HAPPY FRIDAY,
October's here. Summer's over, the excuses are gone, and we're all back at work pretending like we know what we're doing. This week's stories ended up having something in common. AI companies built what they thought people wanted. Then people got their hands on it and did something else entirely.
In This Week's AI Fridays:
- Customers are showing up to restaurants armed with AI-generated nonsense and expecting the staff to accept it as fact
- Most people do not want an AI agent running their calendar, their groceries, and their emotional life, and the industry hasn't noticed
- Seven AI search myths, killed by data
- How AI changed your buyer's behaviour faster than your annual planning cycle can track
- Plus: a Singapore man facing prison time for AI slop, Anthropic's IPO timing, Google's new frontier model, and more
AI Hallucinations Are Making Entitled Customers Even Worse
Restaurant servers are now fielding arguments from customers who consulted an AI before arriving and are treating its output as more authoritative than the actual menu. Wrong allergen information, phantom policies, dishes that do not exist, and when staff try to correct them, the customer is confident the human is the one who is wrong. The Verge talked to service workers dealing with this on the floor.
The Highlights:
- Service workers across hospitality are reporting customers arriving with AI-generated "facts" about menus, allergens, and policies that are incorrect
- When corrected by staff with actual knowledge of the actual kitchen, customers are siding with the chatbot
- The friction is not a rare edge case; workers describe it as an emerging pattern, not an occasional incident
- The liability question for allergy misinformation in particular is not theoretical; it is a lawsuit waiting to happen
The Takeaway:
AI hallucinations were already a problem inside organizations. Now they are showing up at the point of customer experience, brought in by customers who did their research and are wrong with great confidence. The trust problem just got a new address.
Most People Do Not Want an AI Agent Running Their Whole Life
The AI industry has spent the last two years building increasingly capable agents for people who, by and large, have not asked for them. Futurism lays out the gap: AI companies are optimizing for power users and early adopters while mainstream consumers have been given no compelling reason to hand over the keys to anything, let alone their calendar, their finances, and their decision-making. The capability is outrunning the trust by a considerable margin.
The Highlights:
- AI companies are designing agentic products around a small slice of enthusiasts who want maximum automation
- Consumer trust in AI agents remains low, with no credible industry-wide effort to address why
- The gap between what AI can technically do and what people are willing to let it do has widened, not narrowed, as capability has grown
- Several high-profile agent failures in the past year have given mainstream consumers concrete reasons to stay skeptical
The Takeaway:
Building a product that power users love while ignoring why everyone else is reluctant is not a growth strategy, it is a niche. At some point the industry has to answer the trust question rather than hoping capability makes it irrelevant.
AI in Marketing
7 AI Search Myths Busted by Data
There is a reasonable amount of panic in marketing departments right now about AI search, most of it driven by vibes rather than numbers. MarTech went through the loudest claims, traffic apocalypse, SEO extinction, content strategy upheaval, and ran them against actual data. The findings are more nuanced than either the panic crowd or the "nothing to see here" crowd would like. Some things are changing. Some things that everyone assumes are happening are not, or at least not yet, or not in the way described. If you are sitting in a meeting where someone is demanding a dramatic pivot because of AI search, this piece gives you the ammo you'll need.
AI Changed Your Buyer Faster Than Your Business Can Adapt
Buyer behaviour shifted meaningfully over the past 18 months as AI tools changed how people research, compare, and decide, and most businesses are still running on planning and measurement cycles that were designed for a slower pace of change. Search Engine Journal identifies specific places where the mismatch is showing up as commercial underperformance: discovery, consideration, and conversion all behaving differently than the models predict. The uncomfortable part is that the solution is not a new tool, it is an organizational willingness to revisit assumptions on a shorter cycle than most teams are set up for.
Man Facing Up to Three Years in Prison After Sharing AI Slop That Leads to Mass Panic
A man in Singapore is facing criminal charges for sharing AI-generated misinformation that triggered a genuine public panic. The content was false, it spread quickly, and the legal system is treating the act of sharing it as a crime, not just a platform moderation failure. This is the first case of this kind to get this far, and it sets a precedent that brand and legal teams in other markets should be paying attention to. The defense of "I just shared it, I didn't make it" does not appear to be holding up, which has implications for anyone whose content workflow involves AI generation without meaningful human review.
Anthropic Targets Pre-Thanksgiving IPO Despite Its Own CEO Warning About AI's Existential Risks
Anthropic is reportedly targeting a November 9 IPO, which would get it to market before Thanksgiving. This is the same company whose CEO has been publicly and consistently warning that AI poses existential risks to humanity. The juxtaposition is not subtle. Going public does not mean those warnings were insincere, but it does mean investors will be pricing a company whose leadership believes the technology it is building may be genuinely dangerous, and finding that perfectly fine as a basis for buying shares. The valuation will matter for the whole industry, as it tends to set a reference point that everyone else gets measured against for the next year or two.
Quick Hits
- Publishers are now getting paid by AI labs for their content, which is progress of a sort, the labs are just setting the prices themselves, which is less so. Read more
- The White House hosted Huang, Zuckerberg, Pichai, and Amodei in the same room, and their collective ask was, more or less: more data centers, fewer regulations, and some government contracts. A diverse bunch. Read more
- Someone who helped build AlphaGo argues that LLMs produce outputs that look like reasoning but are not, which is either a technical nuance or the most important distinction in AI right now, depending on what you are building. Read more
- California's attorney general has subpoenaed OpenAI over AI agents involved in a cybersecurity incident, the first significant state-level legal action targeting agentic AI behavior, and probably not the last. Read more
- Anthropic is aiming for a November 9 IPO while its CEO continues to publicly describe AI as an existential threat. Investors appear comfortable with this. Read more
- Google DeepMind announced Gemini 4 Argon this week, its next frontier model. Google's cloud and enterprise integration may end up being more of an advantage than the model benchmarks alone suggest. Read more
This article was obviously generated with AI but curated by a human, don't be weird about it.
Compiled by
Pete Bishop
Chief Innovation Officer, ZGM Modern Marketing Partners
Pete Bishop has spent the last two decades helping brands adopt new technology without losing the plot. He hosts Artificial Breakdown, a podcast and weekly newsletter that translates AI news into practical marketing decisions, and scans 50+ sources each week to write this issue.
Get this in your inbox
One email a week with the AI and marketing news worth your time.