BẢN CHÉP ÂM THANH 🔗
WEBVTT
00:00:41.620 --> 00:00:51.140
We have spent a lot of time talking about how brands get mentioned in AI search, but I've been thinking a little more about what happens after that mention.
00:00:51.380 --> 00:00:57.700
Because getting recommended by Chat GPT or showing up in a Google AI overview, that's great.
00:00:57.939 --> 00:01:01.060
But does someone actually click through to the website?
00:01:01.380 --> 00:01:04.659
Can they buy your product directly through an AI assistant?
00:01:04.900 --> 00:01:09.939
And if they do visit your website, can you even track where that traffic came from?
00:01:10.180 --> 00:01:13.140
And that's really the theme of today's episode.
00:01:13.460 --> 00:01:18.099
Google is making some changes to how it displays source links in AI overviews.
00:01:18.340 --> 00:01:23.539
Walmart just launched a new way for AI agents to shop on behalf of customers.
00:01:23.780 --> 00:01:28.500
And we have an update on that chat GPT tracking issue that I talked about last week.
00:01:28.740 --> 00:01:30.340
Hey, I am Cassie Clark.
00:01:30.420 --> 00:01:39.060
I'm an AI Search Visibility Consultant, and this is Found in AI, where we talk about what's happening in AI Search and what all this means for brands trying to get found.
00:01:39.300 --> 00:01:41.219
Today is October 22nd.
00:01:41.460 --> 00:01:47.539
I do have a bit of a cold this morning, so my apologies if I sound rough, but let's get into it.
00:01:56.420 --> 00:02:07.379
On October 15th, Google announced that it's moving source links higher in AI overviews in AI mode and making those links appear more frequently throughout AI generated answers.
00:02:07.620 --> 00:02:12.819
Now, this is interesting because we've been talking about the zero click problem for a bit now.
00:02:13.219 --> 00:02:22.659
Google can pull information from your website, they can summarize it in an AI overview, and then give someone everything they need without that person ever visiting your website.
00:02:22.900 --> 00:02:28.740
And if you're a publisher or brand that's investing in content, that's a pretty big problem.
00:02:29.219 --> 00:02:32.420
So Google's trying to make those source links more visible.
00:02:32.659 --> 00:02:35.140
On the surface, that sounds like a good thing.
00:02:35.300 --> 00:02:39.300
If someone can see the source more easily, maybe they're more likely to click on it.
00:02:39.539 --> 00:02:41.300
But I do have a question about this.
00:02:41.539 --> 00:02:43.539
Are people actually going to click?
00:02:43.860 --> 00:02:54.019
Google has not shared any click-through data showing that these changes are sending more traffic to publishers, and I think that's a pretty important part of this announcement that we really shouldn't overlook.
00:02:54.340 --> 00:03:00.979
We can make all the changes that we want to know how our links are appearing inside of AI answers.
00:03:01.219 --> 00:03:07.859
But until we have the data showing that people are actually using them, we don't know how much of a difference this actually makes.
00:03:08.099 --> 00:03:13.859
Now, the other part of this announcement that I thought was really interesting was preferred sources.
00:03:14.340 --> 00:03:20.419
Google says usage of preferred sources has increased more than seven times since January.
00:03:20.819 --> 00:03:31.379
If you're not familiar with it, preferred sources allows users to select the publishers they want, and they can see those publishers more often in certain Google search experiences.
00:03:32.659 --> 00:03:36.019
From a brand marketing perspective, I think that's interesting.
00:03:36.259 --> 00:03:44.579
We've spent the last year or so trying to figure out how to get AI engines to recognize brands and then retrieve their content and then cite them in the answer.
00:03:44.740 --> 00:03:49.859
But now we're also talking about people actively choosing which sources they trust.
00:03:50.259 --> 00:03:53.539
That kind of changes the conversation just a bit.
00:03:53.780 --> 00:03:54.500
Think about it.
00:03:54.659 --> 00:04:02.900
If someone already knows your brand, trusts your content, and wants to hear what you have to say, they are more likely to seek you out.
00:04:03.219 --> 00:04:11.139
Now I do want to be clear that preferred sources isn't the same thing as choosing which websites appear in every generated answer.
00:04:11.379 --> 00:04:15.699
And Google hasn't given us the actual number of people, you know, using it.
00:04:15.939 --> 00:04:22.660
But I do think this reinforces something that I have been saying across the podcast and everywhere I've been talking about this.
00:04:22.980 --> 00:04:27.460
Brand marketing is becoming an important part of AI search visibility.
00:04:27.699 --> 00:04:32.819
It's not enough to just create good content that an AI engine can retrieve.
00:04:33.139 --> 00:04:38.180
You also need people to recognize your brand and trust the information that you're putting out.
00:04:38.420 --> 00:04:44.019
And, you know, also take the initiative to do something like select you as a preferred source.
00:04:44.340 --> 00:04:50.180
For now, I'd keep my eye on this one and see what happens to organic traffic as Google rolls out these changes.
00:04:50.259 --> 00:04:56.420
And if you're a publisher, I'd also be thinking about how you're building an audience that wants to hear from you specifically.
00:04:56.740 --> 00:04:58.819
Okay, moving on to Walmart.
00:04:58.899 --> 00:05:02.740
And I think this one is a pretty big deal for where AI Commerce is heading.
00:05:02.980 --> 00:05:10.339
On October 21st, so just yesterday, Walmart launched something called Agent Gateway.
00:05:10.659 --> 00:05:20.899
Essentially, it's an MCP server that allows approved AI agents to search Walmart's inventory and complete purchases on behalf of the shopper.
00:05:21.300 --> 00:05:31.620
If you're not familiar with MCP, it stands for Model Context Protocol, and it's a way for AI applications to
Load full transcript
connect with outside tools and systems.
00:05:31.860 --> 00:05:40.659
So in this case, an AI assistant can connect with Walmart's shopping infrastructure, look for products, and help complete a transaction.
00:05:41.139 --> 00:05:44.980
Think about how different that is from how we normally shop online.
00:05:45.540 --> 00:05:52.500
Right now, if you use an AI agent or an AI engine for this, you might ask ChatGPT to recommend a product.
00:05:52.740 --> 00:05:58.420
It'll give you a few options, maybe some links, and then you go to the retailer's website to make the purchase.
00:05:58.740 --> 00:06:08.100
Or if you're like me and you are a Walmart shopper, you just go straight to their website or their app and then spend forever scrolling through results to find what you want to buy.
00:06:08.420 --> 00:06:14.100
But with something like Walmart's Agent Gateway, the process can happen through the AI assistant.
00:06:14.339 --> 00:06:18.980
And next month, Walmart is planning to roll out something called agent profiles.
00:06:19.540 --> 00:06:26.579
This is where shoppers can authorize different AI assistants separately and control what those assistants are allowed to do.
00:06:27.379 --> 00:06:38.420
For example, you could require approval before every purchase, or you could give an assistant a spending limit, or you could give it a repeatable shopping list for your groceries.
00:06:38.659 --> 00:06:42.659
Each authorized assistant would also have its own order history.
00:06:42.980 --> 00:07:04.819
And if we think about AI assisted shopping now as we know it, where an agent or an engine gives a recommendation and then it's on the user to go do the rest compared to what Walmart is doing, there's a really big difference between asking an AI assistant to recommend something and then giving it permission to spend your money.
00:07:05.060 --> 00:07:07.939
But here's the part that really caught my attention.
00:07:08.180 --> 00:07:17.860
While Walmart is building a way for AI agents to access its shopping systems, Amazon has been blocking certain third-party AI shopping agents.
00:07:18.100 --> 00:07:23.219
So we're seeing two very different approaches to authentic commerce.
00:07:23.699 --> 00:07:31.060
Amazon is restricting access, while Walmart is creating a way for approved agents to come in and just complete those transactions.
00:07:31.300 --> 00:07:37.300
I don't think Walmart's approach should be confused with just letting any AI bot access to its website.
00:07:37.540 --> 00:07:42.740
They're building controls around which agents can connect and what those agents can actually do.
00:07:42.980 --> 00:07:51.540
Which brings me back to something I have been talking about recently, which is the difference between AI search visibility and agent readiness.
00:07:51.860 --> 00:08:04.579
Because let's say that you are a retailer, you've invested in GEO, your products are showing up in Chat GPT recommendations, you are getting mentioned when people ask
for the best products in your category.
00:08:04.819 --> 00:08:05.939
That's great.
00:08:06.259 --> 00:08:12.019
But what happens when that same customer asks their AI assistant to go ahead and buy the product?
00:08:12.339 --> 00:08:14.500
Can the assistant actually do that?
00:08:14.740 --> 00:08:17.300
Does it have access to your inventory?
00:08:17.620 --> 00:08:19.860
Does it know whether the product is available?
00:08:20.100 --> 00:08:22.180
Can it complete the transaction?
00:08:22.579 --> 00:08:27.219
Because those are entirely different questions from whether your brand is just being recommended.
00:08:27.540 --> 00:08:31.939
I think this is something that more businesses are going to have to start thinking about.
00:08:32.180 --> 00:08:42.500
We are so focused right now on getting brands into AI-generated answers, and understandably so, but AI agents are starting to do more than just answer those questions.
00:08:42.820 --> 00:08:54.100
And if that continues, the way brands make their products and services accessible to these systems is going to matter just as much as whether they're being recommended in the first place.
00:08:54.500 --> 00:09:03.300
However, if you listen to the Agentic Commerce episode with AJ Gergic, he predicted that Agentic Commerce would increase this holiday shopping season.
00:09:03.540 --> 00:09:05.700
I think we are starting to see that.
00:09:06.100 --> 00:09:11.860
But I don't think we're at a point where every business needs to run out and build an MCP server tomorrow.
00:09:12.100 --> 00:09:17.700
But I do think it is something to start paying attention to how these systems interact with your business.
00:09:18.179 --> 00:09:24.899
Okay, last update to give you, and this one is actually pretty funny because I talked about it on last week's episode.
00:09:25.139 --> 00:09:32.899
If you listen to that news update, you'll remember that I mentioned chat GPT referral traffic was suddenly disappearing from analytics.
00:09:33.299 --> 00:09:38.899
Well, within just a few hours of that episode going live, OpenAI rolled out a fix.
00:09:38.980 --> 00:09:42.899
So apparently all I needed to do was just complain about it on the podcast.
00:09:43.220 --> 00:09:46.179
I'm kidding, obviously, but the timing was pretty funny.
00:09:46.340 --> 00:09:48.100
Anyway, here's what happened.
00:09:48.419 --> 00:09:56.980
After the GPT-6 rollout on October 7th, ChatGPT stopped adding the UTM parameter to outbound links.
00:09:57.220 --> 00:10:01.220
And at the same time, referral information was also missing.
00:10:01.460 --> 00:10:11.620
So between October 8th and October 15th, traffic that would normally have been attributed to ChatGPT was showing up as direct traffic instead.
00:10:11.860 --> 00:10:23.539
Now if you were looking at your analytics during that week, you might have noticed that Chat GPT traffic suddenly dropped, and maybe direct traffic increased around the same time.
00:10:23.779 --> 00:10:29.379
But without
knowing about this bug, you might have just assumed that something had changed with your visibility.
00:10:29.700 --> 00:10:39.220
Adobe looked at traffic across roughly 500 pages and found that ChatGPT attributed visits had dropped by about 31,000 per day.
00:10:39.539 --> 00:10:45.299
Direct traffic on those same pages increased by approximately 26,000 visits per day.
00:10:45.539 --> 00:10:47.379
That's a pretty significant change.
00:10:47.460 --> 00:10:53.700
And while those numbers don't match exactly, they give us a pretty good indication of what was happening.
00:10:54.019 --> 00:10:59.220
OpenAI confirmed that it was a bug and the fix started rolling out on October 15th.
00:10:59.379 --> 00:11:01.860
So the UTM parameter is back.
00:11:02.419 --> 00:11:04.659
But as always, there's a catch.
00:11:04.899 --> 00:11:18.740
The referral information is reportedly still missing on most Chat GPT clicks, which means if you are relying on referral data alone to measure traffic from Chat GPT, you could still be missing some of those visits.
00:11:19.779 --> 00:11:22.100
So what does this mean for marketers?
00:11:22.500 --> 00:11:29.460
First, I'd go back into GA4 and look at what happened between October 8th and October 15th.
00:11:29.779 --> 00:11:36.340
If you saw a drop in Chat GPT traffic during that period, make a note of the bug in your reporting.
00:11:36.580 --> 00:11:42.100
And if you saw an increase in direct traffic around the same time, I think that's also worth looking into.
00:11:42.340 --> 00:11:47.220
Now I am not saying that every direct visit during that week came from ChatGPT.
00:11:47.620 --> 00:11:50.500
We just don't have the data to make that assumption.
00:11:50.740 --> 00:11:55.779
But we do know that a tracking issue was causing some of those visits to be misattributed.
00:11:56.019 --> 00:12:02.580
And I think this is a really good reminder of why we need to be careful about how we were measuring AI search performance.
00:12:02.899 --> 00:12:06.899
Imagine you are reporting to your results for a client or your brand.
00:12:07.139 --> 00:12:14.340
Maybe you've been tracking referral traffic from ChatGPT and now suddenly that number drops by 30 or 40%.
00:12:15.139 --> 00:12:20.580
If you don't know this is a bug, you just might assume that the strategy isn't working anymore.
00:12:20.899 --> 00:12:24.500
Which means maybe you start making changes to your content.
00:12:24.740 --> 00:12:29.379
Maybe you start questioning whether ChatGPT is still recommending your brand.
00:12:29.700 --> 00:12:36.259
When the problem in reality just might be that it has absolutely nothing to do with your visibility.
00:12:36.500 --> 00:12:38.340
It's just a tracking issue.
00:12:38.580 --> 00:12:45.059
That's why I don't really love relying on a single metric to determine whether an AI search visibility strategy is working.
00:12:45.379 --> 00:12:53.860
We also need to look at whether the brands are being mentioned, whether they are being
cited, and how often they are appearing in relevant recommendations.
00:12:54.179 --> 00:12:59.700
And even then, we have to remember that these platforms are changing constantly.
00:12:59.939 --> 00:13:10.659
Sometimes those changes affect how brands show up, sometimes they affect how users interact with the results, and sometimes apparently they just break our analytics for a week.
00:13:10.980 --> 00:13:26.820
So if you are looking over your October performance, keep that October 8th through October 15th window in mind, and make sure you are looking at the chat GPT UTM parameter rather than depending entirely on referral information.
00:13:27.379 --> 00:13:28.740
Okay, that's it.
00:13:28.820 --> 00:13:30.259
Let's close this one out.
00:13:30.500 --> 00:13:40.419
I think the interesting thing about all three updates this week is that we are starting to see more attention on what happens after a brand gets mentioned in AI search.
00:13:40.899 --> 00:13:46.580
Google is trying to make it easier for people to click through to the sources behind AI-generated answers.
00:13:46.899 --> 00:13:51.939
Walmart is building infrastructure that allows AI agents to actually buy products.
00:13:52.419 --> 00:14:00.980
And ChatGPT's tracking issue is reminding us that even when people do click through, we might not be measuring that traffic correctly.
00:14:01.220 --> 00:14:04.899
And I think that's where this GEO conversation needs to go next.
00:14:05.059 --> 00:14:10.500
We need to understand what people are actually doing after they see those recommendations.
00:14:11.139 --> 00:14:21.460
Because at the end of the day, we are not investing in AI Search visibility just to see our brand name appear in a chat GPT or an AI-generated answer.
00:14:21.700 --> 00:14:24.419
We want that visibility to lead to something.
00:14:24.659 --> 00:14:31.379
And I think these three stories are a pretty good reminder that there's a lot more to AI Search than just getting cited.
00:14:31.700 --> 00:14:34.179
That's all I've got for this week's news episode.
00:14:34.340 --> 00:14:39.379
If you want to keep up with what's happening in AI Search, make sure you're subscribed to Found in AI.
00:14:39.539 --> 00:14:50.179
And if you'd like more of my thoughts on what all this means for your content and marketing strategy and all the new research that's coming out, you can subscribe to my weekly newsletter, the visibility report.
00:14:50.340 --> 00:14:51.939
The link is in the show notes.
00:14:52.179 --> 00:14:54.980
And one final quick reminder before I let you go.
00:14:55.139 --> 00:14:59.139
I am hosting a free live training on November 5th at 1 p.m.
00:14:59.299 --> 00:15:00.259
Eastern Time.
00:15:00.419 --> 00:15:03.539
It's called How to Build an AI Search Visibility Strategy.
00:15:03.700 --> 00:15:13.299
If you've been trying to figure out where to start with DEO, how to measure whether your brand is actually showing up in AI Search and what you should be prioritizing, this training is for you.
00:15:13.379 --> 00:15:14.980
And again, it is a free one.
00:15:15.220 --> 00:15:23.620
We're going to talk about how to establish a baseline, how to identify gaps in your AI visibility, and then figure out what makes sense to work on first.
00:15:23.860 --> 00:15:26.659
I'll leave time for questions too, so if you have them, break them.
00:15:26.820 --> 00:15:27.939
It's all completely free.
00:15:28.019 --> 00:15:32.419
You can register at Cassie Clark Marketing.com/slash AI Search Training.
00:15:32.580 --> 00:15:34.659
I'll also put the link in the show notes.
00:15:34.899 --> 00:15:37.220
Okay, now really, that's it for this update.
00:15:37.379 --> 00:15:39.139
Thanks for listening to Found and AI.
00:15:39.299 --> 00:15:40.659
I will see you on the next one.
00:15:40.820 --> 00:15:43.220
Until then, stay visible.