WEBVTT

00:07:23.000 --> 00:07:29.000
I'll, like, that alignment of the response and the topic

00:07:29.000 --> 00:07:32.000
grow to be with the ad.

00:07:32.000 --> 00:07:33.000
even just

00:07:33.000 --> 00:07:48.000
looking at these within the platform, whenever we first started running with that research at the beginning of July, we were constantly saying, this ad does not match up with the topic of the prompt, of the response, and there was just like

00:07:48.000 --> 00:08:08.000
Just seemed very fragmented experiences. But then a month after that, when we really started to monitor these, we saw that ChatGPT released an update in their audience targeting through audience lists. And we were trying to find an example of a fragmented experience for our update to the blog post

00:08:08.000 --> 00:08:15.000
Because it aligned with what we were seeing and we really didn't see as many that were

00:08:15.000 --> 00:08:33.000
Far off from the topic and the ad moving forward. So along with all the other elements of the ads, the advertiser is API and metrics, I think one thing to really like monitor as ChatGPT works to maybe get closer in this audience targeting is just

00:08:33.000 --> 00:08:43.000
See how… how closely that aligns with either your client or what you're looking at personally, in case it's an audience list targeting instead

00:08:43.000 --> 00:08:58.000
Yeah, that's such a good call. And that's relatively new too. I don't think that was capable or that was a capability of ChatGPT's ads platform in the last episode of the signal. But now, yeah, you can upload a lot of granular audience data

00:08:58.000 --> 00:09:14.000
You also have a lot more insight into conversions that are coming from the platform, more so than when this initially launched. I think that's going to make it easier. If you've been on the fence, if you've been like, man, I would love to test this, but there's no way my CFO is going to approve this

00:09:14.000 --> 00:09:27.000
I think with some of these additional features and capabilities, you're getting closer and closer to ChatGPT being a channel that can drive conversions, quality conversions.

00:09:27.000 --> 00:09:41.000
All right, related to that, our last, the last homework assignment we gave was on the topic of measuring GEO. And our stance on this has been there are really not great metrics to measure GEO. This is a really

00:09:41.000 --> 00:09:57.000
new emerging space, and the the advice I tend to give is you've got to go out on a limb here. You've got to pick maybe the the least imperfect metric you can think of that you can get your stakeholders to buy in on

00:09:57.000 --> 00:10:12.000
Because I would argue the more important than picking the right metric is starting to benchmark your data and your activity so you can see as models are released, as you're testing new tactics and strategies

00:10:12.000 --> 00:10:25.000
Is it working or is it not working? So you don't want to spend a year debating what metric to choose and then be like, well, we did a whole bunch of stuff, we have no idea if it worked, because we didn't benchmark any data, we didn't land on any metric.

00:10:25.000 --> 00:10:37.000
So my question for the audience is, is there a metric you have thought of that you wish you could measure? Something that you are uniquely looking at for measuring the impact of the work

00:10:37.000 --> 00:10:42.000
the resourcing you're putting to geo.

00:10:42.000 --> 00:10:44.000
Let's see

00:10:44.000 --> 00:10:55.000
We got some insights from the Sierra marketing team on the query fan out piece. We're seeing a lot of brands in query fan out. This is really interesting. Instead of the query fan outs being a lot of

00:10:55.000 --> 00:10:57.000
You know.

00:10:57.000 --> 00:11:16.000
kind of nascent search terms, you know, like, what's… it's like best geo agency. We're starting to see that shift where more brands are mentioned, whether it's explicit, like, Bain AI Search, or we'll see the site operators, too, you know, site colon bane.com looking for AI content

00:11:16.000 --> 00:11:20.000
So that's interesting.

00:11:20.000 --> 00:11:26.000
All right, you know what? School is still out for summer. The homework

00:11:26.000 --> 00:11:40.000
We're going to put a pin in the homework. And you let us know if you want a homework. Those who wanted homework, you let us know if you want it done in a different way. But this was helpful. Thank you for everybody who shared so far.

00:11:40.000 --> 00:11:57.000
Yeah, Rob, Leslie in the chat just mentioned that he's used a Google Chrome extension. There's a lot of interesting tools. You can hack something together if you're more scrappy, if you've got support from a dev, or you can buy the data keywords everywhere also has a pretty good

00:11:57.000 --> 00:12:10.000
A pretty good Chrome extension that works on top of ChatGPT. And you can buy the data, right? I think that's been our approach, Nick, to buy the data right from ChatGPT is a little expensive, though.

00:12:10.000 --> 00:12:27.000
Yeah, we've primarily been testing either 5.5 with different levels of reasoning to see how those fan outs along with responses citations change over that. It can get, of course, expenses expensive as you choose higher reasoning models

00:12:27.000 --> 00:12:36.000
Just a heads up, I think 5.6 Luna was just released a while back, though, and that's one of their cost-controlled models to check out some of this too

00:12:36.000 --> 00:12:39.000
Nice, nice. Awesome.

00:12:39.000 --> 00:12:49.000
All right, so let's move forward past July's homework. We want to highlight a question we got from a longtime listener.

00:12:49.000 --> 00:13:00.000
And that question was about using AI to create content. And they said, I have some concerns utilizing AI for a large amount of content slash new page creation

00:13:00.000 --> 00:13:18.000
And they asked, first of all, does Google still screen for AI content and penalize, or have they accepted that this is now the norm? And number two, is it possible new content pages could negatively impact the site if Google determines slash cares if this is AI content

00:13:18.000 --> 00:13:24.000
So my POV on this and what I have seen the most is

00:13:24.000 --> 00:13:39.000
AI is going to intrinsically be a part of the technology we use as marketers. And there are ways to use these tools in a way that is very low quality and very easy to identify. And there's ways to use these tools

00:13:39.000 --> 00:13:47.000
in a way that's very high quality. And so Google's stance has been, look, it's not about using AI, it's about the quality of the content.

00:13:47.000 --> 00:14:03.000
That said, just in the last couple of days, as we were preparing for this webinar, we saw a ton of chatter in our SEER Slack channels about something that Claude just launched. So this, I think, might be top of mind

00:14:03.000 --> 00:14:19.000
For many of you on the call, just yesterday, Claude said they in compliance with the EU AI Act, they are going to be watermarking content generated through Claude. And there's a few different ways that they're doing this

00:14:19.000 --> 00:14:23.000
There's a lot of details here in Nuance. So first of all

00:14:23.000 --> 00:14:41.000
Claude models launched in the EU on or after August 2nd are going to support this machine readable marking at launch. If you're not an overly technical person, just assume, you know, the way I think about it is there's some invisible stuff happening when you're copying and pasting

00:14:41.000 --> 00:14:57.000
text from Claude. And what they say is, look, it's probably going to transfer. It's possible if you copy and paste very selectively, or if you're really splitting up the text that's generated through AI, it's possible that the watermark doesn't carry

00:14:57.000 --> 00:15:13.000
Within hours already, we saw third-party tool providers coming out with, like, watermark checkers. So expect that that's going to be part of it. It's a little easier for images created as well, because it's kind of baked into metadata. It's easier to say

00:15:13.000 --> 00:15:30.000
But the high level here is that so many people have convinced their stakeholders, convinced their teams that AI should be a part of your content creation process. And I imagine, especially if you have a skeptical exec team, if you've got a skeptical

00:15:30.000 --> 00:15:49.000
team you're working with and they see an update like this, it's like, all right, are we not going to do this anymore, right? Is this inherently bad? So I want to touch on that. This is the heart kind of what we're talking about here, the embedded watermarks in text. I think that's the biggest thing

00:15:49.000 --> 00:16:05.000
The other thing we're reacting to is graphite did a really interesting study that basically said, look, it's like a 50-50 split at this point between content on the internet that's written by humans versus content on the internet that's generated by AI

00:16:05.000 --> 00:16:14.000
And so that's concerning. It's also concerning to see the upward trend of AI-generated content, the downward trend of human-generated content

00:16:14.000 --> 00:16:23.000
What I would say about this, though, is that this is taking a very binary view at something that I think deserves a lot more of a nuanced

00:16:23.000 --> 00:16:26.000
a nuanced opinion on

00:16:26.000 --> 00:16:40.000
And what we hear from our clients a lot and prospects is, look, if there's an easy button, why shouldn't we press it? And there are great tools. And I like the AirOps team is great. There's a lot of tools out there that

00:16:40.000 --> 00:16:55.000
are hosting these case studies are presenting their tool as, hey, if you've got a lot of pressure coming downward and you need to drive traffic, you need to drive citations, which I would argue citations is not the metric you should be looking at

00:16:55.000 --> 00:16:59.000
But you need to move the needle, whatever the needle may be

00:16:59.000 --> 00:17:06.000
Our tool can help you do it. And for some teams, that is true. For some brands, that could be true.

00:17:06.000 --> 00:17:22.000
But Lily Ray, who we are huge fans of here from AMCIV, Lily Ray published what I think is maybe the most important piece of research this year about this topic. And she dug into over 200 websites

00:17:22.000 --> 00:17:37.000
That followed this pattern of identifying what works. So it could be a listicle, it could be programmatic content, identify what works, and then find the footprint. And by the way, it's very easy to find these footprints once you know what you're looking for

00:17:37.000 --> 00:17:45.000
And what she found is, yes, initially, you see a big spike. The impact is real. The performance goes up.

00:17:45.000 --> 00:17:51.000
But more than half of the sites she analyzed lost 30% or more of their peak organic traffic

00:17:51.000 --> 00:18:08.000
Almost 40% lost half or more of that traffic, and one in 5 sites lost 75% or more of their traffic. And so if you are a new brand, and you are under a ton of pressure to drive performance, I get why you would go in this direction

00:18:08.000 --> 00:18:26.000
But if your brand has authority, if you have a really strong footprint in Google organic search and you've built this momentum, you've built this foundation, you have a lot more at risk than a company that was founded, you know, 12, 18 months ago

00:18:26.000 --> 00:18:37.000
So your mileage may vary. This could still be a viable strategy for many of the brands, especially the brands we tend to work with at Sear. It is a very high risk strategy

00:18:37.000 --> 00:18:54.000
And another piece of this research that Lily Ray generated said there are eight high risk templates to recognize in the pitch. If you're being pitched and you're being promised the world, here's all of the case studies that prove this tool is going to make you

00:18:54.000 --> 00:19:09.000
the star of of your marketing program. Here are the templates that are most obvious to recognize in that pitch. And none of these strategies are inherently bad, by the way. Comparison pages can be great. FAQs can be great

00:19:09.000 --> 00:19:26.000
I would argue off-topic content that might be a bridge too far. That that kind of thing is like the HubSpot caught a lot of flack for that. I think last year they had created so much content, things like what to write in an anniversary card for your coworker

00:19:26.000 --> 00:19:41.000
It's like, do we do we need a piece of content for that? For a while they ranked, they drove traffic, and then you saw the dip as more and more zero click searches became prevalent, and more of those queries became answerable right within Google's results

00:19:41.000 --> 00:19:44.000
But again, all of these templates work

00:19:44.000 --> 00:19:55.000
But it's all about how you're going about it. And if you're applying a system and building a system around this, or if you're just trying to press an easy button, there is no easy button in marketing.

00:19:55.000 --> 00:20:12.000
Our POV is everything that makes AI generated content trustworthy happens before and after the tool. The tool is just that. It is a tool that is part of a system. It is not the most important part of it. But for a sophisticated marketing program

00:20:12.000 --> 00:20:28.000
It could be a critical part of it, and so we don't want to throw the baby out with the bathwater. We don't want to say, hey, we're too worried about what these clawed watermarks, let's not, let's go back to writing content manually. I don't think that is the approach.

00:20:28.000 --> 00:20:48.000
If you are a fan of SmarterX or the marketing AI Institute like we are, Paul rates are recently dropped this framework. And I believe it's going to be his keynote at Macon this year. By the way, Sir is going to be present at MECON. Will is going to be speaking there. It's a really, really

00:20:48.000 --> 00:21:02.000
event for AI forward marketers gives you a really good idea, kind of a benchmark of where you are relative to other marketing teams, which I think can be maybe one of the most useful things you get from the event.

00:21:02.000 --> 00:21:13.000
But I digress. What Paul shared here are these eight pillars that an organization needs to embrace in order to be an AI forward team.

00:21:13.000 --> 00:21:18.000
Technology is just one piece, and I really, really like this because it is so true

00:21:18.000 --> 00:21:28.000
The technology you're using, it is so fun. It's wildly impressive, and it's changing all the time, and there's so much mindshare that we put to it

00:21:28.000 --> 00:21:44.000
But the technology without unique data really is it ends up creating something that all of your competitors can also create. You know, if your UVP is that you write the best prompts, eventually somebody's going to write a better prompt than you

00:21:44.000 --> 00:22:00.000
But if you've got unique data, if you've got this top down vision for how your organization is using AI, if you've got the right governance so teams understand what they can and can't do, what they should and shouldn't do, if you're investing in literacy

00:22:00.000 --> 00:22:15.000
So that even your most junior team members can use the system and create great content and not end up with AI slop. That's where the magic happens. That's the value. And so that's the framework and the mindset I would encourage you all to be in.

00:22:15.000 --> 00:22:19.000
All that said, yes, technology is just one of eight pieces.

00:22:19.000 --> 00:22:36.000
But we also don't want to ignore the upside. Like, don't get it wrong, we are really, really excited and passionate about the value this technology can contribute to marketing. And I've been thinking about this and sitting with us a lot the past few weeks, and this idea keeps coming to mind for me

00:22:36.000 --> 00:22:53.000
Historically, I've been doing SEO for over 15 years. And the way we've always identified content to create has been what are the keywords that have enough volume? And you can make your model more fancy. You can say, okay, we're going to look at more signals, we're

00:22:53.000 --> 00:23:01.000
We're gonna apply conversion data, page search conversion data. So it's not just about monthly search volume. There's other metrics we can use

00:23:01.000 --> 00:23:15.000
But that approach is always going to scratch the surface. It's always going to optimize for quote unquote, the average customer. And there is no average customer. There is no average patient, right? Everybody has their own unique needs.

00:23:15.000 --> 00:23:30.000
So the upside of AI and being able to create, let's just call a spade a spade, you can create a much higher volume of content. And if you're doing it with the right system, that content can be highly valuable. And you can get closer to this bottom line

00:23:30.000 --> 00:23:46.000
Of being able to answer every question that comes with the decision of do I purchase from this brand or do I not purchase from this brand? I don't think you max out. I don't think we ever get to 100%, but I think we can get closer and we can add more value. We can solve more problems

00:23:46.000 --> 00:23:51.000
We can leverage all of our experience. Will and I were just riffing on this earlier today

00:23:51.000 --> 00:24:08.000
We have so much content on the SEER website. We share so much, but it pales in comparison to the Sierra slack. And obviously, we're not going to outsource the entire Slack. But we started to think about what are those byproducts where if I'm having a conversation, if we're riffing on something back and forth

00:24:08.000 --> 00:24:25.000
A few people read it, they learn from it. That's valuable. But what else are we doing with that information? That's the kind of mindset you should be in of, okay, how do I use all of these resources, use them as inputs in a system. AI can help synthesize, can help create the briefs, can help create the outlines

00:24:25.000 --> 00:24:40.000
can even help draft the content as long as you've got your copywriters, your your SMEs in the mix reviewing the content, making sure it is as positive as useful, as valuable as possible.

00:24:40.000 --> 00:24:56.000
All right, and it's very convenient that we invited Nick on here because again, nobody has done more research on this than Nick. So one of the questions we had is like, all right, if we know what not to do, do we have any insight on what we should do

00:24:56.000 --> 00:24:58.000
So, Nick, what do you think about that?

00:24:58.000 --> 00:25:13.000
Yeah, so like Alisa mentioned with those templates that are at risk with AI search along with other metrics that you're going to be monitoring, a lot of these templates aren't inherently bad, like listicles

00:25:13.000 --> 00:25:30.000
People like listicles. I know that if I'm going to be shopping for my family this holiday season and I'm searching for best presents for my mom, like this year, there's probably going to be a site that comes up and it's going to give me some like very solid options. I like the way that those are formatted.

00:25:30.000 --> 00:25:35.000
AI likes the way that those are formatted too, which is why there's a cross there's benefit

00:25:35.000 --> 00:25:40.000
They're… they are beneficial in both sides of this. So

00:25:40.000 --> 00:25:58.000
Throughout, whenever we were looking at a lot of this research, and we saw that first, yes, overall listicles were declining in usage across AI search as citations. But then there were four primary themes that we saw within some of these what we classified as, quote, surviving listicles

00:25:58.000 --> 00:26:13.000
that we're actually still gaining citation share and still being utilized across the different AI platforms, primarily ChatGPT in this instance. The first one was content freshness. I know that content freshness has been very well documented

00:26:13.000 --> 00:26:23.000
by this point, we have a lot of good research from Sunny on our team that is really digging into a lot of how content freshness and recency

00:26:23.000 --> 00:26:38.000
can really impact AI search, and this is all the same. I just gave the example of I'm looking for this holiday season. I don't really care for what was beneficial in 2024, what I should have bought my mom 2 years ago. I'm probably too late to the point on that

00:26:38.000 --> 00:26:41.000
So content freshness is very important

00:26:41.000 --> 00:26:58.000
Here. The second side of it is comprehensiveness and some of the aspects that we saw, the listicles that were consistently declining in citations were commonly just using 5 to 8 different options, products, or brands in

00:26:58.000 --> 00:27:16.000
lists. They were also very thin in content, just giving a really quick description on maybe the potential value of one. And whenever there was a self-serving list tool, it was primarily that number one option that was promoting itself had… was pretty comprehensive, and then all the other options were pretty thin.

00:27:16.000 --> 00:27:32.000
The surviving listicles consistently had pages of 10 to 20 options on average, and they were also much more complex. They were having a lot more pros and cons, and why this is actually impactful to different audiences

00:27:32.000 --> 00:27:41.000
The other side of this is that data validation and methodology, just trust signals and backing up the claims with verifiable information

00:27:41.000 --> 00:28:01.000
For a lot of those self-promoting listicles, if you actually put together a full methodology of we're taking together all of these different inputs to create this list that is backed by data, a lot of times your own brand isn't going to be mentioned as the best option. And so really providing some of those trust signals are what we found were impactful as well

00:28:01.000 --> 00:28:24.000
And then lastly, just structured format. This is table stakes, and just making sure that the list and the content is very easily accessible and readable by AI crawlers, and just a great user experience for the users that are going to be engaging with the content and landing on that page, hopefully, as well.

00:28:24.000 --> 00:28:25.000
Awesome.

00:28:25.000 --> 00:28:27.000
And

00:28:27.000 --> 00:28:28.000
Yeah.

00:28:28.000 --> 00:28:29.000
Thank you, Flatmeet.

00:28:29.000 --> 00:28:30.000
That's good stuff.

00:28:30.000 --> 00:28:33.000
Yeah, I mean, I would underscore

00:28:33.000 --> 00:28:48.000
Within within everything Nick just said, I think if you think about the type of marketing program that represents, that's a marketing program that you should have conviction in. AI can still be a part of generating all of this

00:28:48.000 --> 00:29:04.000
But all of it is still in service of your audience, in service of your customer, if you are creating content in the way they want to ingest it, to Nick's point, we're people, we love listicles, we love things that are organized, we love things that are easy to digest

00:29:04.000 --> 00:29:18.000
So inherently, listicles are not bad. It's just the system in which they are created at times results in really low quality results. And that's not what we're looking for.

00:29:18.000 --> 00:29:32.000
All right, so we're going to assign… how about I don't call it homework? And maybe in September, we'll feel better about homework, and we'll figure out a better cadence of sharing, a better method of sharing. So two things we recommend that you do this month.

00:29:32.000 --> 00:29:47.000
First of all, list the unique inputs you currently include in your content creation workflow, and what else can you add? What I would challenge you to do is not create a new thing. Don't go to your team and say, hey, let's start doing this thing. Think about what is already happening

00:29:47.000 --> 00:30:04.000
And what I would remind you is if you have access to subject matter experts, that is a unique input and that's a unique input that's already a part or could be a part of your resourcing plan. So those unique inputs, you know, you've got to talk to these people

00:30:04.000 --> 00:30:23.000
Right? You've got to have conversations with them. Maybe you have so many, it's difficult to scale those conversations. Again, AI can play a role in that. We've been doing a lot of interesting stuff. Some of you have heard of what we call expert echo. We've been doing a lot of interesting stuff with voice AI, and it has advanced greatly

00:30:23.000 --> 00:30:28.000
In the last 6 to 12 months, there's a lot you can do

00:30:28.000 --> 00:30:43.000
in a low effort way with your marketing team to extract those insights from your experts. The other thing is what I mentioned earlier. Think about slack. We share so much great information. People challenge each other. People

00:30:43.000 --> 00:30:55.000
You know, read things and say, hey, what do you all think about this? And right now, that is such a valuable resource for us at SEER that we honestly don't do anything with.

00:30:55.000 --> 00:31:10.000
Right? And there's so much more we could do with that. So that's the mindset to be in, right? What are the unique inputs that already exist today? How do we take a look at the metaphorical factory floor and see what are the byproducts? How do we try to sweep those up

00:31:10.000 --> 00:31:14.000
and create a system that makes use of them.

00:31:14.000 --> 00:31:35.000
And then the number two assignment we've got for you, review your top 10, your 10 most recently published pages. How many of them match Lily Rae's high risk templates and think about how you created those. If you created this type of content, but you sourced unique quotes, unique insights from your SMEs

00:31:35.000 --> 00:31:41.000
You populated this with unique data that only you have access to

00:31:41.000 --> 00:31:43.000
I think you're probably in a good spot

00:31:43.000 --> 00:32:06.000
But if the secret sauce to your content generation program is, you know, a collection of prompts that works really well today, that is not going to work very well in the future. And you've got to think about, what am I doing? What am I creating that makes this a more compelling experience than somebody just going to ChatGPT or Claude and asking the questions themselves

00:32:06.000 --> 00:32:21.000
All right, so with that, if you have not yet, let me stop my share here. If you have not yet submitted a question and you've got one, please submit one now. I'm going to start from the top here because we already had a few

00:32:21.000 --> 00:32:37.000
So we had one question. Claude has said they are keeping their key secret at the moment that identifying the watermark without that key is impossible. Do you use a checker? Do you trust it? Right now, the team's not using a checker

00:32:37.000 --> 00:32:58.000
In large part because this news is like, you know, 22 hours old, or whatever it is. But I will say the team immediately, some of our scrappier team members were immediately like, hey, I found this thing. I found a checker already. I am sure there is a way to game this system, and I'm also sure it is not worth your team's time to try to game the system

00:32:58.000 --> 00:33:13.000
What I would say is you always want to understand what is your risk level? So if you have been creating content with Claude, and I know a lot of you are, maybe you have a program that you're like, hey, I've got high conviction in this program. We are creating really valuable content

00:33:13.000 --> 00:33:30.000
I'm still nervous about this cloth thing. I think it does behoove you to try to dig in and see, is there any impact to rankings or visibility, whether that is your share of voice in ChatGPT, whether it's your visibility in Google search

00:33:30.000 --> 00:33:39.000
Is there any correlation between content that you created with clawed models released after August 2nd

00:33:39.000 --> 00:33:54.000
And I say that and at the same, I don't even think there are Claude models released after August 2nd. I could be wrong on that. I think this is more of a forward looking thing. But my point in all of this is you want to understand what your risk level is

00:33:54.000 --> 00:34:04.000
So to the extent it's possible, it's probably a correlation, not a causation thing, you want to try to dig in, you want to try to explore what the… your level of risk is

00:34:04.000 --> 00:34:10.000
But if you have high conviction in your overall program, if you're using unique insights

00:34:10.000 --> 00:34:13.000
you should feel confident in that.

00:34:13.000 --> 00:34:28.000
Okay. And thank you for that, Jessica. Yes, Opus 5 and the most recent one came out July 24th. So this is if you have executives who are panicking, that is a really good caveat to say. Like, yes, this already launched

00:34:28.000 --> 00:34:36.000
No, it does not necessarily impact the content we have created prior to today with with Claude content

00:34:36.000 --> 00:34:52.000
All right, the next question we have is they had a question on the graphite study that we shared. They said, what is the definition of AI generated, AI assisted, human edited? This is why I really don't like the

00:34:52.000 --> 00:35:11.000
That binary thought process, right? Because I totally agree. The answer to answer your question, and definitely check out Graphite's site and check out the whole study. They used a tool basically they used Surfer's AI detector with a chunk size of 500 words

00:35:11.000 --> 00:35:18.000
I always think about Ethan Malik, who is one of my favorite people in the space. A lot of you are fans of his

00:35:18.000 --> 00:35:34.000
He's consistently said that any kind of AI detector is if it's looking for patterns, it's it's always going to be. It's never going to be foolproof, and especially for content created by non-English, non-native English

00:35:34.000 --> 00:35:47.000
speaking people. And so I always take that with a grain of salt, right? But if you want to run an analysis like that, you have to… you've got to use something. So again, it's it's probably a correlation versus causation thing.

00:35:47.000 --> 00:36:09.000
All right, I want to tip the next question for you, Nick, because nobody I know has done more query fan out research than you. And this question says they are seeing searches with meaningful impression counts in Google Search Console that look like query fanouts to me. It's things like George Fox University founded 189

00:36:09.000 --> 00:36:23.000
Comma site colon Georgefox.edu. Maybe that's how the kids are searching these days. I don't think it is. So, Nick, what do you think? Is this query fan evidence of query fan outs in Search Console?

00:36:23.000 --> 00:36:41.000
I think that's really scrappy, and to look at it that way and kind of find those patterns. It hasn't been an input that we've looked at consistently, but now it's something that I want to go into our Search Console and look at immediately. So I think it could be evidence. One thing that I would look at to

00:36:41.000 --> 00:37:01.000
try to align those patterns are with some of the releases of these new GPT models. So, as we go from, whenever GPT-5.5 became the default in ChatGPT, which was at the beginning of May, is that when we're starting to see an increase of these site searches in Google Search Console

00:37:01.000 --> 00:37:11.000
That could be a pattern. And whenever 5.6 becomes the default, I know that it was released, but I don't believe it's the default for the platform yet.

00:37:11.000 --> 00:37:31.000
Are we going to see another increase in the site searches in our own data that we've pulled from the API, we know that site search was used about 55% more often in 5.6 compared to 5.5. So it could definitely be be a signal. I would try to line it up with some of those model releases just to see if

00:37:31.000 --> 00:37:47.000
That aligns because, yeah, we've been trying to find other ways to pull query fan outs. This would also layer in the aspect of, you know, the impression side of it and seeing how often some of these are searched. You can see the themes of those

00:37:47.000 --> 00:37:51.000
As well, so yeah, something… something for us to dig into.

00:37:51.000 --> 00:38:07.000
Yeah, Nick, as you're talking through that to plus one to all of that, I think there's also an interesting opportunity to try, again, think about everything we are doing as having byproducts of value, and where my head went with that is

00:38:07.000 --> 00:38:17.000
If you are seeing a sharp spike, let's use the example Rob shared in the chat. Big spike of these types of query fanouts on June 11th.

00:38:17.000 --> 00:38:26.000
That's an interesting signal that you have an audience who's using the paid version of ChatGPT. And

00:38:26.000 --> 00:38:46.000
That state… like, first of all, just sit with that statement, because you might be now wondering, does my audience use primarily the paid version of ChatGPT, or are they mostly using the free version? This is part of the sometimes it feels impossible how fragmented these surfaces are that people are searching on

00:38:46.000 --> 00:39:08.000
Where even if you know you've got the data, you know you pulled something from SparkToro, you know your audience is using ChatGPT, maybe you've asked them yourselves or you're seeing leads come in from ChatGPT. Even if that's true, you don't necessarily know which model they are most often engaging with. And these models vary really significantly, as you can see with this shift with the query fan out

00:39:08.000 --> 00:39:18.000
Right? That one shift can really significantly change the brands that are appearing for a prompt

00:39:18.000 --> 00:39:34.000
And if you ask the same exact prompt in the free version of ChatGPT, maybe you're getting a totally different experience. So I mentioned that because we're all getting hit up by executives or, you know, our clients are telling us, hey, our CEO is searching XYZ, you know.

00:39:34.000 --> 00:39:46.000
But what's old is new again. We're back in the early days of SEO and it's really difficult to answer and diagnose those scenarios when there are so many layers to the surfaces people are using

00:39:46.000 --> 00:39:47.000
All right.

00:39:47.000 --> 00:39:48.000
Perfect

00:39:48.000 --> 00:40:12.000
Yeah, and to that point, you mentioned that we're using SparkToro to see what platforms they're using, but just going back to that, your point on how different the experiences can be within a singular platform, it's the query fan outs, or if they're using the free version that introduces the ads to a much higher degree than others. So there's going to be two very different experiences depending on which version of ChatGPT.

00:40:12.000 --> 00:40:14.000
Yep, absolutely.

00:40:14.000 --> 00:40:30.000
All right, we have time for one last question, and it is on content freshness. So Nick, I'm going to give this one to you as well. The question we were asked is when we talk about content freshness, how fresh are we talking

00:40:30.000 --> 00:40:35.000
How often should site content be updated?

00:40:35.000 --> 00:40:50.000
So, I'm going to give Sonny another shout out for this one with her post that was just updated last month. And in her post, she found that 75% of the pages that LLMs were citing were updated within the past year. So

00:40:50.000 --> 00:41:07.000
That's an overarching number to look for. But then there's also, I think, the case of, you know, looking at this by industry and by vertical, because there are just going to be some aspects where it matters a little bit more for this industry compared to another one

00:41:07.000 --> 00:41:23.000
And so, it's definitely going to be the traditional SEO answer of it depends, and it does. But I think this is where it's important to just dig into your own data and see which pages of yours are getting cited, are getting hit by AI

00:41:23.000 --> 00:41:41.000
crawlers the most, and when were those updated? What's the impact if you take a set of pages and refresh them? What changes from there? So I think that there is going to be some sort of flywheel that needs to get incorporated depending on what you find, because in some cases it may need to be

00:41:41.000 --> 00:41:52.000
6 months in some cases it may need to be six years, but it's just one of the cases where you don't know until you kind of dig into the data yourself.

00:41:52.000 --> 00:41:54.000
Yep, that's perfect.

00:41:54.000 --> 00:42:13.000
All right, that's the last question we have. If you have more questions for us, please check out those emails that you're getting from our marketing team. We have so much fun with these and we want to make sure that we are answering your burning questions and we're talking about the things you all really want to hear about

00:42:13.000 --> 00:42:26.000
And like to give you all a few minutes to decompress, we have the full hour booked, but our intention is always to give you a little time between to go touch grass, go

00:42:26.000 --> 00:42:42.000
Do your emails and your Slack responses if you must, but have a great rest of your day. Thank you all for joining. Thank you, Nick, for being our special guest host. And we'll see you next month.

00:42:42.000 --> 00:42:43.000
Thanks, everybody.

00:42:43.000 --> 00:42:45.000
Bye, everybody