บทถอดความฉบับเต็มของ Google Updates Content Guidance as AI Agents Reshape Search

if you go check that page now, probably not in the answer because it has sat dormant for a while. 00:03:27.659 --> 00:03:30.939 I need to redo that test actually and see what happens. 00:03:31.179 --> 00:03:41.179 But I also really want to be really careful here because this experiment does not prove that just adding a TLDR makes perplexity sight you more. 00:03:41.419 --> 00:03:50.459 So please do not go add 400 TLDRs to your blog post and then tell everyone that I said it was a secret to GEO. 00:03:50.699 --> 00:03:53.819 Like that's not how any of this works. 00:03:54.060 --> 00:04:08.619 What this experiment showed me though is that making information easier to identify, understand, and extract appeared alongside a very fast change and how often that page was being surfaced in that particular test. 00:04:09.099 --> 00:04:15.819 And now we're seeing Google put more explicit language around the quality and organization of the main content itself. 00:04:16.220 --> 00:04:22.300 So to me, this is another place where SEO and GEO absolutely overlap. 00:04:22.939 --> 00:04:29.259 Not because they're the same thing, they are not, but clear structure helps humans understand information. 00:04:29.819 --> 00:04:33.259 It also helps search engines understand information. 00:04:33.579 --> 00:04:41.100 And based on what we're seeing in AI search experiments, it can help AI engines work with that information too. 00:04:41.340 --> 00:04:44.860 So this really isn't about stuffing more headings into an article. 00:04:45.019 --> 00:04:46.620 Please, please don't do that either. 00:04:46.780 --> 00:04:53.500 But it's about making the purpose of the page obvious and making the information somebody came for easier to find. 00:04:53.660 --> 00:04:58.060 And frankly, I think that should have been good content strategy all along. 00:04:58.780 --> 00:05:04.140 Now, Google also made another change to that same helpful content documentation. 00:05:04.460 --> 00:05:06.620 This one's a little less nuanced. 00:05:06.780 --> 00:05:09.259 They just said stop making up authors. 00:05:09.580 --> 00:05:15.259 Google specifically warns against what it calls deceptive authorship information. 00:05:15.740 --> 00:05:26.300 Its examples include AI-generated headshots, fake names, and false credentials that make content look as though it was written by a human expert, even if it wasn't. 00:05:26.540 --> 00:05:32.780 Google says deception like this makes a page untrustworthy to users and it's automated quality systems. 00:05:32.939 --> 00:05:35.660 Now, my reaction to this one is pretty simple. 00:05:35.980 --> 00:05:38.460 You cannot manufacture authority. 00:05:38.780 --> 00:05:45.580 We have spent a long time watching people try to reverse engineer what signals they think Google wants. 00:05:45.819 --> 00:05:53.100 Now we're watching the same thing happen with AI search and these new AI engines, and everyone really just wants the shortcut. 00:05:53.340 --> 00:05:57.340 They're asking, can I add an expert reviewer box and does that count? 00:05:57.580 --> 00:06:00.540 Can I put an author bio on every article? 00:06:00.860 --> 00:06:03.259 Can I manufacture mention somewhere? 00:06:03.420 --> 00:06:07.500 Can I publish enough content that eventually the AI has to notice me? 00:06:08.060 --> 00:06:12.379 But an author box doesn't really make someone an expert. 00:06:12.620 --> 00:06:16.460 A fake headshot certainly does not make somebody an expert. 00:06:16.700 --> 00:06:18.700 And putting reviewed by Dr. 00:06:19.020 --> 00:06:25.340 Somebody on a page doesn't magically give the underlying information any more credibility. 00:06:25.580 --> 00:06:29.340 Authority is something that you build, it's something that you work on. 00:06:29.580 --> 00:06:33.020 It's the expertise attached to the people behind your company. 00:06:33.180 --> 00:06:39.100 It's your original research, it's earned media, it's what credible third parties say about you. 00:06:39.420 --> 00:06:43.980 It's whether the things that you're saying on your website are corroborated elsewhere. 00:06:44.220 --> 00:06:52.939 And that becomes even more relevant in AI Search because these engines aren't necessarily evaluating your brand based on one paint on your website. 00:06:53.180 --> 00:06:55.819 They're pulling information from across the web. 00:06:56.220 --> 00:07:00.140 Which actually brings us pretty nicely to the next story. 00:07:00.379 --> 00:07:04.220 Now, I am probably going to mispronounce this brand. 00:07:04.300 --> 00:07:06.379 I'm not actually sure how you say it. 00:07:06.700 --> 00:07:19.500 But SOCI released its 2026 Local Discovery Index based on a survey of more than 1,000 US consumers, and one number just immediately jumped out of this research. 00:07:20.540 --> 00:07:34.379 The percentage of consumers who said they had used an AI tool to find a local business in the previous 30 days went from 9% in 2025 to 52% in 2026. 00:07:35.020 --> 00:07:39.819 That's close to a six-fold increase in just one year. 00:07:40.140 --> 00:07:46.140 Here's another part of the story that I found particularly interesting or the study. 00:07:46.300 --> 00:07:48.540 Maybe I should say study and not sorry. 00:07:48.860 --> 00:07:54.379 Either way, people are using AI more, but they're not necessarily trusting it more. 00:07:54.700 --> 00:08:02.379 67% of AI users surveyed said they received incorrect information about a local business at least once. 00:08:02.540 --> 00:08:05.180 Now I think I have talked about this on the podcast. 00:08:05.420 --> 00:08:06.780 I have also had this happen. 00:08:06.939 --> 00:08:10.860 It was an incorrect phone number when I was trying to call a doctor out doctor's office. 00:08:11.100 --> 00:08:13.660 I got some random lady's cell phone number. 00:08:14.300 --> 00:08:27.020 And according to this survey, when an AI tool recommends a business, 81% of those respondents said they take some kind of additional step to verify it before

contacting the business. 00:08:27.259 --> 00:08:34.139 They might check reviews or look at social media, maybe run another search, or just check multiple sources. 00:08:34.620 --> 00:08:38.139 SOCI calls this a verification loop. 00:08:38.379 --> 00:08:41.740 And that is the part the marketers need to pay attention to. 00:08:41.980 --> 00:08:48.059 Because we are still talking about AI visibility as though the goal is to get mentioned in an answer. 00:08:48.779 --> 00:08:53.980 But imagine somebody asks Chat GPT for a restaurant recommendation. 00:08:54.299 --> 00:09:03.740 Chat GPT then recommends you, which is great, but then the person goes to Google and sees that your hours don't match what ChatGPT had just said. 00:09:03.980 --> 00:09:08.220 Or they check Instagram and they see that you haven't posted since 2024. 00:09:08.379 --> 00:09:14.779 Or they look at reviews and discover three people saying something completely different than what the AI just told them. 00:09:15.659 --> 00:09:19.500 Did you actually win because you got the recommendation? 00:09:19.980 --> 00:09:21.180 No, not really. 00:09:21.420 --> 00:09:25.659 You entered the consideration set, which is great, but that's different. 00:09:25.899 --> 00:09:31.899 This is why I keep saying that AI search visibility is bigger than SEO and bigger than citations. 00:09:32.139 --> 00:09:38.779 The buyer doesn't experience your marketing channels in the neat little organized boxes that companies use internally. 00:09:39.420 --> 00:09:42.139 And it seems neither does AI. 00:09:42.460 --> 00:09:55.420 Your website reviews, your social presence, your PR, your third-party coverage, your local listing, all of those things are contributing to the information environment around a brand. 00:09:56.220 --> 00:09:59.180 And they all need to be as accurate as possible. 00:09:59.500 --> 00:10:09.420 Which brings me to the last story because we're now moving away from AI helping somebody research a business to AI actually doing some of that research for them. 00:10:09.899 --> 00:10:16.460 So Profound published research this week, it was October 2nd, looking at two personal AI agents. 00:10:16.700 --> 00:10:19.340 That is Muse, and then the other one is Instinct. 00:10:19.580 --> 00:10:22.139 Shout out to not Brown at Profound if you're listening. 00:10:22.300 --> 00:10:24.460 Thank you for sending this straight to me. 00:10:24.779 --> 00:10:29.100 This one actually just caught my attention because I've been playing with Muse myself. 00:10:29.340 --> 00:10:35.580 If you've been listening to the podcast or you're on the visibility report, I've been talking about this a little bit. 00:10:35.899 --> 00:10:42.379 Personal agents are different from answer engines that we've spent the last couple of years talking about. 00:10:42.700 --> 00:10:47.580 With something like ChatGBT search, you go ask a question, you get an answer, and you're done. 00:10:47.820 --> 00:10:56.700 A personal agent, though, like Muse,

or where Gemini is kind of headed, is designed to actually carry out parts of that task for you. 00:10:56.860 --> 00:11:08.779 It can research products, it can compare options, it remembers your preferences, it's very personalized, visits websites for you, and then it can potentially take action on your behalf. 00:11:09.100 --> 00:11:15.019 Profound found that Muse maintains a shopping profile containing preferences by product category. 00:11:15.259 --> 00:11:30.300 So when someone makes a shopping request, Muse checks the profile, it can search its memory for additional context, it browses merchant websites and also searches meta product catalogs before filtering those candidates into recommendations. 00:11:30.700 --> 00:11:34.139 Now, here's where it gets interesting for marketers. 00:11:34.620 --> 00:11:39.180 Muse and Instinct can visit websites live using a browser. 00:11:39.340 --> 00:11:43.580 So they're not simply relying on previously indexed content. 00:11:43.899 --> 00:11:53.580 Profound says they navigate using the accessibility tree, so essentially the structured description the browser creates from your HTML and accessibility labels. 00:11:53.899 --> 00:12:00.220 Semantic HTML and clear labels can therefore help these agents understand how to navigate a website. 00:12:00.460 --> 00:12:14.620 In Profound's small exploratory muse test, and I want to emphasize small here because it was 50 prompts in one retail category, the browser typically visited three domains and returned four products from two merchant sites. 00:12:14.940 --> 00:12:21.180 Meta's product catalog separately returned roughly 60 products from about 30 merchant sites. 00:12:21.340 --> 00:12:24.700 Those candidates were then filtered down into recommendations. 00:12:25.259 --> 00:12:29.820 This is where I think we need to start expanding the AI visibility conversation. 00:12:30.220 --> 00:12:34.139 We have spent the last little bit asking: can AI find my brand? 00:12:34.220 --> 00:12:35.659 Does it understand it? 00:12:35.899 --> 00:12:37.259 Does it cite my content? 00:12:37.420 --> 00:12:38.779 Does it recommend me? 00:12:39.019 --> 00:12:43.340 And they've said this before, and I will say it until I am blue in the face. 00:12:43.659 --> 00:12:48.860 The next question is: can the agent actually interact with my brand? 00:12:49.180 --> 00:12:56.860 If I had to dust off my crystal ball, I would say we are headed to agentic commerce maybe much faster than we realize. 00:12:57.259 --> 00:13:01.820 It might still take a couple of months, but I think that's where we're headed. 00:13:02.139 --> 00:13:11.340 And this is important because being visible to an AI system and being usable by an AI agent are not necessarily the same problem. 00:13:11.899 --> 00:13:19.500 I had a really interesting experience with Muse recently that I'm still digging into and I want to kind of mention it here. 00:13:19.899 --> 00:13:29.100 I had a conversation with it about how it was navigating websites, because sometimes if you just interrogate these AI chatbots enough, they will tell you how they work. 00:13:29.340 --> 00:13:31.659 Muse sung like a canary. 00:13:32.700 --> 00:13:40.139 But but I had this conversation with it about how it was navigating websites and some of the friction it was encountering. 00:13:40.379 --> 00:13:53.820 Several days later, like after I had stopped talking about the subject, it was almost actually two weeks later, Muse proactively sent me a notification with a newly published article related to that earlier conversation. 00:13:54.300 --> 00:14:00.860 I kind of stared at my phone for a minute, I'm not even gonna lie to you, because that was a very different discovery experience. 00:14:01.100 --> 00:14:08.539 I didn't search again, I didn't ask another question, I actually really haven't touched Muse since that first experiment. 00:14:09.019 --> 00:14:15.180 But the agent remembered something that I cared about, it encountered new information, and then it brought it back to me. 00:14:15.420 --> 00:14:25.180 Now, again, I want to do more digging before I make any sweeping claims about how exactly that process works, but think about what that behavior could mean. 00:14:25.420 --> 00:14:28.139 Like today I searched for information when I need it. 00:14:28.379 --> 00:14:33.820 Tomorrow I may tell an agent what I care about once, and then it just keeps looking for me. 00:14:34.060 --> 00:14:39.580 That changes discovery, it changes shopping, and it really changes research. 00:14:39.820 --> 00:14:48.139 I'm over here thinking about how I can use it for my own my own research with the podcast and the news updates and all of the things. 00:14:48.300 --> 00:14:52.300 But for marketers, it changes what it means to be findable. 00:14:53.420 --> 00:15:07.180 We are moving from optimizing for a moment of search toward creating an information environment that both people and their agents can continuously discover, understand, verify, and potentially act on. 00:15:07.420 --> 00:15:19.259 And I still think that we are very ugly in figuring out what that actually requires or what it means, which absolutely means I'm gonna keep poking at these personal assistants until I understand them better. 00:15:19.500 --> 00:15:22.300 And of course, I will report back what I learned. 00:15:22.539 --> 00:15:25.019 So, quick recap before we wrap up. 00:15:25.259 --> 00:15:30.620 Google is putting more emphasis on the quality and structure of actual information on the page. 00:15:30.860 --> 00:15:40.060 It's also making it very clear that you cannot manufacture authority by inventing experts or creating the appearance of expertise that isn't really there. 00:15:40.620 --> 00:15:43.980 At the same time, we're seeing consumer behavior shift. 00:15:44.220 --> 00:15:54.460 More people are using AI for local discovery, but they're still verifying what those systems tell them across reviews, search, social media, and all these other sources out there. 00:15:54.700 --> 00:15:57.740 And then personal agents take this one step further. 00:15:57.980 --> 00:16:10.139 They're not just answering questions, they can browse websites, remember your preferences, research on your behalf, and potentially bring back information to us without us initiating another search. 00:16:10.300 --> 00:16:15.100 And sometimes that can be as much as two weeks later, probably longer, time will tell. 00:16:15.340 --> 00:16:29.100 When we put all that together, I think the bigger story this week is that AI visibility is becoming less about optimizing for a single answer and more about the entire information environment around your brand. 00:16:29.340 --> 00:16:46.139 Your content needs to be understandable, the authority needs to be real, the information about you across the web needs to hold up when someone or their AI agent goes looking, and your website may need to work for agents as well as humans. 00:16:46.379 --> 00:16:57.259 There is still quite a bit that we do not know about how all of this develops or how it's gonna take shape, but I think we are seeing a very clear direction here. 00:16:57.500 --> 00:17:00.620 And with that, that's it for today's news update. 00:17:00.779 --> 00:17:05.180 If you want more of this kind of analysis, you can subscribe to the visibility report. 00:17:05.339 --> 00:17:17.740 I send it every Friday with what I'm seeing in AI surge, discovery, and really the buyer journey, including the experiments in the rabbit holes that don't always make it into the podcast or anywhere else that I'm talking about it. 00:17:17.899 --> 00:17:20.700 I will be back on Tuesday with another guest episode. 00:17:20.859 --> 00:17:23.180 Until then, stay visible.