Latest / Found in AI: AI Search Visibility, SEO, & GEO / Google AI Mode Monitoring + Why AI Citations Don’t Equal Recommendations

Transcript

Google AI Mode Monitoring + Why AI Citations Don’t Equal Recommendations · Found in AI: AI Search Visibility, SEO, & GEO

• 0:41
  SPEAKER_00: Hey, welcome back to Found in AI, the podcast where we talk about what's changing in AI Search, GEO, and how brands get recommended and found inside AI engines.
• 0:53
  SPEAKER_00: I am Cassie Clark, an AI Search Visibility Consultant.
• 0:56
  SPEAKER_00: Today is somehow October 1st, I am still surprised by it.
• 1:01
  SPEAKER_00: And we have four stories today with a bit of a theme running through all of them.
• 1:05
  SPEAKER_00: The way people search is getting more complicated.
• 1:08
  SPEAKER_00: We are moving beyond typing a question into a box and getting Tempo links back, which obviously we've been moving away from for a while and talking about it here on the show.
• 1:19
  SPEAKER_00: But this week we have some really good examples of just how much bigger search is becoming.
• 1:25
  SPEAKER_00: We're talking about AI that keeps searching after you leave, visual searches that finally have their own reporting in Search Console, new research into what AI systems actually need to retrieve in order to answer a question well, and a study showing a pretty big difference between brands that get cited and brands that it just actually ultimately recommends.
• 1:49
  SPEAKER_00: So let's get into it.
• 2:06
  SPEAKER_00: So instead of going back to Google over and over again to repeat the same search, users can essentially give Google an assignment and tell it to keep looking for them.
• 2:15
  SPEAKER_00: This is very, very similar to Meta's Muse that I covered last week.
• 2:20
  SPEAKER_00: I mean, I suppose Meta is the one that's copycatting Google here since Google's been working on this for a while, but you get the point.
• 2:28
  SPEAKER_00: Google's AI system can monitor information across the web, and that includes websites, forums, social platforms, real-time sources, and Google's shopping graph, and then update the user when it finds something relevant.
• 2:42
  SPEAKER_00: So let's say that you're waiting for a particular product to come back in stock, or maybe you're watching prices.
• 2:49
  SPEAKER_00: You don't necessarily have to conduct another search.
• 2:53
  SPEAKER_00: Google can keep doing it for you.
• 2:56
  SPEAKER_00: I think that really creates an interesting shift for marketers because we usually think about search visibility as just a single moment in time.
• 3:06
  SPEAKER_00: Someone has a need, they search Google, Google gives them the answer.
• 3:10
  SPEAKER_00: We try to appear somewhere in that process.
• 3:14
  SPEAKER_00: But what happens when the search just keeps going?
• 3:18
  SPEAKER_00: Now your content potentially needs to be useful not just when someone initially asks the question, but when Google comes back looking for updated information later.
• 3:28
  SPEAKER_00: And that makes freshness even more important.
• 3:31
  SPEAKER_00: I've talked about freshness quite a bit on this podcast because we've seen just how quickly updated information can affect retrieval in AI search.
• 3:41
  SPEAKER_00: This is just another reason brands need to stop thinking about content as something they publish once and then leave their sitting collecting dust forever and ever.
• 3:51
  SPEAKER_00: If AI engines are actively monitoring the web for changes, clearly communicating that something has changed becomes part of that visibility.
• 4:00
  SPEAKER_00: I think that we are going to see more of this, especially as these AI chatbots or agents become a little bit more sophisticated.
• 4:08
  SPEAKER_00: Search doesn't necessarily end when someone closes the browser anymore.
• 4:13
  SPEAKER_00: If that agent is running, it appears that the search just keeps going until it finds what it's looking for.
• 4:20
  SPEAKER_00: Next, I want to talk about a new study of AI shopping results because one finding in there is something that I think marketers really need to pay attention to.
• 4:30
  SPEAKER_00: New research from Graphite looked at nearly half a million AI shopping responses, and that's across Chat GPT and Google.
• 4:38
  SPEAKER_00: And large retailers had a pretty substantial advantage here.
• 4:42
  SPEAKER_00: Depending on whether Web Search was involved, major national retailers accounted for roughly half or more of the top recommendations.
• 4:51
  SPEAKER_00: But here's the part that I found really interesting.
• 4:54
  SPEAKER_00: Large retailers and smaller retailers each represented roughly 38% of the stores that were cited.
• 5:03
  SPEAKER_00: But large retailers appeared as the top recommendation about two and a half times as often.
• 5:11
  SPEAKER_00: So in other words, being cited by an AI system and then being recommended by that same AI system are not the same thing.
• 5:19
  SPEAKER_00: I've been saying that for a while now, that citation presence is useful, but it cannot be the only thing that we're measuring.
• 5:26
  SPEAKER_00: A model can retrieve your website, use your information, cite you as a source, and then still recommend someone else or another brand.
• 5:35
  SPEAKER_00: Those are very different outcomes.
• 5:37
  SPEAKER_00: The study, it just gives us a really nice example of that happening in e-commerce and shopping now.
• 5:44
  SPEAKER_00: Graphite also found something interesting too about the way that prompts changed those recommendations.
• 5:51
  SPEAKER_00: When users specifically ask for an independent retailer, keyword there, independent, smaller stores showed up much more often.
• 6:00
  SPEAKER_00: But terms like local or near me, which we use in traditional Google, didn't create nearly the same shift.
• 6:09
  SPEAKER_00: And I think that really gets into something else that we don't talk enough about with AI visibility, and that's how these models understand attributes about a business.
• 6:19
  SPEAKER_00: It's really not enough for your business to just simply exist online.
• 6:24
  SPEAKER_00: You also need to ask: does the model understand what kind of business you are?
• 6:30
  SPEAKER_00: Does it understand who you serve?
• 6:32
  SPEAKER_00: Does it understand what makes you relevant to specific criteria the user included in their prompt?
• 6:38
  SPEAKER_00: I think that's why prompt tracking needs to go a bit deeper than running just one generic query and then checking whether your brand appears.
• 6:46
  SPEAKER_00: Different languages can produce very different recommendation environments as well.
• 6:52
  SPEAKER_00: So if you are measuring AI visibility, you really need to know whether you're simply present in the information being retrieved or actually making it into that recommendation.
• 7:04
  SPEAKER_00: Those are not interchangeable KPIs.
• 7:08
  SPEAKER_00: Our third update comes from Google Search Console.
• 7:12
  SPEAKER_00: Google has added a new multimodal search filter to its performance reporting, and that includes searches that begin through Google Lens, circle to search, images uploaded directly to Google Search, and Chrome's search this image feature.
• 7:29
  SPEAKER_00: And I really like this update because search behavior has been moving beyond typed keywords for a long time.
• 7:37
  SPEAKER_00: But marketers haven't always had great visibility into what that behavior actually looks like.
• 7:43
  SPEAKER_00: Someone can see a pair of shoes they like and then use lens to find them, or they can circle something on their phone and then find it that way.
• 7:51
  SPEAKER_00: There is still intent there and discovery is still happening, but there may never be a traditional typed query that we are used to just tracking.
• 8:01
  SPEAKER_00: Now though, if those searches are driving traffic to your website, Google is giving you a way to actually isolate and analyze some of that activity.
• 8:11
  SPEAKER_00: And importantly, for this show, Google says multimodal data can appear in both its regular search performance reporting and its generative AI features reporting.
• 8:22
  SPEAKER_00: I think this is another reminder of that keywords are becoming only just one piece of understanding search behavior.
• 8:29
  SPEAKER_00: People can type or they can point cameras at things, and now AI is just sitting in the middle interpreting all of that.
• 8:37
  SPEAKER_00: So for marketers, the takeaway on this is just pretty simple.
• 8:41
  SPEAKER_00: If you have access to that reporting, go look at it, especially if you're e-commerce, travel, local search, home, fashion, or any other category where visual discovery makes sense.
• 8:54
  SPEAKER_00: We finally have another window into how people are actually finding things.
• 8:58
  SPEAKER_00: So if your report says anything helpful, take that into consideration and use it in your strategy.
• 9:04
  SPEAKER_00: And finally, Perplexity published some new research this week around something called contextual embeddings.
• 9:12
  SPEAKER_00: This one gets pretty technical pretty quickly.
• 9:16
  SPEAKER_00: So I'm not gonna drag you through the entire paper or try to pronounce everything at four in the morning when I'm recording this.
• 9:22
  SPEAKER_00: Thank you, cat, for waking me up that early.
• 9:25
  SPEAKER_00: Um but the interesting part for marketers is the problem that they are trying to solve.
• 9:31
  SPEAKER_00: A lot of retrieval systems are essentially trying to find the passage that's most relevant to the user's question.
• 9:39
  SPEAKER_00: Perplexity calls this the gold passage.
• 9:42
  SPEAKER_00: The problem is that the passage containing the direct answer isn't necessarily all the information an AI system needs in order to give someone a good answer.
• 9:54
  SPEAKER_00: It may need surrounding context or supporting information from somewhere else in the document, or it just may need additional passages that help explain or verify what the first passage actually says.
• 10:07
  SPEAKER_00: So Perplexity has been working on contextual embeddings designed to retrieve both the directly relevant passage and the additional context needed to answer the question.
• 10:18
  SPEAKER_00: And again, the paper gets really technical, but I think there's a content lesson buried inside of all of this.
• 10:25
  SPEAKER_00: We've spent the last couple of years watching people try to optimize content for AI by breaking everything into tiny self-contained answer blocks.
• 10:34
  SPEAKER_00: That term chunking comes to mind.
• 10:37
  SPEAKER_00: Really hate that term.
• 10:39
  SPEAKER_00: Um, and clear structure does matter.
• 10:42
  SPEAKER_00: Like it really does matter if you get to the point in the passage.
• 10:46
  SPEAKER_00: I've said this about structure approximately eight million times, but structure is not the same thing as just stripping away context.
• 10:56
  SPEAKER_00: AI engines still need enough information to understand what you're saying, why it's true, how the concepts relate, and when an answer applies.
• 11:05
  SPEAKER_00: So the goal here should not be to create the smallest possible chunk an AI can just extract.
• 11:13
  SPEAKER_00: You still need that supporting context in your content.
• 11:17
  SPEAKER_00: So the goal here should be to just create content that is easy to understand and retrieve without removing the information that makes the answer useful in the first place.
• 11:27
  SPEAKER_00: Perplexity's research is just another indication that retrieval itself is also just evolving in that direction.
• 11:36
  SPEAKER_00: So that's where I'm gonna leave you this week.
• 11:39
  SPEAKER_00: Google is letting searches continue in the background.
• 11:42
  SPEAKER_00: We are getting better visibility into visual search behavior.
• 11:46
  SPEAKER_00: Perplexity is working on retrieving context instead of just the passage containing an answer.
• 11:52
  SPEAKER_00: And shopping research is showing us that appearing in an AI system sources doesn't necessarily mean you're the brand that gets recommended.
• 12:00
  SPEAKER_00: That's why I keep coming back to the same thing.
• 12:03
  SPEAKER_00: Yeah, visibility is not just one single metric.
• 12:07
  SPEAKER_00: You have to understand how these systems retrieve information.
• 12:10
  SPEAKER_00: You need to understand how they understand your brand.
• 12:14
  SPEAKER_00: And you also need to understand what happens between being found and actually being recommended.
• 12:20
  SPEAKER_00: Okay, that's all I got for you today.
• 12:22
  SPEAKER_00: If you found this episode helpful, make sure you're subscribed to Found and AI wherever you listen to your podcast.
• 12:29
  SPEAKER_00: And if you want these kind of updates in your inbox, you can subscribe to the visibility report where I break down what's changing in AI search and what marketers actually need to pay attention to that goes out to emails and inboxes every Friday.
• 12:43
  SPEAKER_00: I'll be back Tuesday with another guest episode.
• 12:46
  SPEAKER_00: Until then, stay visible.