Agent Beat for October 19, 2026

Welcome to Agent Beat. It's Monday, October 19, 2026. This is the daily briefing on how AI finds brands and how agents choose what to buy.

Google has explained how PDFs are handled in Search and in AI Mode. The guidance covers conversion for crawling and indexing, links, images, file limits, and ranking. The Search Engine Journal piece says the guidance matters for SEO and GEO because PDFs can be retrieved and cited in AI answers. It also tells site owners how to check whether a PDF is set up to be indexed.

On the tracking side, ChatGPT stopped adding UTM parameters to redirect links for a week. The article's headline calls it surprising that a player like OpenAI is exposed to a bug as basic as removing UTM. Julien Ribourt, senior partner for data, AI and agentic at Converteo, said the impact on CAC 40 company sites was moderate. The article examines how the missing tags affected tracking of ChatGPT referrals. It follows OpenAI's statement that the interruption was a bug.

A new paper describes the ChainIT Authority Protocol, which it calls a pre-execution authority and evidence control plane for transactions involving software agents. The protocol aims to give counterparties, payment rails, marketplaces, and agent runtimes compact proof that the party presenting a transaction holds current organizational authority under applicable policy, budget, and approval state. The author says agentic commerce makes the gap urgent because an agent can sign a valid transaction outside its delegated purpose or after its sponsor's authority changes.

In shopping recommendations, Jellyfish's Share of Model shopping tool tracked product recommendations from Alexa for Shopping, ChatGPT, and Google AI Mode across eight categories in the US, UK, Australia, and Singapore. When asked for toys, Alexa for Shopping showed products from 177 brands, almost all from Amazon's own store, while ChatGPT recommended 24 retailers and Google AI Mode 37, according to the analysis. The piece says a brand's visibility depends on whether its products are sold by the retailers each assistant draws on.

Being mentioned by a generative AI and being described well are two different things. The article on AI reputation draws on a 2026 barometer of CAC 40 companies to separate whether a company appears in an AI answer from how that answer portrays it. It says executives should not treat a mention as proof of good positioning.

In B2B marketing, brands are exploring creator video to build credibility and get attention from large language models, even though B2B influencer marketing lacks the direct-response playbook common in B2C. The article describes longer sales cycles and niche audiences as obstacles. It treats LLM attention as one reason B2B marketers are trying the format.

For paid search teams, a PPC column says click-based measurement leaves out part of the buying journey as more of that journey moves into AI. It recommends conversion tracking, impression signals, and business records to validate attribution. The advice is for PPC teams adapting measurement to AI-mediated search.

The article on Meta reports a change to Meta's business terms that experts told Nieman Lab could amount to permission to scrape websites, including for training its Muse models. Meta communications head Andy Stone denied that the change was meant for that purpose, according to the article. It also says The Verge reported that Meta is changing the passage again.

For exporters, GEO work should start before the company opens in a foreign market. The article says a small company can be known to AI systems in a market without being recommended there. It separates being known to an AI from being recommended by one.

A survey by the media company MERY found that fewer than 20% of respondents buy a product based only on an AI assistant's recommendation, Web担当者Forum reports. Young women said they check social media and stores before buying, and also consult AI when choosing cosmetics and clothes. The outlet gives no detail on the survey's sample or method.

In Korea, a TBWA digital marketing column says ChatGPT ads went live in August 2026. The ads appear in a separate ad area below the answer and do not change the AI answer itself. The piece argues brands still need GEO because the ads reach a limited audience and some categories are restricted. It also describes context hints, extra detail advertisers supply to help the system judge when an ad fits a conversation.

Replit CEO Amjad Masad said OpenAI would replace Anthropic as the platform's default model provider, TechCrunch reports. Claude will still be available to customers on the platform.

TIME has named Google SynthID one of the Best Inventions of 2026 for its work on AI watermarking and verification. Google announced the recognition in a blog post. The item does not describe a new product change.

Jev is the fastest-adopted model in the history of Vercel's AI gateway, TechCrunch reports. The figure comes from the gateway, a unified API service similar to OpenRouter.

That's the briefing for Monday, October 19, 2026. Thank you for listening to Agent Beat. For the full written brief, go to agent-beat.com.