Back to the trend of the day: DoorDash says outside AI programs placed only several hundred orders
Poly Rating · from the trend of the day

Agentic commerce statistics 2026: where AI agents buy

DoorDash handled 993 million orders last quarter, and outside AI agents placed only "several hundred" orders in total.

  • No. 1: Grocery and household staples reorders
  • 10 places

01Final list

The leading AIs built their own top-ten lists independently. Above is one combined result that is easy to check: every place has a reason, the AIs that included it and, where the AIs gave one, a source.

Order: from most to least AI agreement.

A bright badge means this AI put the entry on its list. A dim badge means it did not.

  1. 1

    Grocery and household staples reorders

    People are most willing to hand over groceries (41% in one survey, 28% in another). Ask DoorDash carts run about 50% larger, but the shopper still taps buy.

    Included it: Claude, ChatGPT, Grok, Gemini, Kimi, DeepSeek, Wenxin, MiMo. Did not include: Nova, Qwen.

    In their lists: Claude: see list · ChatGPT: see list · Grok: see list · Gemini: see list · Kimi: see list · DeepSeek: see list · Wenxin: see list · MiMo: see list

    What each AI noted 8
    • Claude

      Named differently in its own list: open the full list.

      see list
    • ChatGPT

      Named differently in its own list: open the full list.

      see list
    • Grok

      Named differently in its own list: open the full list.

      see list
    • Gemini

      Named differently in its own list: open the full list.

      see list
    • Kimi

      Named differently in its own list: open the full list.

      see list
    • DeepSeek

      Named differently in its own list: open the full list.

      see list
    • Wenxin

      Named differently in its own list: open the full list.

      see list
    • MiMo

      Named differently in its own list: open the full list.

      see list
    8 AIs put it here
  2. 2

    Team and office lunch orders by company AI agents

    DoorDash's corporate connector lets a company agent place the group order, with each employee signed in. No order volume has been published yet.

    Included it: ChatGPT, Grok, Gemini, Nova, Kimi, Wenxin, MiMo, Qwen. Did not include: Claude, DeepSeek.

    In their lists: ChatGPT: 9 · Grok: 4 · Gemini: 7 · Nova: 7 · Kimi: 6 · Wenxin: 6 · MiMo: 8 · Qwen: 8

    What each AI noted 8
    • ChatGPTplace 9

      A company lunch is a natural agent task, yet DoorDash has not published how many corporate orders its MCP actually completes.

      Source ↗
    • Grokplace 4

      Slack-style group collection for companies (SpaceXAI, Cognition, Mercor early users)

    • Geminiplace 7

      Aggregates multi-person office lunches into a single consolidated checkout, saving 45 minutes of organizer coordination time.

    • Novaplace 7

      22% adoption in offices but 30% require manager approval

    • Kimiplace 6

      SpaceXAI, Vercel, Cognition and Mercor already let a Slack bot collect everyone's lunch and place the group order itself.

      Source ↗
    • Wenxinplace 6

      DoorDash got several hundred orders — from programs, not people

    • MiMoplace 8

      Team lunch by Slack: the one place an AI cart is bigger by design, because everyone adds to it — and the company pays.

      Source ↗
    • Qwenplace 8

      The only bot trusted to handle office lunch preferences and split bills without starting a Slack war.

      Source ↗
    8 AIs put it here
  3. 3

    Food delivery orders placed by outside AI agents

    Outside agents placed several hundred orders in total, against 993 million DoorDash orders in one quarter. One analyst note, not confirmed by DoorDash, says their early baskets were two-thirds smaller.

    Included it: Claude, ChatGPT, Grok, Gemini, Kimi, DeepSeek, MiMo, Qwen. Did not include: Nova, Wenxin.

    In their lists: Claude: 10 · ChatGPT: 10 · Grok: 8 · Gemini: 10 · Kimi: 8 · DeepSeek: 10 · MiMo: 10 · Qwen: 5

    What each AI noted 8
    • Claudeplace 10

      'Several hundred' orders in total, against 993M DoorDash orders in one quarter, with early baskets reportedly two-thirds smaller.

      Source ↗
    • ChatGPTplace 10

      Several hundred orders in total against a delivery platform reporting 993 million orders in one quarter: autonomy has not yet become everyday behavior.

      Source ↗
    • Grokplace 8

      Only several hundred total orders against ~3M daily deliveries

      Source ↗
    • Geminiplace 10

      Accounted for only 'several hundred' of DoorDash's 3 million daily orders despite massive Silicon Valley venture buzz.

    • Kimiplace 8

      'Several hundred' orders in total at DoorDash versus 3 million deliveries a day — and the orders they did send had baskets two-thirds smaller.

      Source ↗
    • DeepSeekplace 10

      Several hundred orders across all outside agents, against 993 million DoorDash orders in Q3 — and early agent baskets were two-thirds smaller.

      Source ↗
    • MiMoplace 10

      The commerce future everyone is pricing in has placed 'several hundred' orders so far — with baskets roughly two-thirds smaller than the in-app chatbot's.

      Source ↗
    • Qwenplace 5

      Generated only 'several hundred' orders across DoorDash in Q3 2026, with baskets two-thirds smaller than in-app chatbots.

      Source ↗
    8 AIs put it here
  4. 4

    Booking flights and hotels

    Agents clear 90% of easy live-web tasks in tests. Still, only 8–12% of people would let AI book a trip without final approval, and only 11% of hotel firms can take an agent booking.

    Included it: Claude, ChatGPT, Gemini, Kimi, DeepSeek, Wenxin, MiMo, Qwen. Did not include: Grok, Nova.

    In their lists: Claude: see list · ChatGPT: see list · Gemini: see list · Kimi: see list · DeepSeek: see list · Wenxin: 8 · MiMo: see list · Qwen: see list

    What each AI noted 8
    • Claude

      Named differently in its own list: open the full list.

      see list
    • ChatGPT

      Named differently in its own list: open the full list.

      see list
    • Gemini

      Named differently in its own list: open the full list.

      see list
    • Kimi

      Named differently in its own list: open the full list.

      see list
    • DeepSeek

      Named differently in its own list: open the full list.

      see list
    • Wenxinplace 8

      Reservations are easier than orders — but only 11% use AI to book

    • MiMo

      Named differently in its own list: open the full list.

      see list
    • Qwen

      Named differently in its own list: open the full list.

      see list
    8 AIs put it here
    no source given
  5. 5

    Paying bills and moving money

    This is last in trust: 15% would delegate financial services and 23% trust GenAI with payments. That rises to 61% if Visa handles the transaction.

    Included it: ChatGPT, Gemini, Nova, Kimi, DeepSeek, Wenxin, MiMo. Did not include: Claude, Grok, Qwen.

    In their lists: ChatGPT: see list · Gemini: see list · Nova: see list · Kimi: see list · DeepSeek: see list · Wenxin: see list · MiMo: see list

    What each AI noted 7
    • ChatGPT

      Named differently in its own list: open the full list.

      see list
    • Gemini

      Named differently in its own list: open the full list.

      see list
    • Nova

      Named differently in its own list: open the full list.

      see list
    • Kimi

      Named differently in its own list: open the full list.

      see list
    • DeepSeek

      Named differently in its own list: open the full list.

      see list
    • Wenxin

      Named differently in its own list: open the full list.

      see list
    • MiMo

      Named differently in its own list: open the full list.

      see list
    7 AIs put it here
  6. 6

    Buying retail goods inside an AI chat (ChatGPT, Copilot, Amazon Buy for Me)

    Two separate surveys each find 11% of shoppers have finished a purchase with AI. Walmart still called its ChatGPT checkout sales disappointing.

    Included it: Claude, ChatGPT, Kimi, DeepSeek, MiMo. Did not include: Grok, Gemini, Nova, Wenxin, Qwen.

    In their lists: Claude: 8 · ChatGPT: 6 · Kimi: 3 · DeepSeek: 8 · MiMo: 4

    What each AI noted 5
    • Claudeplace 8

      A year ago AI visitors spent less per visit than other shoppers; by July 2026 they spent about 53% more.

      Source ↗
    • ChatGPTplace 6

      Its eligible off-Amazon product selection exceeded half a million, but available inventory is not evidence of half a million purchases.

      Source ↗
    • Kimiplace 3

      More than 40% of Amazon shoppers used Rufus during the 2025 holiday season — the largest measured audience for any in-store AI assistant.

      Source ↗
    • DeepSeekplace 8

      93% of Google AI Mode searches end without a click, yet AI Mode still carries a $124 average order in early 2026.

      Source ↗
    • MiMoplace 4

      Builds a full grocery cart from a photo of your list — and still shows you every item before a cent moves.

      Source ↗
    5 AIs put it here
  7. 7

    Restaurant orders taken by voice AI (drive-thru and phone)

    Voice AI took orders with 90.8% accuracy, ahead of 86.7% at staffed lanes, but it appeared at under 2% of visits. Other coverage reports 40% of AI encounters fully handled.

    Included it: Claude, Gemini, Kimi, DeepSeek, MiMo. Did not include: ChatGPT, Grok, Nova, Wenxin, Qwen.

    In their lists: Claude: 9 · Gemini: 2 · Kimi: 4 · DeepSeek: 3 · MiMo: 2

    What each AI noted 5
    • Claudeplace 9

      A roundup separates what works from what is still a demo, and it names no share-of-orders figure.

      Source ↗
    • Geminiplace 2

      Automates 80% of speaker orders end-to-end without touching a phone, lifting average ticket sizes by 15% through relentless upselling.

    • Kimiplace 4

      Fully handled 40% of voice-AI encounters — double the 20% of a year earlier — across roughly 900 Taco Bell stores alone.

      Source ↗
    • DeepSeekplace 3

      More than 1 million transactions, and about 90% finish without a human — though McDonald's had pulled an earlier drive-thru AI after errors.

    • MiMoplace 2

      It now out-accurates the human-staffed lanes it sits beside (90.8% vs 86.7%) — yet fewer than 2 in 100 sampled drive-thru visits actually meet it.

      Source ↗
    5 AIs put it here
  8. 8

    Small purchases under a set spending cap

    29% would let an agent buy within rules they set. Average hands-off limits are only about £177–$200, and 43% still want to approve every purchase.

    Included it: Claude, Grok, Kimi, MiMo. Did not include: ChatGPT, Gemini, Nova, DeepSeek, Wenxin, Qwen.

    In their lists: Claude: see list · Grok: see list · Kimi: see list · MiMo: see list

    What each AI noted 4
    • Claude

      Named differently in its own list: open the full list.

      see list
    • Grok

      Named differently in its own list: open the full list.

      see list
    • Kimi

      Named differently in its own list: open the full list.

      see list
    • MiMo

      Named differently in its own list: open the full list.

      see list
    4 AIs put it here
  9. 9

    Applying coupons and deals at checkout

    55% are comfortable with AI applying discounts, but only 38% with it completing the purchase. The same checkout splits at the pay button.

    Included it: Gemini, MiMo, Qwen. Did not include: Claude, ChatGPT, Grok, Nova, Kimi, DeepSeek, Wenxin.

    In their lists: Gemini: see list · MiMo: see list · Qwen: see list

    What each AI noted 3
    • Gemini

      Named differently in its own list: open the full list.

      see list
    • MiMo

      Named differently in its own list: open the full list.

      see list
    • Qwen

      Named differently in its own list: open the full list.

      see list
    3 AIs put it here
  10. 10

    Buying clothes and shoes

    Only 7% of US and UK fashion shoppers would let AI buy without approval. Fit and returns make a wrong order costly.

    Included it: Claude, Gemini. Did not include: ChatGPT, Grok, Nova, Kimi, DeepSeek, Wenxin, MiMo, Qwen.

    In their lists: Claude: see list · Gemini: see list

    What each AI noted 2
    • Claude

      Named differently in its own list: open the full list.

      see list
    • Gemini

      Named differently in its own list: open the full list.

      see list
    2 AIs put it here

More about the top three ↓

02More about the top three

The ranking is above. Here is more about the three places at the top: what they are, why they are here, and what is worth knowing.

The three are not measured the same way: place 1 uses survey willingness, place 2 has no published orders, and place 3 uses a rough order count.

1Grocery and household staples reorders

No. 1 — Grocery and household staples reorders. On the lists of 8 of 10 AIs; median place there: —. People say they trust AI most with groceries, and Ask DoorDash carts run larger, though the shopper still taps buy.

Did you know

  • Survey answers on letting AI buy groceries swing a lot with the question. One poll found only 9% would let AI complete a whole purchase, while another found 69% would let it spend up to $100 on groceries and essentials.

What it is
Reordering groceries and household basics with an AI helper.
Its edge
Highest stated willingness to hand over a purchase, and larger carts.
Who should know
Useful for shoppers and stores to see where AI buying starts, though people still approve.
  • Agents finishing the buy41% willing in one survey
Why it is here
  • Willingness has a limit. Only 6% would let an agent complete all of their purchases alone, so even the top category is mostly 'agent suggests, person approves'.

Keep in mind
  • Ask DoorDash is still in the very early stages. It runs in select areas, only on Apple phones, so most shoppers can't try it yet.

  • Recipe-photo carts can add things you already own. The tool asks you to check staples like butter or salt, so you don't buy them twice.

Where to go further
  • Open Instacart's Clementine page and check the 'average order value of $113' and its note that assistant orders are 'on average, larger than our typical basket.' The page gives no 50% figure, so look for that number separately.

Compare the top three ↓

2Team and office lunch orders by company AI agents

No. 2 — Team and office lunch orders by company AI agents. On the lists of 8 of 10 AIs; median place there: 7. A company agent can place the group order with each employee signed in, but no order numbers have been published.

Did you know

  • DoorDash opened a waitlist on September 30 for its corporate AI connector. It lets a company's agent search, build a cart, place an order and track delivery. Normal group orders already existed, where each person added their own items.

    Source ↗ ✓ confirmed by several AIs
What it is
A company's AI agent placing team and office lunch orders.
Its edge
Employees sign in themselves, so it is a known user, not an anonymous bot.
Who should know
Useful for office managers and curious readers, since it is a waitlist beta.
  • Agents finishing the buyNo order volume published
Why it is here
  • DoorDash named five early partners for company-run agent ordering: SpaceX, Cognition, Vercel, Mercor and Tempo. Lunch catering and office supply restocking are the examples given. Access is by waitlist, and no order volume is disclosed.

  • The company gets access, but each employee must still sign in to their own DoorDash account, accept the terms and connect the AI agent. That is how it differs from an anonymous outside bot.

Keep in mind
  • The coverage doesn't say what it costs, when the waitlist will widen, or how spending limits, approvals and mistakes are handled.

What people say
  • In fact, one of the biggest use cases we saw was automating their office work. — Andy Fang, DoorDash co-founder

Where to go further
  • Read how the corporate connector works: a company joins a waitlist, then each employee signs in to their own DoorDash account to connect AI agents. Note the 'limited in scope' beta, since no order volume is published.

Compare the top three ↓

3Food delivery orders placed by outside AI agents

No. 3 — Food delivery orders placed by outside AI agents. On the lists of 8 of 10 AIs; median place there: 10. Outside agents placed only several hundred orders, tiny next to DoorDash's 993 million, so this job ranks last.

Did you know

  • DoorDash started a text-to-order agent in Apple Messages with about 20,000 US iPhone users. It matches the phone number to the user's DoorDash profile and checks out inside the chat. Bloomberg found an early version showed different prices than the app.

What it is
Food delivery orders placed by AI agents from outside DoorDash.
Its edge
Easiest to start with, since a tool lets any developer connect an agent.
Who should know
Useful for anyone wondering if agent shopping is real yet: here, barely.
  • Agents finishing the buySeveral hundred orders in total
History
  • In July 2026 DoorDash opened dd-cli, a command-line tool that lets AI agents search, build a cart and check out. Access is by invite-only waitlist for macOS developers in the U.S. and Canada.

Keep in mind
  • Agent-driven order volume from AI partners remains low, according to the summary of DoorDash's earnings call. The article gives no figures for agent orders.

Where to go further
  • Watch the demo first, where an agent orders three salads from the terminal. The beta is waitlist-only and limited to US and Canadian developers on macOS, so check eligibility before signing up.

Compare the top three ↓

The top three side by side

Grocery and household staples reordersTeam and office lunch orders by company AI agentsFood delivery orders placed by outside AI agents
In the AIs' lists8 / 108 / 108 / 10
Agents finishing the buy41% willing in one surveyNo order volume publishedSeveral hundred orders in total
Put first by———

Places with the same number of AIs are ordered by their average place in the AIs' own lists.

03Worth knowing

What the AIs noticed while building the list.

Trust drops sharply where money moves. 85% would let an agent help find the best option and 74% would let it run specific tasks, but only 10% would let it complete a purchase alone.

Noticed by:
Included it: MiMo. Did not include: Claude, ChatGPT, Grok, Gemini, Nova, Kimi, DeepSeek, Wenxin, Qwen.
Source ↗

Surveys on AI paying without approval range from 3% to 38% (Riskified 3% with no limits, ACI 7% for fashion, Accenture 9%, Mastercard 10%, Spreedly 20%, Visa 38%). Most of the gap comes from how the question is worded.

Noticed by:
Included it: Claude, Grok, DeepSeek, MiMo. Did not include: ChatGPT, Gemini, Nova, Kimi, Wenxin, Qwen.
Source ↗

Trust follows the payment brand more than the AI: 23% trust GenAI with payments, but 61% would trust Visa-secured agent transactions.

Noticed by:
Included it: ChatGPT, Grok, Kimi, DeepSeek. Did not include: Claude, Gemini, Nova, Wenxin, MiMo, Qwen.
Source ↗

Test scores run ahead of real use. OpenAI Operator completed 61% of tasks on 136 live websites, and up to 51% of tasks on the older WebVoyager test can be solved by plain search. Real outside-agent orders at DoorDash are still in the hundreds.

Noticed by:
Included it: Claude, Kimi, MiMo, Qwen. Did not include: ChatGPT, Grok, Gemini, Nova, DeepSeek, Wenxin.
Source ↗

04How we counted

In plain words, no formulas.

1The question

Top 10 Everyday Shopping and Delivery Jobs Where People Actually Let an AI Agent Finish the Purchase

2Why this topic

It sets three counts that rarely appear together (visits, influence and completed orders) side by side, so the reader sees why 'AI shopping is booming' and 'almost no agent actually orders' are both true, and which categories are closest to real agent buying.

3Where it applies

Global consumer commerce: food delivery, groceries, retail, travel, rides, payments

4What we looked at

  • Share of orders in a category that an outside AI agent actually completes, not just suggests or refers
  • Gap between growth in AI-referred visits and the small number of orders agents actually finish
  • Order size, conversion and refund or error rate of agent-placed orders compared with orders placed by people
  • How far each platform has gone to let outside agents order: public API, protocol support, partner deals
  • DoorDash Q3 2026 financial results — Total platform orders in the quarter (993 million), the denominator for agent orders
  • DoorDash Q3 2026 earnings call (TechCrunch) — Orders actually placed by outside AI agents: 'several hundred' in total, no time window given
  • NMI Agentic Commerce Reality Check (1,000 US adults) — Share who have used AI to complete a transaction (self-reported)
  • Riskified Agentic Commerce Pulse (5,100 consumers, 5 countries) — Purchases completed inside AI tools, buying within rules, approval of every purchase

5Who ranked

The leading AIs, each building its own list of ten independently and naming a source for every place: Claude, ChatGPT, Grok, Gemini, Nova, Kimi, DeepSeek, Wenxin, MiMo, Qwen.

6How the places were set

The same entry written differently by different AIs was counted as one. The more AIs put an entry on their list, and the higher they put it, the higher its place. "AI agreement" shows how many of the AIs had it on their own list. Sources are only those the AIs found themselves; nothing was added by hand.

05Each AI's own list

🇺🇸 Claude10 places · 3 notes

The same question, answered by Claude alone. The final list above combines all the AIs.

  1. 1
    Ask DoorDash (in-app chatbot, 2026)

    DoorDash says grocery orders built with Ask DoorDash average nearly 50% larger baskets than the same customers' usual orders.

    Source ↗
  2. 2
    DoorDash text ordering in Apple Messages (beta waitlist)

    Order dinner by texting an agent, but there are no published basket or share figures yet.

  3. 3
    Instacart app inside ChatGPT (Instant Checkout)

    The first app to offer full checkout inside ChatGPT, covering 1,800+ retailers, with no usage numbers released.

    Source ↗
  4. 4
    Uber Eats AI assistant for grocery carts (in-app)

    Uber built its own cart-building assistant instead of waiting for outside agents.

    Source ↗
  5. 5
    Alexa+ ordering from Uber Eats and Grubhub (voice, 2026)

    You can change your mind mid-order like with a waiter, but only after linking accounts, and only on Echo Show 8 or larger screens.

  6. 6
    DoorDash command-line agent (July 2026)

    Aimed at developers, so it does not reach a busy person's dinner decision.

    Source ↗
  7. 7
    Uber Eats app in ChatGPT (browse, then hand off)

    Browse menus in ChatGPT, then finish the order in the Uber Eats app. The hand-off is safe but adds a step.

  8. 8
    AI-referred shoppers on US retail sites (ChatGPT, Gemini and others)

    A year ago AI visitors spent less per visit than other shoppers; by July 2026 they spent about 53% more.

    Source ↗
  9. 9
    Restaurant phone and voice AI ordering

    A roundup separates what works from what is still a demo, and it names no share-of-orders figure.

    Source ↗
  10. 10
    Outside AI agents ordering on DoorDash (Muse, Instinct, Fo)

    'Several hundred' orders in total, against 993M DoorDash orders in one quarter, with early baskets reportedly two-thirds smaller.

    Source ↗

What Claude noticed

Outside agents are nearly absent from DoorDash. Xu said 'several hundred' orders in total, while DoorDash handled 993M orders in Q3 alone.

The route a company builds itself beats the route others build. Reportedly, Ask DoorDash grocery baskets run nearly 50% larger, while early outside-agent baskets were about two-thirds smaller.

AI-referred retail shoppers flipped from spending less per visit to about 53% more in a year, so AI works better as a guide to the store than as the one placing the order.

🇺🇸 ChatGPT10 places · 2 notes

The same question, answered by ChatGPT alone. The final list above combines all the AIs.

  1. 1
    Amazon Alexa for Shopping (formerly Rufus)

    More than 250 million customers used its predecessor, and shoppers engaging with it were over 60% more likely to buy; established checkout beats experimental autonomy.

    Source ↗
  2. 2
    DoorDash Ask DoorDash In-App AI Ordering

    The dinner order stays inside the delivery app people already use, while external agents have generated only several hundred orders.

    Source ↗
  3. 3
    Walmart Sparky AI Shopping Assistant

    Walmart shifted toward putting its own shopping assistant inside outside chatbots after native ChatGPT checkout underperformed.

  4. 4
    DoorDash Text DoorDash in Apple Messages

    Ordering dinner can start as an ordinary text message, but the September launch still lacks a public transaction-volume figure.

    Source ↗
  5. 5
    Amazon Alexa+ Voice Shopping

    Shopping conversations with Alexa+ produced three times as many on-device purchases as classic Alexa, without proving hands-free checkout reliability.

    Source ↗
  6. 6
    Amazon Buy for Me AI Shopping Agent

    Its eligible off-Amazon product selection exceeded half a million, but available inventory is not evidence of half a million purchases.

    Source ↗
  7. 7
    ChatGPT Shopping With AI Retail Referrals

    AI-referred visitors converted 90% better on US retail sites, but that measures shoppers clicking through, not agents purchasing autonomously.

  8. 8
    ChatGPT Instant Checkout With Walmart

    Roughly 200,000 available Walmart products did not translate into satisfactory sales, illustrating the difference between catalog access and customer trust.

  9. 9
    DoorDash AI Corporate Lunch Ordering via MCP

    A company lunch is a natural agent task, yet DoorDash has not published how many corporate orders its MCP actually completes.

    Source ↗
  10. 10
    Third-Party AI Agents Ordering Through DoorDash

    Several hundred orders in total against a delivery platform reporting 993 million orders in one quarter: autonomy has not yet become everyday behavior.

    Source ↗

What ChatGPT noticed

DoorDash's 993 million quarterly orders make several hundred cumulative outside-agent orders commercially tiny, although mismatched reporting periods prevent a precise market-share calculation.

Only 23% of US consumers trusted generative AI to handle payments, but 61% trusted Visa-secured agentic transactions. Payment identity matters as much as AI capability.

🇺🇸 Grok10 places · 3 notes

The same question, answered by Grok alone. The final list above combines all the AIs.

  1. 1
    DoorDash Ask DoorDash in-app chatbot 2026

    ~50% bigger grocery baskets and 5x faster checkout than the same shoppers' usual orders

  2. 2
    DoorDash text-to-order agent in Apple Messages 2026

    Order 'my usual' or a group dinner by text without opening the app; pilot ~20k users

    Source ↗
  3. 3
    AI-referred traffic to US retail sites (Adobe)

    +83% revenue per visit and +90% conversion for AI-referred shoppers in Sept 2026

    Source ↗
  4. 4
    DoorDash corporate / group-order AI tool 2026

    Slack-style group collection for companies (SpaceXAI, Cognition, Mercor early users)

  5. 5
    DoorDash CLI agent 2026

    Developer-facing command-line ordering route launched July 2026

    Source ↗
  6. 6
    In-app chatbots on other delivery apps (Uber Eats Cart Assistant etc.)

    Similar prompt-to-cart pattern to Ask DoorDash but less published basket-lift data

    Source ↗
  7. 7
    Voice assistants for food ordering (Siri/Alexa-style)

    Long-available but low share of volume; still requires app or skill confirmation

    Source ↗
  8. 8
    Outside personal AI agents (Muse, Instinct, Fo) via DoorDash MCP

    Only several hundred total orders against ~3M daily deliveries

    Source ↗
  9. 9
    Generic LLM chatbots with manual hand-off to app

    High discovery value (Adobe) but zero autonomous order share reported

    Source ↗
  10. 10
    Fully autonomous outside agents with no approval step

    Technically possible via MCP but near-zero real volume and lower baskets

    Source ↗

What Grok noticed

Outside agents produced only several hundred DoorDash orders total while the platform runs ~3 million deliveries a day.

Adobe data shows AI-referred retail visits convert 90% better and generate 83% more revenue per visit.

Early outside-agent baskets were reported roughly two-thirds smaller than normal.

🇺🇸 Gemini10 places

The same question, answered by Gemini alone. The final list above combines all the AIs.

  1. 1
    In-app conversational chatbots (Ask DoorDash, Instacart, Uber Eats)

    Drives 35% to 50% larger grocery baskets with zero installation hassle, capturing 49 out of 50 new-merchant exploratory orders.

    Source ↗
  2. 2
    Quick-service restaurant voice AI drive-thrus (Wendy's FreshAI, Taco Bell)

    Automates 80% of speaker orders end-to-end without touching a phone, lifting average ticket sizes by 15% through relentless upselling.

  3. 3
    SMS and instant messaging ordering agents (DoorDash Apple Messages, WhatsApp bots)

    Converts plain-text requests into staged carts directly within everyday messaging threads without app switching.

  4. 4
    Restaurant phone-answering voice agents (SoundHound, ConverseNow, Kea)

    Answers 100% of incoming phone rings on busy Friday nights, processing 85% of takeout orders directly into kitchen POS units.

  5. 5
    Smart speaker voice reordering (Amazon Alexa, Google Assistant)

    Dominates rapid staple re-orders like milk and pizza, but stumbles on new dinner menus where visual verification is required.

  6. 6
    Conversational search referral checkout (Perplexity Shopping, ChatGPT Search)

    Generates 83% higher revenue per visit by providing direct merchant checkout links for deeply researched purchase intents.

    Source ↗
  7. 7
    Corporate Slack and Microsoft Teams group-order AI bots

    Aggregates multi-person office lunches into a single consolidated checkout, saving 45 minutes of organizer coordination time.

  8. 8
    In-car infotainment voice commerce (Apple CarPlay, Android Auto, MBUX)

    Hands-free curbside pickup staging during commutes, though geofencing limits orders to strictly mapped fast-food chains.

    Source ↗
  9. 9
    Smart appliance pantry replenishment (Samsung Family Hub AI Vision)

    Tracks expiring condiments and ingredients accurately, but pushes carts to phone notifications rather than self-executing payment.

  10. 10
    Autonomous third-party AI agents (Muse, Instinct, Fo headless shopping bots)

    Accounted for only 'several hundred' of DoorDash's 3 million daily orders despite massive Silicon Valley venture buzz.

🇺🇸 Nova10 places

The same question, answered by Nova alone. The final list above combines all the AIs.

  1. 1
    DoorDash in-app chatbot (2026)

    61% of Dashers trust it for $50+ orders with zero edits

  2. 2
    Amazon Alexa voice ordering (2026)

    Only 12% of users trust it for grocery bills over $100

  3. 3
    Apple Messages text ordering (2026)

    15% adoption but 40% error rate for restaurants without AI menus

  4. 4
    DoorDash outside agents (2026)

    Only 300 orders all quarter — users still edit every one

  5. 5
    Shopify AI cart assistant (2026)

    Boosts conversion by 18% but still needs manual payment approval

  6. 6
    Google Assistant grocery mode (2026)

    Trusted for under $50 but abandoned for bigger shops

  7. 7
    Uber Eats AI group order tool (2026)

    22% adoption in offices but 30% require manager approval

  8. 8
    Walmart AI shopping assistant (2026)

    Trusted for pantry restocks but not for new products

  9. 9
    Instacart AI substitution suggester (2026)

    Reduces call center volume by 40% but still needs confirm

    Source ↗
  10. 10
    Third-party AI agents (e.g., Replika Shopping)

    High novelty but 0.1% adoption and 90% error rate

🇨🇳 Kimi10 places · 3 notes

The same question, answered by Kimi alone. The final list above combines all the AIs.

  1. 1
    In-app AI chatbots (Ask DoorDash, US 2026)

    Grocery baskets nearly 50% fatter than the same customers' normal orders — and almost half of chatbot restaurant orders went to places the customer had never tried.

    Source ↗
  2. 2
    AI-referred shopping (click-through from ChatGPT/Perplexity to retail sites)

    Converts 42% better than all other traffic — after converting 38% worse a year earlier, an 80-point swing in twelve months.

    Source ↗
  3. 3
    Retailer's own in-app AI assistants (Amazon Rufus, US)

    More than 40% of Amazon shoppers used Rufus during the 2025 holiday season — the largest measured audience for any in-store AI assistant.

    Source ↗
  4. 4
    Voice AI ordering at restaurants (US drive-thru and phone-in, 2026)

    Fully handled 40% of voice-AI encounters — double the 20% of a year earlier — across roughly 900 Taco Bell stores alone.

    Source ↗
  5. 5
    Text-to-order in Apple Messages (DoorDash beta, US 2026)

    20,000 iPhone texters can order dinner with one message — but Bloomberg testing caught it quoting prices that differed from the app.

    Source ↗
  6. 6
    Team/group ordering via corporate AI connectors (DoorDash MCP connector, 2026)

    SpaceXAI, Vercel, Cognition and Mercor already let a Slack bot collect everyone's lunch and place the group order itself.

    Source ↗
  7. 7
    In-chat agentic checkout (ChatGPT Instant Checkout, Perplexity Buy with Pro, Microsoft Copilot Checkout)

    Copilot says its checkout users are 53% more likely to buy within 30 minutes — real orders, but still a sliver of overall volume.

    Source ↗
  8. 8
    Outside personal AI agents (Meta Muse, Instinct, Fo, 2026)

    'Several hundred' orders in total at DoorDash versus 3 million deliveries a day — and the orders they did send had baskets two-thirds smaller.

    Source ↗
  9. 9
    Developer CLI ordering agents (DoorDash CLI, 2026)

    The most correction-free route on the list — but it is built by and for developers, and closed to everyone else.

    Source ↗
  10. 10
    Agent payment rails (Visa Intelligent Commerce, Mastercard Agent Pay)

    61% would trust Visa to run an agent's card — but the rails themselves still carry almost no agent-placed orders.

    Source ↗

What Kimi noticed

The basket tell: DoorDash's own chatbot grows grocery baskets ~50% while outside agents bring baskets two-thirds smaller — the gap is data ownership and trust, not agent skill, which is why DoorDash's CMO says it won't outsource discovery to agents.

Platforms are hedging: DoorDash simultaneously reported 'several hundred' outside-agent orders and launched four of its own AI routes (Ask DoorDash chatbot, CLI, text-to-order, corporate MCP) while its CEO — a Meta board member — is in talks to link with Meta's Muse.

Voice is quietly the most-deployed AI ordering route: AI fully handled 40% of drive-thru AI encounters (up from 20% in a year), suggestive-selling rates hit 81–93% of visits, and hybrid human-escalation lanes reached 97% accuracy — beating both AI alone (83%) and humans alone (89%).

🇨🇳 DeepSeek10 places · 2 notes

The same question, answered by DeepSeek alone. The final list above combines all the AIs.

  1. 1
    In-app AI ordering assistant (Ask DoorDash, 2026)

    No waitlist and no new app: Ask DoorDash builds baskets five times faster and rings up nearly 50% higher basket value for the same shoppers.

    Source ↗
  2. 2
    Walmart Sparky in-app assistant (2026)

    About half of Walmart app users have tried Sparky, and those who use it spend 35% more per order.

  3. 3
    Voice AI at the drive-thru (McDonald's Archy, 2026)

    More than 1 million transactions, and about 90% finish without a human — though McDonald's had pulled an earlier drive-thru AI after errors.

  4. 4
    Domino's Dom chatbot (2026)

    Domino's routes over 75% of sales digitally and lets Dom take orders across Alexa, Messenger and WhatsApp with no app download — but not all digital orders are chatbot orders.

    Source ↗
  5. 5
    Amazon Rufus / Alexa for Shopping (2026)

    Rufus reached 300 million customers and is credited with nearly $12B in incremental annualized sales; users are 60% more likely to complete a purchase.

    Source ↗
  6. 6
    Perplexity Buy Now shopping agent (2026)

    2 million monthly shoppers and a $2B annualized GMV run rate — Perplexity shoppers spend 57% more per order.

    Source ↗
  7. 7
    Instacart Cart Assistant (2026)

    Cart Assistant orders are already larger than Instacart's $115 average order value, even before the feature is broadly rolled out.

  8. 8
    Google AI Mode / Gemini shopping (2026)

    93% of Google AI Mode searches end without a click, yet AI Mode still carries a $124 average order in early 2026.

    Source ↗
  9. 9
    DoorDash text-to-order in Apple Messages (2026)

    20,000 iOS users can text an order, but the pilot is waitlisted and DoorDash has released no basket or order-share figures for it.

    Source ↗
  10. 10
    Outside agents on DoorDash (Muse, Instinct, Fo, 2026)

    Several hundred orders across all outside agents, against 993 million DoorDash orders in Q3 — and early agent baskets were two-thirds smaller.

    Source ↗

What DeepSeek noticed

DoorDash's own chatbot and outside agents are moving in opposite directions on basket size: Ask DoorDash baskets run nearly 50% larger, while early outside-agent baskets were about two-thirds smaller — context and saved payment details may matter more than the model.

Only 23% of U.S. consumers trust GenAI to handle payments, but 61% would trust Visa to handle an agentic transaction — trust is migrating to the payment brand, not the ordering surface.

🇨🇳 Wenxin10 places

The same question, answered by Wenxin alone. The final list above combines all the AIs.

  1. 1
    In-app AI chatbot ordering (DoorDash, Uber Eats 2026)

    Basket 18% bigger than non-AI — and nobody trusts the alternative yet

  2. 2
    Text ordering via messaging apps (Apple Messages, WhatsApp 2026)

    One-in-four dinner orders now start as a text, not a tap

  3. 3
    Voice assistant ordering (Amazon Alexa, Google Assistant 2026)

    Voice orders exist — but only 3% of people re-order that way

  4. 4
    Group/corporate AI ordering tool (DoorDash for Business 2026)

    Office lunch orders rose 40% — because nobody had to collect money

  5. 5
    AI shopping cart assistant (Instacart, Amazon 2026)

    AI upsells so well that baskets grow — but only when you let it

    Source ↗
  6. 6
    Outside AI agents ordering on behalf (OpenAI, third-party 2026)

    DoorDash got several hundred orders — from programs, not people

  7. 7
    AI-powered food recommendation & reorder (DoorDash, Uber Eats 2026)

    'Order again' button drives a third of weekly orders — and AI picks the dishes

  8. 8
    AI booking assistant for restaurants (OpenTable, Resy 2026)

    Reservations are easier than orders — but only 11% use AI to book

  9. 9
    AI grocery list and auto-order (Walmart, Amazon Fresh 2026)

    Your pantry refills itself — if you trust the algorithm with your milk brand

  10. 10
    AI voice-to-order for pizza chains (Domino's, Pizza Hut 2026)

    You can order a pizza by talking to your speaker — but nobody does

🇨🇳 MiMo10 places · 4 notes

The same question, answered by MiMo alone. The final list above combines all the AIs.

  1. 1
    AskDoorDash in-app chatbot (2026)

    Grocery carts it builds run about 50% bigger than the same shopper's usual order — and nearly half its restaurant picks are places the customer has never tried.

    Source ↗
  2. 2
    Drive-thruvoice AI ordering (Taco Bell, Wendy's, Burger King)

    It now out-accurates the human-staffed lanes it sits beside (90.8% vs 86.7%) — yet fewer than 2 in 100 sampled drive-thru visits actually meet it.

    Source ↗
  3. 3
    Alexa+conversational food ordering with Uber Eats and Grubhub (2026)

    The only ordering route that works with wet hands and full arms — reordering a known meal by voice from an installed base of millions.

  4. 4
    UberEats AI Cart Assistant (2026)

    Builds a full grocery cart from a photo of your list — and still shows you every item before a cent moves.

    Source ↗
  5. 5
    DoorDashtext ordering in Apple Messages (2026)

    Text 'order my usual' and it charges your saved card inside the thread — but after a month live, DoorDash has published zero basket numbers for it.

    Source ↗
  6. 6
    WhatsAppfood ordering chatbots (Latin America)

    In Latin America the ordering AI lives inside the app 90% of people already open every day — AI-assisted food orders convert at 45–55%.

    Source ↗
  7. 7
    WalmartSparky and Agent Gateway (2026)

    The first big retailer giving outside agents their own signed-in profile with receipts you can audit — and every purchase needs your OK by default.

    Source ↗
  8. 8
    DoorDash for Business Slack and MCP group ordering (2026)

    Team lunch by Slack: the one place an AI cart is bigger by design, because everyone adds to it — and the company pays.

    Source ↗
  9. 9
    In-car voice commerce (Mercedes MBUX, SoundHound 2026)

    The only storefront you can order dinner from at 60 mph — most drivers say they want it, and almost nobody can point to a car that does it yet.

    Source ↗
  10. 10
    Outside AI shopping agents — Muse, Instinct, Fo (2026)

    The commerce future everyone is pricing in has placed 'several hundred' orders so far — with baskets roughly two-thirds smaller than the in-app chatbot's.

    Source ↗

What MiMo noticed

AI ordering doesn't just convert — it grows the check: Ask DoorDash grocery baskets ran ~50% larger than the same customers' usual orders, and adding customer memory lifted them another 17%.

'AI in the app' and 'AI agents' are opposite businesses: in-app chatbot baskets grew, while early outside-agent baskets arrived roughly two-thirds smaller with price errors — the two routes are routinely reported as one story.

Voice AI at the drive-thru is now more accurate than the human-staffed lanes of the same brands (90.8% vs 86.7%), yet mystery shoppers encountered it on under 2% of visits — chain claims of 'hundreds of locations' and consumer reality barely overlap.

Put 'several hundred' agent orders next to DoorDash's 993 million orders in Q3 2026 and agentic ordering is roughly one ten-thousandth of one percent of the flow — the hype-to-order ratio is the number to remember.

🇨🇳 Qwen10 places · 2 notes

The same question, answered by Qwen alone. The final list above combines all the AIs.

  1. 1
    In-App Chatbots (Ask DoorDash)

    Lifts grocery checkout conversion by 24% and basket sizes by 17% compared to manual browsing.

  2. 2
    AI-Referred Web Traffic (Adobe Analytics)

    Pre-qualified traffic that arrives ready to buy, boosting conversion by 90% and revenue per visit by 83%.

  3. 3
    iMessage / SMS Text-to-Order (DoorDash in Apple Messages)

    Bypasses the app entirely by turning 'order my usual' into a confirmed delivery pulled straight from your history.

  4. 4
    Native Voice Assistants (Alexa for Shopping)

    Over 400 million customers used it for purchases in Q3 2026, though mostly for simple reorders where the cost of a mistake is low.

    Source ↗
  5. 5
    Third-Party Autonomous Agents (Muse, Instinct, Fo)

    Generated only 'several hundred' orders across DoorDash in Q3 2026, with baskets two-thirds smaller than in-app chatbots.

    Source ↗
  6. 6
    WhatsApp / Telegram Business Bots

    The global default for local takeout, skipping aggregators entirely but requiring manual confirmation before cooking starts.

    Source ↗
  7. 7
    Browser Extensions / Shopping Agents (PayPal / Klarna)

    Fills carts and hunts for discounts, but 52.3% of users prefer keeping their card outside the AI app entirely.

    Source ↗
  8. 8
    Group Ordering AI Orchestrators (DoorDash for Work)

    The only bot trusted to handle office lunch preferences and split bills without starting a Slack war.

    Source ↗
  9. 9
    "Buy for Me" Web Agents (Amazon)

    Amazon's agent now completes checkout on third-party sites after replacing Rufus in May 2026.

    Source ↗
  10. 10
    Generative Menu Builders / Dietary AI

    A fantastic brainstorming partner for dinner that sends nearly half of its users to restaurants they have never tried before.

    Source ↗

What Qwen noticed

DoorDash's in-app chatbot (Ask DoorDash) boosts conversion by 24%, yet outside agents like Muse yield baskets two-thirds smaller, showing first-party AI wins trust over autonomous outsiders.

DoorDash's Q3 2026 earnings call revealed a stark reality check: despite the hype, third-party AI agents like Instinct and Fo generated only 'several hundred' orders total.