Will AI agents make marketing disappear? What the evidence says, claim by claim

What does the Forbes article claim, and what do its sources say?

Every statistic in the article exists. None of them measures agents completing purchases without a person in the loop, which is what the thesis needs. The piece is a Forbes Technology Council post, a paid contributor program, written by the chief technology officer of a moving company, and it cites no primary links.

Two omissions matter more than any single number. OpenAI wound down ChatGPT’s Instant Checkout in March 2026; The Information’s source said users “weren’t using the chatbot to actually help them make purchases” (TechCrunch). And ChatGPT now sells ads, which is where marketing budgets go when a channel works.

A subtler problem: the article calls product-data optimization “rules-based, measurable” and therefore automatable. The evidence in the next sections says it is stochastic. The same prompt returns the same brand list twice in fewer than one run in 100.

Where does AI sit in the buying journey today?

AI is a research tool with a person still deciding. Every dataset that separates research from delegated purchase says so, and the gap is roughly five to one.

Question wording differs across surveys, so read the columns as a range. The ranking is the same everywhere: groceries and replenishment first, personal care and pet supplies next, apparel and electronics last. Bain found actual AI purchases were “mostly small-ticket grocery and household goods.”

The product that tested the thesis failed

ChatGPT Instant Checkout launched in September 2025 with Etsy and Shopify. By March 2026 OpenAI had moved checkout back to merchants’ own sites. Walmart said in-chat conversion ran about three times worse than clicking out to walmart.com (Search Engine Land). Etsy “did not see a large volume of sales”; Instacart saw “only a very small percentage” of orders from agents (Modern Retail). No US platform has disclosed agent-completed purchase volume. The one mass-scale claim is Alipay’s 120 million “AI Pay” transactions in Chinese New Year week (Ant Group), company-reported and undefined.

Traffic is growing fast from a small base

Adobe Analytics has AI-referred retail traffic up 393% year over year in the first quarter of 2026 and converting 42 to 60% better than non-AI traffic by mid-2026 (Digital Commerce 360). Retailers themselves report AI at 2 to 5% of referrals, a few approaching 10% (Digital Commerce 360). The only independent academic study, 973 sites and 50,000 ChatGPT-attributed orders through July 2025, found ChatGPT under 0.2% of sessions and converting worse than Google organic (University of Hamburg). The two findings cover different periods; both agree AI is a low-single-digit share. Similarweb puts it plainly: people use AI to narrow options and search to make the decision. 89% of AI shoppers still use search.

How do AI shopping agents actually choose?

Agents are not the rational, unpersuadable buyers the Forbes piece imagines. In controlled experiments they are more swayed by marketing cues than people are, just by different cues. Three independent labs (Columbia, MIT and Microsoft Research) reached the same conclusions with different setups.

The caveat on all of this: the bias studies are sandboxes with 2025 and 2026 model versions, not live ChatGPT. The directions replicate across labs; the magnitudes change with every model release.

What this means for a company: generic superlatives do nothing. Citations, statistics, named reviewers and badges move agents a lot, whether or not they are true. The practical question is not “is our copy persuasive” but “which third-party sources does the model trust, and are we in them.” That is the same conclusion as our guide to content that LLMs retrieve and cite, reached from the buying side.

How are the AI platforms making money from shopping?

The AI platforms tried a transaction cut, it failed, and they moved to ads. Marketing budgets are flowing into the agent platforms in the shape of retail media: labeled ads beside answers, sponsored units inside product carousels, and now sponsored brand agents.

No platform has disclosed any mechanism where paying changes the organic recommendation, and all state the opposite. The lab evidence cuts the same way: a “Sponsored” tag lowered agent selection from 10% to about 8% in ACES. Consumers say they would trust AI shopping less if results were sponsored (75%, Harris Poll for Quad), and 78% already assume they are.

So the grade splits. “Marketing spend moves to the platforms as ads” is Solid. “Paying buys the pick,” the slotting-fee version, is Conjecture, and the incentives on both sides push against it for now.

The pattern after Instant Checkout’s failure is convergence: discovery in the AI, checkout on the merchant, merchant stays seller of record, no platform fee. Google, Microsoft, Perplexity and now OpenAI all run that model. Walmart’s Daniel Danker calls it “intermediation”: Walmart’s own agent, Sparky, sits inside ChatGPT and Gemini and “decides which products appear and in what order” (Digital Commerce 360). The retailer keeps the cart, the checkout, the loyalty data and the ranking.

Who is blocking AI agents, and why?

Whether an agent may enter a store is being settled by terms of service, agent architecture and alliances, not by hacking law. Amazon against Meta’s Muse is the case that shows it, and it changes the picture from “retailers versus agents” to “platforms fighting over who owns the customer.”

What Amazon versus Muse changes

It is platform against platform. Amazon is at once the largest walled garden and an aggressive agent operator. Buy for Me enters other retailers’ checkouts, fills in the shopper’s name, address and payment details, and enrolled brands without asking. The consent standard Amazon now demands of Muse, advance notice and self-identification, is one it did not apply to itself. Meta has no store, so it needs every store open. Neither side is defending “retail”; both want to be the layer the customer talks to.

The money explains the lines. Amazon’s objection list includes “bypassing personalization.” An outside agent that strips sponsored placements and Amazon’s own ranking threatens an ad business near $68 billion (Forbes). The platform with the most retail media to lose is the one blocking hardest.

The legal line moved from “is it hacking” to “where does the agent run.” The Ninth Circuit protected a browser on the user’s device. Amazon’s amended complaint targets Comet for iOS precisely because it runs in Perplexity’s cloud. Muse runs in a Meta-hosted “Secure VM” per user, which is architecturally closer to Comet for iOS than to the desktop browser the court protected. The precedent that saved Perplexity may not cover Muse. Untested so far; Amazon has declined to say whether it will sue.

Alliances follow Amazon exposure. Walmart, Shopify, Target, Etsy, Stripe and PayPal all gain from any channel that weakens Amazon as the default store. They co-developed Google’s protocol, signed with Muse, and are drafting the identification standard. Retailers that block (Williams-Sonoma, Kohl’s, Delta, United, Yelp) have direct relationships to protect and no Amazon problem to solve. Tapestry sits in between: it sells through Google’s AI Mode with explicit approval but bars agents from its own checkout, citing fraud and inventory hoarding (PYMNTS).

For most companies the real problem is accidental blocking. A September audit of 200 UK and US retailers found 28 refused an ordinary browser agent, every one of them through bot defenses rather than stated policy; only 38 had a working agent checkout endpoint (Blend Commerce). Walmart, which wants agents, blocked one by mistake. Security teams tuned for scrapers are turning away customers. Our own study of 100 law firm websites found the same pattern for AI crawlers: 3 of 100 block in robots.txt, 13 of 78 block at the firewall.

Regulators are moving, slowly. The UK Competition and Markets Authority published guidance in March 2026: existing consumer law applies to agents, and firms are liable for third-party agents acting on their sites. FTC Chair Ferguson said in September that companies “cannot blame AI agents” (PYMNTS). Senator Warner’s AI AGENT Act (S.5051, July 2026) would require platforms to admit authorized agents “on the same terms as a user” and bar favoring affiliated agents. It is pending in committee.

Do AI agents cancel subscriptions and switch brands?

Cancellation friction is worth roughly half the revenue of an average subscription service. Agents that remove it now exist. Nobody has measured the effect yet, and the market has already priced it.

The friction is measured. Einav, Klopack and Mahoney used card-replacement months as a natural experiment across ten services: inattention and switching costs roughly double seller revenue on average, ranging from +14% to over 200% (Stanford SIEPR). A regulatory sweep found about 70% of auto-renewal traders use at least one friction practice (ICPEN). The counterweight: a 1.4-million-reader field experiment found auto-renew defaults reduced sign-ups 35% and total subscribers 23% over 20 months, because sophisticated consumers pre-empt the trap (Stanford GSB). Friction helps this quarter and hurts acquisition.

Agents that cancel are live and early. Muse audits card statements and cancels unwanted subscriptions (CNBC). Rocket Money and Pine AI do the same by phone and email. Success rates are undisclosed. In Muse’s launch week Planet Fitness fell 17%, the New York Times 10%, Booking and TripAdvisor 6%, on the theory that an agent makes switching free (Reuters via The Star). One analyst called it “far-fetched.” No company has attributed churn to agents on an earnings call.

The regulatory backstop is gone for now. The Eighth Circuit vacated the FTC’s click-to-cancel rule in July 2025. The FTC reopened rulemaking in March 2026 but has issued no proposed rule (Crowell). Enforcement continues case by case.

Do agents re-shop every reorder? Not the ones that ship. Instacart’s Clementine reorders “my usuals” from order history (PYMNTS). Alexa reorders the item you bought last time. Walmart’s Danker describes recurring purchases through Sparky as “a more sophisticated version of” subscriptions, designed to lock in repeat buys. And in the lab, agents take the first acceptable offer 60 to 100% of the time. The loyalty-erosion channel that is real is different: a model update that reshuffles recommendations with no change on your side. 81% of retail executives expect generative AI to weaken brand loyalty by 2027 (Deloitte via Retail Dive); that is an expectation, not a measurement.

Grades: friction’s value to revenue, Solid. Agents eroding it today, Conjecture. Agents re-shopping every reorder, Contradicted.

Who owns agent readiness inside a company?

The readiness gap is measured. The ownership gap is inferred from it. No survey asks which function owns agentic commerce.

• Only 37% of commerce and marketing leaders have a plan to retain the customer relationship when agents intermediate, against 78% who call first-party data critical (HFS Research and Cognizant, n=101).
• One in five organizations says it is prepared to redesign processes for agentic operation (Deloitte, n=501).
• AI agents are built by central IT in 47% of enterprises and by a cross-functional team in 16% (Contentstack).
• 89% of merchants say they are “actively preparing” while 3% of transactions involve agents; the report’s own phrase is “nobody wants to own the risk” (Checkout.com via FinTech Weekly).
• Retailer readiness scores 4.4 out of 10 (Mirakl and Merkle). A Fortune panel in June concluded “no one is ready” and liability is “wide open” (Fortune).

New titles are appearing, mostly in payments rather than retail marketing: Citi’s Head of Agentic Commerce (May 2026), Capital One’s Business Director, Agentic Commerce Strategy, Target’s first Chief AI Officer, Morrisons’ Chief AI Officer in October. That the first owners are at card networks and banks, not brands, is itself a signal about where the leverage is thought to sit.

Hidden pricing. The claim that agents punish “contact sales” pricing has no study behind it. What exists: half of B2B software buyers now start research with AI chatbots (G2, n=1,038), and 67% of B2B buyers prefer a rep-free purchase (Gartner via GSPANN, n=646). An agent can only compare what is published, so the mechanism is plausible. Nobody has tested what an agent does when the price is behind a form. Conjecture.

Could purchase delegation grow nonlinearly?

Muse’s first month shows fast adoption of an agent, not fast delegation to it. Downloads have run at a steady 230,000 a day since launch; the share of those users letting it buy anything is unpublished and, on every survey, small. The nonlinear case rests on mechanisms that are real but have not yet shown up in any number.

That is 6.6 million in 29 days, about 230,000 a day, with the pace between any two readings between 150,000 and 270,000. The curve is a launch spike followed by a straight line. Trackers disagree on exact totals, so read the slope, not the level.

• Daily users: 1.8 to 2.0 million, about 30% of downloads (Sensor Tower via Bank of America, Barchart). ChatGPT’s mobile daily users: 664 million. Muse is 0.3% of that.
• Growth rate: downloads grew 55% day over day in the first two weeks against ChatGPT’s 24% over its first ten days. #1 on both US app stores since September 18 and 19.
• Promotion: within ten days Muse had the majority of Meta’s house-ad slots and ranked among the top ten brands by ad spend, yet ads were only 6% of impressions. More than 95% of Muse users already use Facebook.
• Forecasts: Evercore expects 100 million users within 6 to 12 months. Bank of America expects no meaningful revenue before 2028. Meta’s stated model is a small commission on commerce through Walmart, Best Buy, Wayfair and Shopify.
• Not disclosed: tasks completed, purchases made, subscriptions cancelled. Meta reports third-quarter earnings on November 4.

Why delegation could still bend upward

1. Distribution is a step function. Meta reaches 3.6 billion people a day and has not yet put Muse inside WhatsApp or Instagram. An operating-system agent from Apple or Google would be the same kind of step. ChatGPT went from zero to 100 million monthly users in two months with no distribution at all (UBS via Taipei Times).
2. Trust compounds through low-stakes wins. 31% of consumers say one successful low-risk purchase would build comfort (Accenture). Subscription cancellation is the ideal wedge: immediate, measurable, satisfying, no purchase risk. Each win raises the ceiling for the next category.
3. Memory raises switching costs. An agent that knows your sizes, addresses, brands and card gets better with use. The second delegation is cheaper than the first, and moving to a rival agent means starting over.
4. The rails are being removed one at a time. Purchase protection (Amex, Stripe Link), payment tokens (Visa, Mastercard), identification standards (Web Bot Auth, the Meta and Walmart draft), checkout protocols (Google’s UCP, OpenAI’s ACP). Each one removes a reason to say no.
5. The platforms are paid to push it. Meta wants a commission on volume. Google, OpenAI and Microsoft want checkout. Every supply-side incentive points toward more delegation.
6. It has happened once. Alipay reported 120 million AI Pay transactions in one week when the agent sat inside the payment app. Company-reported and undefined, but a step function at scale.

What is holding it back

• Design: Muse asks before every purchase. The product is built for assisted buying, not autonomous buying.
• Trust shocks: a Mac zero-day found by Patrick Wardle that could expose the account token (Ynet); prompt injection succeeding 0.5 to 8.5% of the time; Meta conceding its Secure VM “does not prevent Meta from accessing data when necessary.”
• Access: the largest store is closed to it, and others block it by accident.
• Economics: in-chat checkout converted three times worse at Walmart; no analyst expects commerce revenue before 2028.
• Intent: an 11% ceiling on letting AI decide (Gartner); three-quarters uncomfortable with autonomous payment (Forrester).

If it did bend: what it would mean for people

• The brief becomes the point of control. Whoever is the default agent sets the consideration set. The fight among Amazon, Meta, Google, OpenAI and Apple is over that position. People become the prize rather than the customer.
• Persuasion moves out of sight. Agents are more manipulable than people in the lab. A person who delegates never sees the fabricated “clinically proven” claim that steered the agent. Fraud and manipulation shift to a layer with no human witness, which is why the CMA’s “firms are liable for their agents” line matters.
• Money moves back toward consumers. Average US subscription spend is $1,887 a year (Mastercard and FT Strategies via Yahoo); friction is worth about half the revenue of an average service. A widely used cancellation agent transfers some of that back and forces subscription businesses to compete on value or raise list prices.
• Loyalty becomes stored preference. The brand in your agent’s memory wins repeat purchases by default. The one that is not is invisible. Lab data already shows concentration toward big brands and national chains, with no local store in the top twelve.
• Exclusion is silent. Businesses that block agents, on purpose or by accident, lose customers who never learn they were turned away.
• One point of surveillance and failure. Card statements, email, calendar and shopping history in a single agent, hosted by a company that concedes it can access the data. Liability when the agent errs (Amex covers registered agents only) and recourse when it is wrong are unresolved.
• The upside is time and price. Agents are more price-sensitive than people. In commodity categories, delegation transfers surplus to consumers and makes replenishment invisible.

Grade: the mechanisms for nonlinear delegation are Solid. Evidence that it is happening is Conjecture. What would change the grade: a Muse engagement or commerce metric on November 4; Muse inside WhatsApp or Instagram; an operating-system agent from Apple or Google; the first platform to disclose purchases completed by an agent.

Scorecard: every claim graded

The same 26 rows are in claims.csv with a key source for each.

How was this review done?

1. Starting point: the Forbes Technology Council post of October 13, 2026, read in full, with every statistic and named company listed.
2. Three research passes on October 16, 2026, each by a Claude agent with web search: one on the article’s own sources and the state of adoption, one on the experimental literature about how shopping agents choose, one on platform economics, blocking, retention and readiness. A fourth pass covered Amazon versus Muse and Muse’s adoption figures.
3. Source order: peer-reviewed or preprint experiments first, then primary documents (press releases, court filings, regulator guidance, platform documentation), then independent reporting, then vendor studies, which are labeled as such wherever used.
4. Grading: Solid needs an independent primary source or an experiment replicated by more than one group. Mixed covers evidence that is partial, correlational, single-source or vendor-funded. Conjecture is a plausible mechanism with no measurement. Contradicted means the available measurements point the other way.
5. Pages that refused automated reading (Morgan Stanley, OpenAI, Gartner, CNBC, Forbes, Dealroom) were taken from syndicated copies or secondary reporting and are noted as such in the text. No figure on this page comes from memory.
6. Our own claims were graded too. Six claims we made before checking turned out mixed or conjecture, and two were contradicted by the lab evidence. They are kept in the scorecard.

Limitations

• The behavior studies are sandboxes. ACES, ABxLab and Magentic Marketplace measure agents in mock stores with 2025 and 2026 model versions, not live ChatGPT, Gemini or Rufus. Directions replicate; magnitudes move with every release.
• Most adoption surveys are vendor-funded. The independent ones (Gartner, Forrester, the Hamburg and Frankfurt study, NBER, the arXiv papers) are named as such. Vendor surveys are kept because they are the only data on several questions.
• Muse figures are tracker estimates. Sensor Tower, Apptopia and Appfigures disagree by up to a million downloads on the same day. Meta has published no number.
• Some sources were read through syndicated copies. Listed in the method section. Where a figure comes from a secondary report of a primary page, the text says so.
• This is a review of other people’s evidence, not original data collection. It is current as of October 16, 2026 and will age quickly: Meta’s earnings on November 4, the Amazon decision on suing, and any platform disclosure of agent-completed purchases would change several grades.

FAQ

Will AI agents make marketing departments disappear?

Not on the evidence available in October 2026. The claim depends on AI agents completing purchases without a person in the loop, and that is about 3% of transactions by merchants’ own count. OpenAI shut down ChatGPT’s in-chat checkout in March 2026 because people researched in ChatGPT and bought elsewhere. ChatGPT now sells ads, Google shows sponsored deals inside AI Mode, and Amazon runs sponsored products inside Rufus, so marketing spend is moving into the AI platforms rather than vanishing. What does change is where persuasion is aimed: at the third-party sources the models trust, and at the person writing the brief, rather than at the buyer in the moment of choice.

How much online shopping do AI agents actually do in 2026?

Very little, end to end. Merchants surveyed by Checkout.com in June 2026 said about 3% of their transactions involved an AI agent. Gartner found 11% of US consumers would let AI make the purchase decision even for household supplies; Accenture found about one in ten would let an agent buy on its own; Forrester found three-quarters uncomfortable letting an agent pay. Research use is far higher: 45% of consumers in IBM’s 23-country survey use AI during the buying journey, and 53% of US adults in NMI’s survey researched a purchase with AI in the past month. No US platform has disclosed the number of purchases completed by an agent.

Are AI shopping agents harder to persuade than people?

No. In controlled experiments they are easier to persuade, by different cues. In MIT’s ABxLab study of 17 models, a “Best seller” badge or a buy-one-get-one offer shifted agent choice by 10 to 60 percentage points at matched price and rating, while human participants moved about 10 points on all nudges combined. In Columbia’s ACES study, moving a product from the second row to the first was worth roughly a doubling of price. A fabricated “clinically proven” claim broke an incumbent brand’s 100% recommendation lock in a 2026 study. Agents weight citations, statistics and authoritative-sounding reviewers heavily and rarely check whether they are true.

Sources

All pages were read on October 16, 2026. Sources are also linked where each figure appears above.

Research papers

• Allouah, Besbes, Figueroa, Kanoria and Kumar, “What Is Your AI Agent Buying? Evaluation, Implications and Emerging Questions for Agentic E-Commerce” (ACES), Columbia Business School and Yale, December 2025. arxiv.org/abs/2508.02630
• Cherep, Ma, Xu, Shaked, Maes and Singh, “A Framework for Studying AI Agent Behavior: Evidence from Consumer Choice Experiments” (ABxLab), MIT Media Lab, ICLR 2026. arxiv.org/abs/2509.25609
• Bansal, Hofman, Amershi, Horvitz et al., “Magentic Marketplace,” Microsoft Research and Arizona State University, October 2025. arxiv.org/abs/2510.25779
• Chu and Hou, “Incumbent Advantage,” June 2026 (v3 October 10, 2026). arxiv.org/abs/2606.17443
• Dziemian, Lin, Fu, Zou, Kolter et al., large-scale prompt injection competition, 13 frontier models, March 2026. arxiv.org/abs/2603.15714
• Malthouse, Lee, Yang, Pal and Feng, “Evaluating Brand Retrieval and Ranking in LLM Recommendations,” September 2026. arxiv.org/abs/2609.16304
• Uberti-Bona Marin et al., AI product recommendations across 2,528 real queries, September 2026. arxiv.org/abs/2609.18729
• Khan, Gummadi, Ravichander et al., “In Agents We Trust, but Who Do Agents Trust?”, February 2026. arxiv.org/abs/2602.15456
• Shopping Companion benchmark (failure analysis of GPT-5 shopping runs), March 2026. arxiv.org/pdf/2603.14864
• Aggarwal et al., “GEO: Generative Engine Optimization,” KDD 2024. arxiv.org/abs/2311.09735
• Einav, Klopack and Mahoney, “Selling Subscriptions,” NBER, summarized by Stanford SIEPR. siepr.stanford.edu
• Miller, Sahni and Strulov-Shlain, auto-renew field experiment with 1.4 million newspaper readers, Stanford GSB. gsb.stanford.edu
• Kaiser and Schulze, “ChatGPT and e-commerce,” 973 sites, University of Hamburg and Frankfurt School, November 2025. bwl.uni-hamburg.de
• Wharton 2026 follow-up to ACES (six 2026 models), reported by The Decoder, August 27, 2026. the-decoder.com

Adoption and intent

• IBM Institute for Business Value and NRF, consumer study, January 7, 2026. newsroom.ibm.com
• McKinsey, “The agentic commerce opportunity,” October 17, 2025. mckinsey.com
• Morgan Stanley agentic commerce outlook, reported by Digital Commerce 360, December 9, 2025. digitalcommerce360.com
• Gartner, consumer survey press release, May 27, 2026. gartner.com
• Accenture, “Talk to My AI Agent,” June 2026, via Yahoo Finance. finance.yahoo.com
• Forrester, “Consumers aren’t ready to delegate payments to AI agents,” June 15, 2026. forrester.com
• Bain, “Agentic AI commerce hinges on consumer trust,” July 2025. bain.com
• NMI consumer research, August 26, 2026 (vendor). nmi.com
• Semrush, “AI tools and the modern buyer journey,” March 5, 2026 (vendor). semrush.com
• DOSS consumer survey via Chain Store Age, August 13, 2026 (vendor). chainstoreage.com
• Checkout.com, agentic commerce consumer and merchant research, June 2026 (vendor). checkout.com
• Adobe Analytics AI-referral data via Digital Commerce 360, August 19 and September 16, 2026. digitalcommerce360.com, retailer reports
• Similarweb, State of Ecommerce 2026, September 2026. martechedge.com
• Ant Group, “Alipay AI Payment Exceeds 120 Million Transactions in One Week,” February 13, 2026. businesswire.com
• UBS estimate of ChatGPT reaching 100 million monthly users, February 2023, via Taipei Times. taipeitimes.com

Platforms and protocols

• OpenAI, “Buy it in ChatGPT,” September 29, 2025. openai.com
• TechCrunch, “OpenAI’s plans to make ChatGPT more like Amazon aren’t going so well,” March 24, 2026. techcrunch.com
• Modern Retail, “What went wrong with ChatGPT’s Instant Checkout,” March 27, 2026. modernretail.co
• Search Engine Land, Walmart on ChatGPT checkout conversion, March 2026. searchengineland.com
• PYMNTS, “Shopify merchants to pay 4% fee on sales made through ChatGPT checkout,” January 21, 2026. pymnts.com
• OpenAI, “Our approach to advertising and expanding access,” January 2026, and regional expansion posts. openai.com
• TechRepublic, “ChatGPT Sponsored Agents,” September 2026. techrepublic.com
• Google, “New tech and tools for retailers to succeed in an agentic shopping era” (Universal Commerce Protocol), January 11, 2026. blog.google
• Google Search Central, “AI features and your website.” developers.google.com
• Marketing Brew, Google Marketing Live 2026 ad formats, May 2026. marketingbrew.com
• Microsoft Advertising, “Copilot Checkout and Brand Agents,” January 2026. about.ads.microsoft.com
• Visa, “Visa Introduces Trusted Agent Protocol,” October 14, 2025, and “Visa and Partners Complete Secure AI Transactions,” December 18, 2025. usa.visa.com, corporate.visa.com
• Mastercard, first live agentic transaction in Singapore, March 4, 2026, via The Asian Banker. theasianbanker.com
• American Express, Agentic Commerce Experiences developer kit and Agent Purchase Protection, April 14, 2026, via Nasdaq. nasdaq.com
• Meta, “Introducing Muse: The World’s First Personal AI Agent Built for Everyone,” September 8, 2026. about.fb.com
• Shopify, “AI and organic search are doing different jobs,” August 11, 2026. shopify.com
• Digital Commerce 360, Walmart on agentic commerce, March 19, 2026. digitalcommerce360.com
• Harris Poll for Quad, trust in sponsored AI shopping results. quad.com

The access fight

• TechCrunch, “The next hurdle for AI agents: getting websites to let them in,” October 6, 2026. techcrunch.com
• The Next Web, “Amazon blocks Meta’s Muse and accuses Perplexity of misleading a court,” September 21, 2026. thenextweb.com
• TechTimes, “Amazon Blocks Meta Muse Using Standards It Ignores for Its Own Shopping Agent,” September 23, 2026. techtimes.com
• Forbes, “Amazon’s $68 Billion Reason To Block Meta’s Muse,” September 23, 2026. forbes.com
• Pearl Cohen, “Ninth Circuit Vacates Injunction Against Perplexity’s AI Shopping Agent,” August 2026. pearlcohen.com
• CNBC, “Amazon wins court order to block Perplexity’s AI shopping agent,” March 10, 2026. cnbc.com
• Modern Retail, “Brands are upset that ‘Buy For Me’ is featuring their products on Amazon without permission,” January 6, 2026. modernretail.co
• KQED, “Too many bots in the kitchen: Williams-Sonoma bans AI shoppers,” October 2026. kqed.org
• PYMNTS, “AI agents knock; retailers decide how wide to open the door,” October 5, 2026. pymnts.com
• Blend Commerce, AI shopping audit of 200 retailers, September 2026. blendcommerce.com
• Cloudflare, AI traffic defaults, September 15, 2026, and “Signed Agents,” August 28, 2025. blog.cloudflare.com, signed agents
• Reuters via The Star, “Meta’s Muse rekindles fears over winners and losers,” September 24, 2026. thestar.com.my
• Competition and Markets Authority, “Complying with consumer law when using AI agents,” March 9, 2026. gov.uk
• PYMNTS, “FTC Chair says companies cannot blame AI agents for their actions,” September 2026. pymnts.com
• Davis Wright Tremaine, the AI AGENT Act (S.5051), July 2026. dwt.com
• Crowell, FTC moves to revive click-to-cancel rule, 2026. crowell.com
• ICPEN, Dark Patterns Sweep public report, July 2024. icpen.org (PDF)

Muse adoption and readiness

• TechCrunch, “Meta is putting its muscle behind Muse as the AI app takes off,” September 25, 2026. techcrunch.com
• Forbes, “Meta’s Muse AI Assistant Hits 5 Million Downloads,” September 30, 2026. forbes.com
• Dealroom, “Meta’s Muse hits 6.6M downloads and 1.8M daily users,” October 2026. dealroom.co
• Barchart via Yahoo Finance, “Muse Is Nearing 2 Million Daily Users,” October 9, 2026. finance.yahoo.com
• CNBC, “Meta’s Muse agent is attacking one of the economy’s most profitable weak spots,” September 27, 2026. cnbc.com
• Ynet, Muse downloads and the Mac zero-day, September 2026. ynetnews.com
• Mastercard and FT Strategies subscription spending report via Yahoo, 2026. yahoo.com
• HFS Research and Cognizant, agentic commerce readiness, 2026. cognizant.com
• Deloitte, AI readiness and agentic AI survey, 2026. deloitte.com
• Contentstack, Agentic Enterprise Report 2026. contentstack.com
• Mirakl and Merkle, retailer agentic commerce readiness, 2026. mirakl.com
• Fortune, “AI shopping agents are coming. No one is ready for them,” June 12, 2026. fortune.com
• G2, “Half of B2B software buyers now start their research with AI chatbots,” March 2026. prnewswire.com
• Yelp, 2025 Trust and Safety Report, February 25, 2026. businesswire.com

How to cite: Copy for Originals, “Will AI agents make marketing disappear? What the evidence says, claim by claim,” October 16, 2026, https://tools.copyfororiginals.com/ai-shopping-agents-marketing-evidence/. Data licensed CC BY 4.0.