On Friday, Wall Street Journal reporter Patrick Coffee explored retailers’ divergent approaches to AI shopping agents, with consumer demand still uncertain.
Echoing elements of the debacle between Amazon and Perplexity or Amazon and Muse over agentic shopping, Coffee previewed his piece on LinkedIn:
- Restricted access: “eBay blocks any agents that don’t have explicit permission to browse or shop its site, because the company doesn’t trust all AI to navigate the nuances of buying handmade and pre-owned items.”
- Browse-only access: “Kate Spade and Coach let agents browse but block them from shopping due to concerns about bot-driven fraud. Parent company Tapestry, however, is designing its framework to optimize for agents whenever that strategy changes.”
- Product feed strategy: “QVC is basically rolling out the red carpet for AI, and it also proactively sends its catalogs to OpenAI, Google, Microsoft etc. to make sure their agents know all about its products.”
Read more. (Oct. 2)
Emarketer analyst Sky Canaves offered a reminder about consumer control: “Consumers still like to make their decisions themselves and be in charge, especially where money is concerned.”
Read: “AI Agents Aim to Change Shopping. Some Retailers Are Locking the Doors.” (Oct. 2) – The Wall Street Journal (subscription)
From tipsheet: Another consideration: retailers also have a high-margin advertising business to protect. Retailers including Amazon sell access to shoppers’ attention, adding profit to the retail business. If agents take over product selection, purchases could continue while that advertising profit shrinks. But consumers may delegate the transaction without giving up discovery or the pleasure of shopping.
LLMs & CHATBOTS
Could Anthropic enter ad tech or martech?
On the Marketecture podcast, hosts Ari Paparo and Aperiam Ventures’ Eric Franchi discuss Anthropic’s reported confidential IPO filing and what the frontier labs’ expansion could mean for ad tech and martech.
Paparo calls Anthropic’s investment a “huge bet” and asks whether the frontier labs will also build applications for advertising and marketing technology (lightly edited):
[27:01]
“I think what’s interesting here is this question about whether these giant labs are going to be platforms and/or have vertical-specific applications. That’s the classic computing problem…
And so with Anthropic, their main product is Claude Code. They’ve also talked about legal and other verticals. Will there be a Claude martech or ad tech offering at some point? — the CMO-replacement-in-a-box that many startups have been investing in — which Claude would be in an interesting position for as well as OpenAI. OpenAI seems more likely because they’re in the advertising business.
So that’s what to really watch here for us as a community. Will someone try to dive in here?”
Anthropic has also advertised martech-related roles for its own internal marketing capabilities over the past two months, including these positions.
Listen: “Episode 193: Adam Epstein on Muse & Agentic Commerce, Plus TTD Blocked on Safari and Amazon UnBoxed” (Oct. 2) – Marketecture Podcast
From tipsheet: We noted last month that OpenAI’s “Head of Marketing Technology” listing set an ambition to build “the most AI-enabled marketing operations organization in the world.” These internal investments don’t establish plans for commercial products, but they make the labs’ own marketing operations worth watching.
What prevents the frontier labs from owning verticals? For ad tech and martech companies, the answer has to be more than access to a model. Proprietary data, customer integrations and feedback from running campaigns could provide an advantage. The question is whether those advantages endure as the labs expand.
DATA
Apple tightens access to ad data, devices
Apple’s restrictions on data access came into sharper focus Friday, following Tuesday’s report by AdExchanger’s James Hercher on ad tech restrictions in Safari.
Friday’s follow-up suggests the restrictions could extend to hundreds of programmatic companies.
Read: “Apple Has Far-Reaching Plans To Block Hundreds Of Programmatic Data Companies From iOS” (Oct. 2) – James Hercher on AdExchanger
Reaction
W Media Research analyst Karsten Weide commented about Friday’s news on X:
“If true, it would be a massive assault on the open Internet. And, if Apple did block all three-letter vendors but not Google, it would be unfair, too.
This wouldn’t just be bad for vendors, it would also be bad for buyers (less competition, worse advertiser-facing products, higher prices, worse conditions), publishers (less competition for their inventory, lower prices), and users (worse content, less content diversity, more bias).”
Read on X. (Oct. 2)
Commenting on Apple’s moves last week which did not include Google’s Doubleclick ad server, Chalice CEO and co-founder Adam Heimlich said on the AdTech AdTalk podcast:
[14:26]
“There’s the cumulative effect of browser companies messing with our industry. They’re doing the job of government regulators in a way that government regulators never would — just shutting things down totally without argument, without accountability. What LiveRamp does is legal. UID2 is legal, and you (the browser company) might have a claim that your customers don’t want it, but it’s a weak claim. You certainly can’t claim that the customers don’t want it from Trade Desk, but they want it from Google.”
Listen on YouTube. (Oct. 2)
Related?
Separately, Apple is tightening macOS “Full Disk Access” controls, citing risks from AI agents. The news comes amid allegations that Meta’s Muse agent read an Inc. columnist’s private messages without permission. Meta denied the report.
Read: “Apple says it’s tightening macOS ‘Full Disk Access’ controls due to new risks from AI agents” (Oct. 2) – TechCrunch
From tipsheet: As AI agents seek more access to our devices and data, Apple’s privacy and security rules will also shape who can compete. What gets blocked, how vendors can appeal and whether the rules apply equally deserve closer scrutiny.
CREATIVE
YouTube prioritizes originality for Shorts reach
YouTube is updating Shorts recommendations to favor original content over videos copied from other creators, 9to5Google’s Andrew Romero reports.
The platform says its recommendation systems will “further prioritize original content and reduce the reach of content re-uploaded from other creators without adding anything of your own.”
The distinction between copying and transforming matters as YouTube expands its AI creation tools. Its Reimagine feature lets users turn a frame from an existing Short into a new eight-second AI video. How those remixes will fare under the recommendation change remains unclear.
Read: “YouTube says Shorts recommendations won’t surface unoriginal content” (Oct. 2) – 9to5Google
Source: “Prioritizing original content on Shorts” (Oct. 1) – YouTube Help
Related: “Google Search now says it’s critical to manually factcheck and review all AI-generated content for accuracy and trustworthiness” (Oct. 1) – Search Engine Land
From tipsheet: Google is making AI creation tools, but still wants creators to add something of their own and check their work.
COMMERCE MEDIA
Criteo on tracking AI chatbot referrals
September research from commerce media and digital ad platform Criteo found that “from January through July 2026, ChatGPT’s share of CPG product-page landings was 1.6x that of search engines, on average. In July, 86% of Criteo CPG clients recorded at least one ChatGPT-referred transaction.”
ChatGPT referrals are producing purchases across Criteo’s CPG client base, though Criteo’s president of retail media Sherry Smith cautioned in a written response to tipsheet: “It’s still early days.”
Asked whether Criteo can distinguish purchases driven by ChatGPT Ads from those following an unpaid recommendation, Smith said:
“There’s an important distinction between paid and unpaid discovery, although the lines can sometimes be less clear in an AI-assisted journey than they are in more traditional channels.
For paid ChatGPT Ads campaigns, brands can measure campaign-associated signals such as traffic, engagement and conversions, depending on the measurement setup and reporting available. Brands can also use path-to-purchase reporting to understand how ChatGPT Ads interact with other channels across the broader media mix. That provides context around the role AI-assisted discovery may be playing within the overall path to purchase.
Organic AI recommendations are different. The path between a recommendation and an eventual purchase can be harder to attribute directly, particularly when a shopper moves across multiple platforms or completes a purchase in a store.
The important thing is to focus on what can be measured consistently. Brands can look at retailer-level sales, units and product-page views before, during and after campaign activity to provide directional context, though that should not be interpreted as direct conversion attribution.”
Her response highlights the difficulty of distinguishing paid influence from unpaid recommendations across an AI-assisted shopping journey.
More broadly, Smith says: “More than 70% of LLM-referred users land directly on product pages, bringing shoppers into high-intent retailer environments where they can continue to evaluate products, convert and engage with relevant media.”
That product page finding also appeared in a Criteo post earlier this year.
From tipsheet: Search has long sent shoppers straight to retailers’ product pages. The potential difference is how much of the shopping decision an AI assistant helps resolve before the shopper arrives. That could leave retailers with less opportunity to influence the choice and makes distinguishing paid influence from unpaid recommendations more consequential.
LLMs & CHATBOTS
Developments
- Brian Chesky interview: AI agents need their own operating system (Oct. 1) – TechCrunch
- The Mulleted, Meme-Loving Billionaire Behind Meta’s Hit AI App (Oct. 4) – The Wall Street Journal (subscription)
- Changes to Gemini model access and limits (October) – Gemini Apps Help
MORE
- Sizable advertisers: “Marketing’s upper middle embraces AI creative production” (Oct. 2) – Digiday (subscription)
- GEO for cars: “Toyota rewrites online marketing playbook to grab AI bots’ attention” (Oct. 2) – Automotive News (subscription)
- Corporate rebrand: “Paramount to change its name to Skydance after Warner Bros. Discovery deal closes” (Oct. 2) – Ad Age (subscription)
- DMV data: “Judge Tosses Suit Against Meta Over Data Collection From DMV” (Oct. 3) – MediaPost
- Agenda: Advertising Week NY, October 5-8 – Advertising Week NY