Agent Beat for October 2, 2026

Welcome to Agent Beat. It's Friday, October 2, 2026. This is the daily briefing on how AI finds brands and how agents choose what to buy.

DoorDash opened agentic ordering to workplaces at its Dash Forward event in San Francisco. DoorDash says companies can point their own AI tools at a connector that automates lunch catering and office supply replenishment. Access is by waitlist. DoorDash named SpaceX, Cognition, Vercel, Mercor and Tempo as early partners. Marketers who sell into offices or buy supplies through DoorDash should expect ordering to move into company AI assistants. They should check whether their catalog and account terms work through that connector.

While DoorDash moves ordering into company assistants, Google is making it easier to see when its assistant sends people to a website. Gemini now adds UTM parameters to the links it sends to websites, so site owners can attribute visits from Google's AI assistant in their own analytics. Search Engine Journal reports the change and says it gives clearer attribution for traffic from the AI platform. Marketers should look for the new parameters in their analytics and report Gemini visits separately from other Google referrals before judging what AI search does for their pipeline.

Attribution matters more as AI systems decide which sources to cite, and video is becoming a major source. Meltwater data reported by Digiday show YouTube was the most cited platform across eight AI systems in August. Those systems include ChatGPT, Claude, Copilot, Google AI Mode, Google AI Overviews, Gemini, Perplexity and Grok. YouTube stayed ahead of Reddit, which had held the top spot after Google agreed to use its content in AI products. Marketers who want their products named in AI answers should treat YouTube transcripts as a citable surface, since these systems are reading them and pointing users to the platform.

If you built tools inside ChatGPT, there is a migration deadline ahead. OpenAI will retire custom GPTs on 11 December 2026 and replace them with plugins, which carry skills, files and connected apps, AI4Business reports. The article says the plugin is also the format OpenAI's new Dots agents work with. It walks through what survives the migration and what has to be rebuilt. Marketers who built a custom GPT to reach ChatGPT users should plan to rebuild it as a plugin, since that is the surface ChatGPT and the Dots agents will call.

Plugins and agents do not only rely on owned content. Third-party signals matter too. A column says Wikipedia entries, podcasts and Reddit threads now shape which brands AI tools cite, and it argues PR teams usually own those signals. The column recommends working with PR on the pages and mentions that language models pick up. Marketers should list the third-party pages that already describe their brand and treat them as citation assets that need the same care as owned content.

That shift in citations is showing up in lead quality. At MarkeZine Day 2026 Autumn, HubSpot Japan's marketing lead described how the company built a four-step answer engine optimization practice after its own site traffic fell by as much as 80%. The session says leads that arrive through generative AI turn into sales talks about three times as often as other leads. Marketers should track AI-referred leads as their own channel and compare that conversion rate with search leads.

As AI becomes a place where buyers talk to sellers, shopping features are moving into chat. OpenAI is rolling out shopping features in ChatGPT. Users can try on clothing and accessories using their own photos, and save products to a Favorites library. The features put product images and saved items inside the chat window instead of on a retailer's site. Marketers with apparel catalogs should check that their images and product data render in the try-on view. They should also watch the Favorites library for signals about what ChatGPT keeps in front of buyers.

But as AI answers more queries, some sites lose the click entirely. Search Engine Roundtable reports that Google answers some team-roster searches with an AI Overview instead of sending the searcher to a site that publishes the roster. The example follows the pattern seen in other AI Overviews, where the answer replaces the page it drew from. Marketers who publish structured lists or reference data should expect fewer clicks from queries Google can answer itself. They should track which of their pages lose impressions that way.

Some publishers are getting paid when their content appears in those answers, but the money is uneven. A new report says Google's AI contribution pilot pays publishers widely different amounts. Some participants get under 0.1% of what they earn from ads. Several say they cannot tell how payouts are calculated. The pilot pays sites whose content appears in AI answers. Marketers at publishers should ask for the payout formula and current participant amounts before opting in. They should treat the program as a small addition to revenue rather than a replacement for lost search traffic.

AI Overviews are also showing up in more places, including brand searches. Search Engine Roundtable reports that AI Overviews are now showing for many large brand-name searches, except for Google's own name and news publications. The outlet says its data is limited and the pattern is not a confirmed rollout. Brand marketers should search their own brand names and check whether an AI Overview now sits above their homepage result.

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