrunch. Delta told TechCrunch it has no partnership that lets outside agents book, United's terms bar automated access without written permission, and Yelp only allows non-human traffic that has paid for its data TechCrunch.

Payment is the hard part. Handing an agent a card number is the step people hesitate over, and stores hesitate too. Walmart's new rules for agents say they never receive card details and need the customer's approval by default Walmart.

Deals aren't signed yet. Uber and DoorDash both describe talks with agent makers, not finished integrations. Until a company connects its system directly, an agent has to click through a website built for people, which is slow and breaks easily.

Are AI Shopping Agents the Same as AI Search Traffic?

No, and mixing them up is the most common mistake in coverage this month. AI search traffic means a person asks ChatGPT, Gemini or Perplexity a question, sees a link and clicks it. An AI shopping agent goes further: it visits the store and places the order itself.

The first kind is growing fast. Adobe, which sells analytics software to retailers, reports that AI traffic to US retail sites rose 71% year over year in September 2026, and that those visitors converted 90% better than visitors from other sources. A year earlier, in August 2025, they converted 9% worse. Adobe's figures cover more than 1 trillion visits (Small Business Trends, reporting Adobe's data; Adobe's post).

So the honest summary for November 2026: people increasingly use AI to decide what to buy, then buy it themselves. Very few let the AI finish the job. We saw the same pattern from the publisher side when Reddit blamed AI answers for falling search visits in its login requirement announcement.

Which Big Sites Block AI Agents? (Our Check)

Most big US shopping sites don't address AI agents at all in their public crawler rules. On November 5, 2026, we read the robots.txt file of 50 large US retail, grocery, delivery and travel sites. This is the public file where a site tells automated visitors which pages they may open. We could read 42; the other 8 returned an error to our automated request.

"A shopper's agent" means the four named fetchers that open pages when a person asks an AI tool to: ChatGPT-User, Claude-User, Perplexity-User and Meta-ExternalFetcher.

What stood out:

• 28 of 42 sites name no AI bot at all. That includes Walmart, Target, Costco, Home Depot, DoorDash, Uber Eats and Booking.com. Their files only carry general rules for every robot.
• Amazon is the strictest. It closes the whole site to all four shopper's agents and to the training crawlers we checked from OpenAI, Anthropic, Perplexity and Google.
• Training and shopping are treated differently. Five sites close the whole site to OpenAI's training crawler, GPTBot: Amazon, eBay, Instacart, Yelp and Airbnb. But eBay and Airbnb still let ChatGPT's user-requested fetcher open most pages.
• Travel is the most welcoming. Delta's file opens with a section titled "LLM / AI Crawlers - Explicitly Permitted." Expedia allows the user fetchers almost everywhere, and Kayak opens its flight, hotel and car pages to ChatGPT.

There is a large catch. This file is a request, not a lock. OpenAI says that for ChatGPT-User, "robots.txt rules may not apply" because a person started the visit OpenAI. Perplexity says its user fetcher "generally ignores robots.txt rules" Perplexity, and Meta says its fetcher "may bypass robots.txt rules" Meta. Anthropic says its bots honor the file Anthropic. The real blocking happens elsewhere: bot checks, account rules and error pages like Amazon's. The 8 sites that refused our plain request are a small sign of that.

What Is Walmart Doing Differently?

Walmart is building a front door for agents instead of a wall. On October 21, 2026, it announced Agent Profiles, which give each AI agent its own profile inside a customer's account, so its purchases show up separately. Customers can approve every purchase or set a spending limit; Walmart's examples are $50, $100 and $200 Walmart.

The rules are specific. Agents see the same prices people see, never receive card details, can't add payment methods, and get only the information the task needs. Agents that break the terms get blocked. Profiles were due "in the next month," so they should arrive by late November. A beta of the Walmart Agent Gateway, a direct connection for agent makers, is open to companies and developers.

This is the mirror image of Amazon's approach, and it makes the next few months a real test: if agent orders grow at Walmart and stay flat elsewhere, access was the problem, not demand.

What This Means If You Write About AI Agents

The useful framing for any niche, whether you cover food, travel, retail or personal tech, is "interest is high, completed orders are tiny, and access is the bottleneck."

• Quote counts, not hype. "4 rides in a week" is more convincing than any forecast, and it comes straight from Uber's earnings call.
• Separate the two trends. AI-referred shoppers (growing fast) and AI agents that buy (barely started) are different stories with different numbers.
• Name who lets agents in. Walmart and travel sites like Delta and Expedia are opening doors; Amazon, Yelp and Instacart are closing them.
• Date every number. These figures will change quickly after Walmart's profiles launch and the holiday season ends.

For more numbers on how AI is changing what people read and write online, see our AI writing statistics.

What to Watch Next

• Walmart Agent Profiles launch, expected by late November 2026, and whether Walmart reports any order numbers.
• Holiday results. Adobe and others will publish season totals in early January, which may split out agent purchases for the first time.
• Partnership deals between Uber, DoorDash and the 8 to 10 assistant makers Khosrowshahi described.
• Amazon's next move, since its block is the single biggest closed door in this market.

Methodology

On November 5, 2026, we chose 50 large US retail, grocery and pharmacy, food delivery and restaurant, and travel, rides and ticket sites, and downloaded each site's robots.txt file with a standard browser request. We could read 42 files; Best Buy, Macy's, Adidas, REI, Kroger, Pizza Hut, OpenTable and United returned errors, and we did not try to get around their protection. For 14 AI bots from OpenAI, Anthropic, Perplexity, Google, Meta, Amazon, Apple and Common Crawl, we recorded whether the site names the bot and whether the rules that apply to it close the whole site, some pages, or none. A site counts as closing a page set to a shopper's agent when any of ChatGPT-User, Claude-User, Perplexity-User or Meta-ExternalFetcher is closed from the whole site, or all of it except a handful of pages. This is a one-day snapshot of public rules files, not a test of whether agents can actually shop on each site.

How to Cite

Credit BestWebsiteTemplate.com and link to this page. For the company figures, link to the original source listed next to each number.