Search Agents, Universal Cart, and the End of Passive Discovery

📬 You like this podcast? You’ll love the newsletter.
Join the weekly 3-2-1 on AI search + marketing: subscribe

Google I/O was Tuesday. There's a lot to sort through — and most of it isn't the hardware story. It's the search story. This week, Cassie breaks down the I/O announcements that actually matter for visibility strategy, unpacks the llms.txt drama that played out in a LinkedIn comment section yesterday, and covers Perplexity updates that got buried under all of it.

In this episode:

• Google I/O: the Search box redesign, AI Mode at 1 billion users, Search agents, and Universal Cart
• Why Search agents change what "being discoverable" actually means
• Google's GEO guide said stop using llms.txt — then Lily Ray pointed out Google uses it on their own docs. John Mueller's response draws a distinction the guide never made.
• Why GEO isn't just an SEO conversation — it's the infrastructure that makes agentic commerce work
• Bing on trust as infrastructure in AI-powered search
• Perplexity: GPT-5.5 as the new default in Computer, Teams integration, and Personal Computer at $20/month

Google is building toward a world where your website is the backend. This episode is about what that means for your strategy right now.

Let’s connect:

LinkedIn → Cassie Clark | Fractional Content Strategist
Website → https://cassieclarkmarketing.com

Download Freshness, Structure, Authority: The Framework for AI Search Visibility:

Amazon

P.S. Is your brand losing its "Answer Authority"?

Most series A/B and enterprise brands are being "nudged" out of AI search results because of entity gaps and "stale" content. I am opening a limited number of specialized audit slots to help you reclaim your Share of Voice using the FSA Framework (Freshness, Structure, Authority).

Request your 7-Day AI Search Visibility Audit: https://cassieclarkmarketing.com/ai-search-visibility-audit/

If you're listening to this and thinking I need someone to lead this for me, that's what I do.

I'm an AI search optimization expert and a fractional content strategist for startups and enterprise brands. If that sounds like the kind of help you're looking for, email me at cassie@cassieclarkmarketing.com.

Transcript

SPEAKER_00 0:00

Hey, welcome back to Found in AI. I'm Cassie Clark, a fractional content strategist and an AI search optimization expert. I'm also the host of this show where we break down AI search, GEO, AEO strategies, and what all this means so we don't get left behind in this new wave of user search behavior. Today is Thursday, May 21st, and we have a lot to cover this week, and I do mean a lot. Google I.O. was yesterday, which means there's a pile of announcements to sort through, not all of which are equally interesting for our purposes. So I'm gonna skip over the stuff that's more hardware related or model news and just focus specifically on what's changed for search, visibility, and AI-driven discovery. Then we're gonna talk about something that happened in the comment of an ex post yesterday that I think is actually the more important framing of everything that has happened this week, maybe more important than any individual product announcement, because it changes how we should be thinking about what GEO is even for. Then we'll close with perplexity, who had a significant few weeks that got completely buried under the Google coverage. Okay, let's get into it. Okay, Google I.O. There was a lot, and I'm gonna give you the four announcements that are actually worth paying attention to if you're in the marketing space. First, that search box. Google called this the biggest upgrade to the search box in over 25 years. The new field dynamically expands as you type, so it's built for longer, more conversational queries. And the AI powered suggestions now go beyond just simple autocomplete. They're designed to anticipate your intent, not just complete your phrase. This is Google officially acknowledging, out loud, that keyword search is over. Users have already moved on to full-on sentences, conversational queries, they've been doing that in Chat GPT and Perplexity in Gemini. Google is just catching up to that behavior. Second, we need to talk about AI mode. Now they did not officially pull the plug on this and make this the default thing for search, but AI mode has 1 billion users monthly now, and it now runs on Gemini 3.5 Flash. So this means that this is not a feature, but this is the whole product. If you're still treating AI mode as an experimental overlay on traditional search, consider this your sign to update that mental model. Third, they announced information agents. This is the one that I think is the most fascinating. So they call these things information agents. It's an AI that runs in the background 24-7, scanning blogs, news, social posts, real-time data, and notifying you in something that matches what you've told it to track. The consumer examples they gave were apartment hunting and sneaker drops. But I want you to think about this from a B2B lens for a second. A buyer could set a search agent to surface the moment a solution for their specific problem appears just somewhere, anywhere on the web. Not when they think to search it, but just an agent continuously, automatically looking for it on their behalf. That changes what discoverable means in a fundamental way. Your content doesn't just need to rank, but it needs to be the thing the agent finds, flags as relevant, and then can do something with it later. We'll come back to why this matters when we talk about the GEO guide. Again, I know we talked about it on Friday. We need to talk about it again, but we'll talk about it in a minute. Fourth, Google announced Universal Cart. Universal Cart is a Gemini powered shopping experience across search and eventually in the Gemini app, YouTube, and Gmail. It aggregates products for multiple retailers, it monitors price drops, flags incompatibilities, and then draws on Google Wallet for payment and loyalty data. The checkout lane is now inside the answer layer. For anyone in e-commerce or product marketing, this is where that transaction is headed. But here is the through line for everything that Google announced this week. Google is moving from a platform that sends people to websites to a platform that keeps people inside of Google, answering the questions, completing the task, closing the transaction. Your website is just the back end of all this. That's not criticism, that's just the architecture that they're building now. And it has real implications on how we think what optimization is even for. Which brings me to segment two. This is the segment I want you to pay the most attention to because it kind of reframes the entire conversation around this GEO guide and everything we've been talking about. Google is being sneaky once again. Last week, before I.txt files, content tucking, AI specific rewriting, and special schema. They said, nay nay, these are not for us, just ignore these. Now, if you have been paying attention to the GEO conversation over the last year, you know that some of these tactics, particularly that LLMS.txt file, has been circulating as things that you should be doing for AI visibility. Entire tools have been built around generating these files, but Google just said, uh, whatever, that's not for us. But then yesterday morning, Lily Ray, BP of SEO Algorithmic, posted something that just kind of cut through all of us. She tagged John Mueller on X, John Mueller, he's from the Google search relations team, and pointed out that Google itself uses the LLMS.txt files and markdown files on its own developer documentation, despite having just told the rest of us just last week that these things do not matter for search performance. Now, bless him, Mueller responded. His answer is worth taking a hard look at because it adds a layer that Google's official guide did not include. His answer, in short, was lms.txt files isn't done for search. And he says, there's more to websites than just SEO. Gonna pause there for a second. Now he drew a distinction between two different things. He says that discovery or getting found by search engine in the first place is different than functionality or what happens once an agent or an AI system has already found your page and needs to do something with it. He compared it to CTAs on a traditional page. So what that means is like you're not optimizing a CTA to get found, but you're optimizing for what happens after someone lands on that page. And for developer documentation specifically, when AI coding systems are trying to read reference material and execute against it, a structured, easily parsable format helps these systems work more accurately. He was pretty clear that for non-technical sites like maybe a shoe brand or a B2C product page, that a markdown version of your content is probably not gonna do much. But a markdown version of a shoe spec, as he puts it, is not gonna get you more sales, which is interesting that he says that. I have more commentary on this in later episodes with AJ. But anyway, here's why I think this exchange is actually the most important thing that's happened this week, despite everything else. I've been saying a version of this in every agentic commerce conversation I've had for months. GEO is not just an SEO versus GEO debate. It's not, these are not things that we need to be arguing about because GEO is just the underlying infrastructure that makes agentic commerce work. Google's official guide was talking about discovery, getting your content found and cited in AI answers. That is one layer. Mueller's response was talking about functionality. What happens once an agent has found your site and needs to take action? Those are two different optimization surfaces. And right now, most of the GEO conversation is focused on the first one, while the second one is becoming more important than anyone else is talking about. Think about what Google announced on Tuesday. Search agents that run 24-7, universal cart that completes purchases inside search interfaces, the ability for AI to book, buy, and schedule on your behalf. None of that works if your site is only optimized to be found. It also has to be ready for an agent to do something once it arrives. That, my friends, that is the shift. And Google is explicitly building toward it. They have been on and on about agentic search for months. These announcements are just the infrastructure. In the GEO conversation, in my humble opinion, it needs to catch up to where all of this is going. Alright, that's enough about Google for a minute. Let's talk about Microsoft because they dropped something last week on their Bing blog and I think it deserves a mention. They published a post framing what they call trust as infrastructure and AI-powered search. The core argument is that as AI reshapes how answers are generated and surfaced, keeping trusted visible content isn't a secondary feature. It's a foundational to how AI-powered search should function. Now, their specific focus is on safety, like public safety announcements that surface at the top of results when a query signals risk. Safe search controls with transparency about when and why content is filtered. And the principle that when risk is detected, support should be more visible than harm. Now, this might seem like a different kind of story than the GEO and visibility conversations we usually have here, but I think it connects in a meaningful way because Bing is consistently the engine doing the work of making AI search governable and transparent. They launched the AI Performance Dashboard earlier this year. They're publishing on trust infrastructure. They released their AEO good back in January. They are thinking out loud about how these systems should behave at a policy level. While Google is saying, oh, it's just LSEO, Microsoft is out here building the accountability layer around AI Search as its own channel. Those two postures keep showing up in contrast to each other, and I think it's both worth tracking. It is interesting for sure. Alright, let's close with Perplexity because they also had a significant few weeks and it kind of got glazed over by the Google noise. I do want to give it the time it deserves, especially if you're using these models in your work. So the May 4th changelog is kind of where it all started, and there were a few updates in it that matter in relation to AI search. First, Perplexity swapped in GPT 5.5 as the default orchestration model inside Computer for Pro and Mac subscribers. It's designed to handle complex tasks, multi-step browsing, and long-running agent workflows are faster and more reliable. Now, here's a small operational note I want to call out. When the default model underneath your workflow changes, prompts that were working before can stop working the same way. So if you've built repeatable workflows in computer or are using it for an agent function, it's worth taking the time to check to make sure everything is okay. Now, second, the Microsoft Teams integration. Computer is now available as a Microsoft Teams app. It lets Teams orchestrate research, document creation, and workflow automation directly inside Teams, either by DM or by mentioning computer in a channel. So if your team or your coworkers are already living inside of Team Teams, this is a big change. And it means we need to start thinking about how brands show up inside other AI models, not just Google's. Finally, personal computer came to pro. This is an update that probably should have gotten more coverage, but it did not. On May 7th, Personal Computer, which is Perplexity's local Mac agent, it went from being a Macs only feature at $200 a month to available on all pro plans at just $20 a month. Now, what does Personal Computer actually do? Well, it can access your Macs file system and your native apps like mail, calendar, Slack, Notes, Finder, and then execute entire workflows. It controls apps directly, the way a person sitting at your computer would, with actions that are auditable and reversible. Now, the demo that Perplexy showed, you can open notes, you can have three things on your to-do list, like schedule a meeting, send a Slack update to your comms team, push a new design file to homepage, you type take care of my to-do list, and then it just does it for you. Now, whether that sounds exciting or slightly unsettling depends on how your Tuesday is going. Both of those reactions are valid. Okay, so if I had to pull a single thread through everything that we just talked about is this the Google announcements, like those search agents, Universal Cart, personal intelligence, these are not isolated product features. They are the infrastructure for the future where agents don't just find information for you, but they act on it. They book, they buy, they schedule, they execute on your behalf. The GEO conversation needs to evolve to match that architecture. Discovery absolutely matters. You need to be visible inside of these AI agents for any of this to work. Being cited matters, but functionality, being ready for what happens when an agent finds you, that is the next layer, and not enough people are talking about it yet. John Mueller gave us a glimpse of this this week, even if it wasn't framed that way. The distinction between optimization for discovery and optimization for agent functionality is real. And if Google is building the infrastructure for agent e-commerce, which looking at Tuesday's announcements and everything they have said up until this point, they clearly are. Then the brands that win are going to be the ones who understood that distinction early. Okay, that's it for this week. I know it was a lot. If you found this episode useful, I love if you would hit subscribe and leave a review. It does actually help the show reach more people who are also trying to figure this out. And if you're wondering where your brand actually stands in AI Search right now or how ready your site is for the agentic layer that we just talked about, that's exactly what an AI Search Visibility Audit is for. You can find more at CassieClarkmarketing.com or more information in the show notes. I will see you on Tuesday. Until then, stay visible.