Details emerged on Friday about a test of Anthropic’s LLM chatbot Claude and its ability to manage a physical “storefront.”

tl;dr: The mixed results showed that the AI struggled with strategies around pricing and inventory management.

TechForge’s AI News further explored the details: “…the AI was easily persuaded to offer discounts on products from the business. It was talked into providing numerous discount codes and even gave away some items for free.” Read more.

The source of the story was Anthropic itself which shared results of the test on X and added its own sense of bewilderment:

“Some of those failures were very weird indeed. At one point, Claude hallucinated that it was a real, physical person, and claimed that it was coming in to work in the shop. We’re still not sure why this happened.”

Reminder: Anthropic is backed by online retail giant Amazon.

Versus programmatic

On the New Digital Age podcast published last Thursday, Scope3 CEO Brian O’Kelley was ready for artificial intelligence and no more “status quo.”

O’Kelley said, “AI is a better matching system with no training than the entire $120 billion programmatic ad industry. So, maybe we should do that. And, from a sustainability perspective, having AI make that one decision was dramatically more energy efficient than thousands and thousands of bid requests making bad decisions.”

Hear the rest on Apple Podcasts or read the auto-transcript.

Web visits for Google AI

Data company Similarweb says that its web analytics show a significant rise in web traffic to Google’s AI properties in just one month.

From Similarweb on X last Thursday, month-over-month traffic growth by total visits (from April to May 2025) were as follows:

Google DeepMind: 4,258,300 → 11,189,832 (+162.78%)
Google Labs: 9,796,459 → 19,792,943 (+102.04%)
Google AI: 1,349,027 → 2,042,165 (+51.38%)
NotebookLM : 48,340,200 → 64,906,295 (+34.27%)
Gemini App : 409,432,013 → 527,738,672 (+28.90%)

Commenting on the data, Google DeepMind CEO Demis Hassabis said on X on Friday, “good trajectory….”

Google AI traffic

Analyzing Google AI, ads

New Street Research analyst Dan Salmon continued his deep dive on all things Google and AI in a note to investors on Friday.

Among his “takes,” Salmon said, “We don’t think AI Mode will be made default for US users anytime soon (nor the new test group in India), but we think US users could be given the choice to do so eventually.”

Also of note coming out of his recent visit to the SMX search engine marketing conference: “…the new AI Max adds to Performance Max and Demand Gen in [Google’s] quiver of AI-automated buying tools.”

On emerging competitors to Google in the AI + search + monetization race, “Perplexity is the second most frequently cited. We think there should be more focus on Meta AI.”

AI yield optimization

Free smart TV provider Telly has integrated sell-side startup Swivel’s (was PilotDesk) AI-driven yield optimization tools – “including trafficking, inventory management, and partner workflows” – according to a June 26 press release. Read more.

Swivel, which includes former execs from SpringServe/Magnite and Freewheel, just raised a $5.8 million round in April after a $3.2 million seed raise in May 2024.

Also last Thursday, AdExchanger’s Anthony Vargas spoke to Swivel CEO Frans Vermeulen about his company’s latest updates. Swivel was initially aiming to target CTV and streaming publishers with their product line, according to Mr. Vermeulen. But, he “anticipates that the company’s focus will ‘expand dramatically’ by next year. Traditional print and web publishers, as well as retail media, digital audio and out-of-home sellers, are all on the radar.”

Podcast: cardinality

On the Adtech Adtalk podcast with ad tech CEOs Adam Heimlich of Chalice AI and Gareth Glaser of Gamera, the hosts get in the ad tech “weeds” and discuss the meaning of cardinality and its involvement in ad tech and AI.

Heimlich said [1:00:26, lightly edited]:

“…cardinality means there’s too many values for a machine learning algorithm to work(…) -they work in orders of numbers.”

”So something like the topic of a web page had to be aggregated into categories [for machine learning] … it’s a ‘sports page’…. -[that’s] weaker than all the differences within sports. It’s a ‘hockey“ page, a ‘soccer’ page… The cardinality gets too high when you start throwing in all these words that have nothing to do with each other…”

“[Large] language models are the solution for [cardinality]. It’s a model of the language that captures the distance between ‘hockey“ and ‘soccer’ and ‘football’ so that all these values can be ordinal values used by machine learning to make predictions and lookalike modeling and all that. So at Chalice, we have solved many cardinality problems such as website URLs using large language models.”

“You can think of it as wringing more predictive power out of the same data sets that have always existed. So there’s a lot of companies that talk about AI and one of the real differences when you are talking to Chalice or someone else is -using AIs, they will be able to show you how much better predictions get when the very cardinal data like language gets harvested and utilized.”

Hear more on Apple Podcasts app and read the auto-transcript.

Prompt: Why does cardinality matter in ad tech?

Response from OpenAI’s ChatGPT:

1. Audience Segmentation & Targeting

High-cardinality fields (e.g., cookies, device IDs, hashed emails) are essential for precise targeting.
Enables creation of highly specific audience segments, such as “people who visited a specific product page in the last 3 days.”

2. Frequency Capping & Reach Measurement

Unique identifiers (high cardinality) help ensure users don’t see the same ad too often.
Accurate reach and frequency metrics depend on being able to count distinct users.

3. Attribution Modeling…

podcast: AI creative opportunity

DTC marketer Andrew Faris of AJF Growth said on the June 27 episode of his eponymous podcast that if you’re building with AI and creative today, there is no reason to make things look more “real“ or “human.“

With the new tools at the creator’s disposal, Faris said it’s time to seize the unique opportunity – which may not last forever – and advised [17:47]:

“The right way to think about AI creative: Do all kinds of crazy stuff that you could never do because now you could do AI. I don’t know if that will last forever or it will become too obviously AI and everybody will get over it…. but the way I’ve thought about this is, I basically have the capacity of a major Hollywood movie studio for CGI, for $100/month, at my fingertips.”

Hear the podcast titled, “The AI Ad Creative Opportunity Isn’t What You Think It Is” on Apple Podcasts.

Faris’ statement aligns well with recent ad creative such as the much-discussed Kalshi ad produced for $2,000 and released two weeks ago.

related: AI-manipulated photos are upsetting house plant communities – The Verge

IP and the law

Musk’s X Settles Data-Scraping Dispute Against Israeli Data Firm (June 23) – Bloomberg Law (subscription)
Microsoft sued by authors over use of books in AI training (June 25) – Reuters
Meta Won Its AI Fair Use Lawsuit, but Judge Says Authors Are Likely ‘to Often Win’ Going Forward (June 28) – CNET

Google’s Offerwall

Publishers have been using various ways to bridge between gated subscriptions and advertising and last Thursday Google joined the party with Offerwall,

The Google Keyword blog explained, “When publishers choose to use Offerwall, they can offer audiences a number of ways to access content. People might decide to watch a short ad, complete a quick survey or pay in micropayments. Publishers can even add their own options, like newsletter sign ups. These options empower audiences to decide how they want to access publishers’ sites and help ensure diverse content remains available to everyone.”

TechCrunch took another approach to coverage of the news and reported, “Google’s AI search features are killing traffic to publishers, so now the company is proposing a possible solution.” Read that one.

AI economic opportunity

On Friday, Anthropic announced its “Economic Futures Program” (June 27) and said “Our goal for this program is to contribute to the development of new research and potential responses to the impacts of AI on the labor market and global economy.”.

See examples of recent data that Anthropic – maker of the LLMs for the Claude chatbot – said would be used for an “index” here.

OpenAI launched something similar in January called “Economic Blueprint” which “outlines policy proposals for how the US can maximize AI’s benefits, bolster national security, and drive economic growth.” This followed a similarly named 2024 white paper.

For its index, Anthropic said it isn’t just going to use its own data for input, but also indicated it will be reaching out to experts “including economists and policy professionals, on this initiative.” It’s unclear how these selections will be made.

thought bubble: Both OpenAI and Anthropic’s tools appear to be political AND media tools to advance interests of each company, respectively. Loosely:

OpenAI is taking the glass-half full position with AI.
Anthropic is taking the glass-half empty position with AI.

more: “The Promise and Perils of OpenAI’s Proposed AI Economic Zones, Part 1: Introduction” (March 5) – Daniel Goetzel, Harvard Kennedy School

Doomer op-ed

AE Studio CEO Judd Rosenblatt and research director Cameron Berg outline a lengthy, bleak vision of LLMs and, in particular, OpenAI’s chatbot in a Wall Street Journal op-ed titled, “The Monster Inside ChatGPT.”

After numerous examples of what Rosenblatt and Berg said were problematic responses from the chatbot, the team wrote, “We have to do what America does best: solve the hard problem. We need to build AI that shares our values not because we’ve censored its outputs, but because we’ve shaped its core. That means pioneering new alignment methods.”

More stuff

Calls for Australia-wide crackdown on real estate ads that use AI to hide faults and lure in renters (June 29) – The Guardian
Op-ed: “Here’s a Tip to Companies: Beware of Promoting AI in Products” – quote: “Consumers have less trust in offerings labeled as being powered by artificial intelligence, which can reduce their interest in buying them, researchers say” (June 29) – The Wall Street Journal
In the race for ad dollars, platforms seek to redefine themselves as AI companies (June 25) – Digiday
OpenAI Leadership Responds to Meta Offers: ‘Someone Has Broken Into Our Home’ (June 29) – Wired
MCP: An (Accidentally) Universal Plugin System (June 29) – Works On My Machine blog

Anthropic Tries To Close The Storefront Loop