Agent Beat for October 23, 2026

Welcome to Agent Beat, the daily briefing on how AI finds brands and how agents choose what to buy. Today is Friday, October 23, 2026.

The number of advertisers on ChatGPT rose from about 20,000 at the end of August to roughly 60,000 in early October, according to Dan Salmon, a partner at New Street Research. Clicks on those ads cost about $5 each, against $30 to $50 for similar commercial software ads on Google. The Information reports that OpenAI projected $2.4 billion in ad revenue to investors for 2026 and is likely to miss that figure. Those are the advertiser count and the revenue projection in the report.

Google chief executive Sundar Pichai said he expects AI agents to reach websites in two ways: by browsing pages and through protocols. He said protocols scale better but will take longer to build. The distinction is between agents browsing pages themselves and agents using protocols. Search Engine Journal reports the remarks.

A MyBest survey reported by MarkeZine found that 72% of people who had used generative AI for shopping in the previous six months said they want to use automatic ordering. The survey result is about that group of shoppers and their interest in automatic ordering. The item gives no sample size and no further detail.

1Password launched a way for AI agents to sign in to a customer's personal accounts on the customer's behalf. The credentials that secure those accounts stay private. The customer's password stays private while the agent signs in. TechCrunch reports the launch and describes the agent signing in without seeing the password.

Search Engine Roundtable rounds up three studies published in the past week. All three found that AI search services send a small fraction of the traffic and conversions that Google Search sends. The post says the findings call into question the time, money and resources spent on AI visibility. The findings concern traffic and conversions from AI search services compared with Google Search. The roundup covers those three studies together.

Marie Haynes looked at 12 months of GA4 data from 100 websites. She concluded that chasing visibility inside AI answers misses where the return is as the web becomes agentic. She argues that AI is rewriting how the internet works, but that GEO as commonly sold is a distraction from the agentic web. Her conclusion comes from that year of data across those 100 sites.

Semrush published a case study on how the feedback tool Feature Upvote corrected an incorrect fact Google's AI Overview gave about its API. The company says the fix took days. The post walks through what the company changed to get the AI Overview to say it offers an API. The case study is from Semrush, and it walks through the changes Feature Upvote made.

A TBWA digital marketing column says AI agents are moving past recommending products. They are searching, comparing, checking delivery terms and paying. The column cites Korean figures: it reports Naver's shopping AI agent has lifted transaction value 82% since its June launch and conversation volume 60% as of September. It argues that e-commerce brands need product data and page structures an agent can read. That extends existing GEO work on accessibility, structure and trust. The advice focuses on product data and page structure.

Mike King takes on three studies that found AI assistants send very little traffic. The studies are from Marie Haynes, SEO Works and Kevin Indig. King argues they grade a brand channel with a traffic ruler. The volume figures are sound, he writes, but AI answers work more like advertising than like a source of clicks. His argument is about the measurement, not the traffic counts.

Search Engine Journal says September's rank tracking outages exposed gaps in keyword-position reporting. The piece sets out brand visibility, AI citations and revenue as the measures to use instead. The recommended measures are brand visibility, AI citations and revenue. The piece says those measures should replace keyword-position reporting.

Bing Webmaster Tools added a Topics Portfolio. It shows how well a site performs on broader topics in AI search and where the potential is, SEO Südwest reports. The post describes the addition as a step in AI tracking and site monitoring. Topics Portfolio is the name of the addition.

Google's new robots.txt video, presented by Gary Illyes, says the standard covers crawling. It says nothing about how content is used afterwards. It also recommends percent-encoding non-English paths in robots.txt rules, Search Engine Journal reports. The video separates crawling rules from later use of content. The standard covers crawling, but not how content is used after it is crawled.

That's today's Agent Beat. Thank you for listening. For the full written brief, go to agent-beat.com.