Are you a marketer struggling to make sense of GA4? You’re not alone! In this insightful webinar, analytics expert Dana DiTomaso breaks down what marketers still get wrong about Google Analytics 4 – and why AI tools can’t magically fix your data headaches.
Discover how to avoid common GA4 pitfalls, build reports that actually matter, and turn your analytics from a source of frustration into a powerful decision-making tool. Whether you’re in-house or at an agency, you’ll walk away with practical tips to audit your setup, focus your energy, and finally feel confident in your data.
Key Takeaways
- GA4 is a powerful tool, but requires proper setup; don’t expect AI to fix a bad configuration.
- Universal Analytics and GA4 are fundamentally different; treat GA4 as a new tool, not an upgrade.
- Many sites receive traffic from AI tools (ChatGPT, Perplexity, Gemini); create a dedicated “AI Tools” channel in GA4 to track this.
- Use Google Tag Manager for advanced tracking and to capture meaningful user interactions.
- AI can help uncover patterns in your data, but it’s not a substitute for human expertise or statistical reasoning.
- Always filter out low-volume and irrelevant data before analysis to avoid misleading insights.
- Protect user privacy: never upload sensitive or personal data to AI tools, and ensure your privacy policy covers third-party processors.
- Document your analytics processes and outcomes for future reference and team alignment.
- Use explorations in GA4 for deeper insights, and always provide business context when prompting AI for analysis.
- Consent management is critical; ensure your cookie banners and consent tools are properly implemented and connected to your analytics.
Episode Transcript:
Emanuel: [00:00:00] Hi, and welcome everyone to another. How about marketing session? It’s my pleasure and privilege to have a special guest this evening. Evening, because it’s 7:00 PM Eastern. Dana DiTomaso, we’re just gonna give just one minute to get everyone a chance to join. We already have eight people, nine people in our session.
If you are kind enough to simply say hello. There’s a comments section there, so if you wanna drop a hello and tell us where you are joining us from, that will be I would be curious to know. Thank you so much. Everyone who has registered for this, for the ones that are watching the recording, for whatever reason, they couldn’t, wouldn’t join us tonight. We have Brian from Toronto. Delia from Toronto. I’m from Toronto as well. Where are you joining us from? [00:01:00] Dana?
Dana: I live on Vancouver Island, so I’m about 45 minutes outside Victoria, British Columbia.
Emanuel: That’s a beautiful place. I’ve never been there myself, but I know for sure.
Dana: Actually grew up in Hamilton, near Toronto.
Yep. Lived in Burlington for a while. Went to Western, lived in Toronto for a while, so grew up in southern Ontario, but then moved to Edmonton for 10 years and now living on the island and the island is incredible.
Emanuel: I heard, and it’s on my list sooner rather than later. Hello from Toronto. Sara Yasmine from Toronto, and Kevin, our friend from Arkansas.
We have 11 people joining. I think we are good to start. I’m just gonna double check agAIn to see if we are streaming everywhere else. And apparently we are, , not somehow because Streamy Yard and all these tools, we all know and we can read. They have a mind on of their own.
Hello from Oakville!
Dana: Oh, hey. Perfect. Yep. See, I was just saying, I went to high school in Burlington right next door.
Emanuel: My name is Emanuel. I’m the [00:02:00] founder of, how about some marketing? Thank you for being here. How about marketing?
It’s a community. It’s a place for marketers to get better at their marketing. We have webinars. Soon we’ll have podcasts. We have a newsletter. And many things to come. Thank you for being here, and thank you for showing interest in our activity. Now, I do have a thing, I wouldn’t be much of a marketer if I wouldn’t ask everyone here to go to.
How about marketing.com and sign up for the newsletter if you aren’t already. That’s where you’ll find this recording, previous recordings, future events, and all the information that I’ll try me and my team to put together for you to become better at your marketing. The reason I started this project is because I had questions and I also was educating many of the business owners that they were like paying for some services, but didn’t quite understand.
So I’ve created this as a resource I wish I had. Thank you so much. [00:03:00] Now I like to give a shout out to all of our friends and our partners, Aspiration Marketing Joachim, who has been with us since the beginning. Kevin Carney Organic Growth, he’s backstage. He has an incredible tool. Toronto Marketing Hub, Level Up Media, Global Center for FinTech Innovation Canada and Toronto B2B marketing community. Many of our audience tonight are joining from these communities. Thank you so much there. Last, before we dive right in, I do wanna remind everyone that we have a comment section.
It is a privilege to have Dana here. She’s the leading authority when it comes to GA 4 trAIning and resources. I would, which your permission, take advantage of this opportunity to get, to ask her the tough questions that all of us marketers had. That being sAId, this is my pleasure and privilege to introduce Dana DiTomaso.
So I don’t think anything else is to be added, but then adjust a couple of words [00:04:00] before you start the presentation and I pass you the virtual stage.
Dana: Yeah, absolutely. Yeah. Hi everyone, Dana DiTomaso. As we mentioned in the beginning, I live on Vancouver Island now, but I did grow up near Toronto and my whole family is still there.
So very much enjoy Toronto. So one of the things that I do a lot with is GA4 now, but I actually started working in SEO 25 years ago and what I find is that I became a GA4 expert because GA4 happened as I was getting into analytics, and so I had the opportunity to get really deep into GA4.
GA4 is a great tool. It was not necessarily introduced well to the wider marketing world, so a lot of people are like. “I hate GA4. It’s not, my favorite tool.” It’s, it’s a tool and I think that you can get a lot out of it, but GA4 makes you do a lot of setup and where Universal Analytics held your hand and sAId, here, let me show you what I think is important,
in GA4, you can decide what’s important. And I think because of that difference, people really don’t like the tool. So that’s, [00:05:00] one of the things here, but I’m gonna be talking about today is uncovering AI insights. And using AI to uncover insights from your GA4 data. So I think that’s a really critical use of AI tools and also understanding how people are getting to your website from, AI tools. So that’s what I’m gonna be walking through in the presentation. So Emanuel, do you wanna toss the, I’m sharing my screen if you wanna toss that up. Great. Okay. So I also have my link tree link there. If you have any other questions, feel free to submit them there or put them in the chat here.
But if you’re too embarrassed, you can always message me privately. That’s okay. Also, there’s no silly questions. Please don’t tell me you’re asking a silly question. No one is born an expert to ask away for sure. What is GA4? GA4 is something that everybody hates now. It’s what I find too with GA4.
By the way, if you were a universal analytics expert, forget everything you ever knew about Universal Analytics, just approach GA4 as a brand new tool. It is not an upgrade. It is a completely new tool, and I think that really helps people as well. So one of the things I find whenever I do a GA4 audit.[00:06:00]
Every site has traffic from AI tools. Even if it’s just a few sessions, it’s definitely there. I haven’t seen a site in at least six months. It didn’t have at least some traffic and some sites like those in higher education or home services could have as much as 10% of their traffic coming to their site.
From AI tools such as Chat GPT, and Perplexity. The problem is that it’s hiding and usually it’s in the referral channel or an unassigned. So this is a screenshot of what it typically looks like. You got some Chat GPT referral, Perplexity referral. You got Gemini Chat GPT not set. That happens when there’s a UTM parameter appended from Chat GPT, and they only set the source and not the medium.
You have to set both, or else it shows up as not set. Perplexity and Chat GPT do not know how to create proper UTM tag so someone could let them know that. That’d be great. And the rest is referral because it’s a referral from their website. So it’s there already. And some people might say maybe GA4 will make an AI tools channel.
I don’t know if they will, don’t wAIt on them to do it. Maybe they will. But if they do, it’ll probably just include Gemini or something silly like that because Google loves talking about [00:07:00] Google and nothing else. So instead, what I want you to do is I want you to create a channel called AI Tools and yank all that traffic out of these referral and unassigned channels and make your own AI channel.
I have a post on my website on exactly how to do this. It takes five minutes. So it’s on how to track and report on traffic from AI tools and GA4. We can toss the link into the comments after my presentation or into the show notes later on for people who are re watching this later. And it is incredibly easy process.
I promise it only takes five minutes. I recorded a step-by-step video and that is how long it took. So definitely check that post out. The core part of it is this enormous piece of RegX, which is, if you’re unfamiliar with RegX, it’s pattern matching and it’s looking for this, all these different domAIns, and people give me all these different domAIns, and if you have other domAIns to add to the list, emAIl me and let me know. Happy to toss that on there. And you can also use this in Looker studio. Essentially what you’re saying is if the source matches the set of conditions, it’s [00:08:00] an AI tool. That’s it. Now once you’ve created this channel for AI tools, one of the things you can do in GA4 is you can also create an audience of people who have come to your site via AI tools.
You can use this audience to see if they come back agAIn later on, because I have found personally that people who come to site via AI tools are very well engaged and they’re likely to sign up. They’re likely to engage with you. They’re likely to be interested in your products or services, or join your newsletter.
And they definitely come back agAIn and agAIn. But you can tell that by creating an audience of people who come to your website, where they’re coming via AI tools at any point in time. And then you can also see how engaged they are with other traffic sources as well. And you can use ’em as a remarketing audience if you have enough of them, which would probably only be for high volume sites, but certAInly it is something you can do and as tools like Chat GPT et cetera, do run ads in the future, it’s gonna be really interesting to say, I’ve pulled these people in from these ads on this tool, I’m gonna remarket to them and then remarket to them on other tools. So I’m gonna [00:09:00] make you a to-do list. First thing you’re gonna do is you’re gonna add channel for AI tools in GA4, and then you’re gonna create an audience of visitors from those AI tools in GA4.
And of course, AI tools themselves are not the only AI features that are bringing people to your website. For example, we have AI overviews. So if you search for “how to track URLs with hash marks or hashtags in GA4” we come up twice with our YouTube video and our post, which is great, and people do actually click through, which is nice.
You won’t always get a URL that looks like this. Sometimes you will. And when you do get a URL that looks like this highlights the text on the page so you can see it’s got the URL for this particular post, and then it’s got the hashtag colon, tiled colon text equals part that tells the browser to highlight this piece of text to works in Chrome, but other browsers as well.
You’ll have that click through and it’ll show that information. And so you can also see. It starts the first five words and then the last five words of the [00:10:00] highlighted text as well. So this highlighted part is what we’re gonna grab using Google Tag Manager, and you do have to use Google Tag Manager for this, unfortunately.
But agAIn, I have a walkthrough. This one takes 10 minutes. So this post has step-by-step instructions. Code, you can copy the whole shebang. It is really easy to set up. And then once you’ve set this up using Google Tag Manager, then you can make a report or an exploration in GA4 or a report on Looker Studio of the landing pages where these text highlights were detected.
And so this is a, just a quick snapshot from our own website and then you can see which pages are returned and then what the highlighted text was. Then once you have this information, then you can combine this data with Google Search Console or other SEO tools and see what possible queries cause those AI overviews to come up.
And then you can tell, are you serving that audience well? How are people engaging with this? Of course. As always, I want you to create an audience of these users as well. And then you can also use them agAIn in remarketing. So [00:11:00] for example, if you have some pages that people are coming to there a lot and you know they’re looking for specific searches.
So for example, if I hop back to the ones that we have, you can see we have this post about Google Ads credits. Lots of people wanna know about Google ads credits. If I published a part two to this, then I could use that as a remarketing audience to say, show this to people who have looked at this particular snippet and landed on this page, for example.
And then agAIn, you also can see how engaged they are compared to other traffic sources, especially over the long haul. Of course, remember not every AI overview link includes that text highlighting and people also ask and featured snippets do use text highlighting sometimes. So it’s not perfect, what is perfect in analytics, nothing.
So instead of we’re gonna track what we can, and this is really useful stuff to track for sure. Also, I find that senior leadership who have heard about AI overviews and are desperate to know if they’re coming to the site, love seeing these reports. So I would definitely put them in for your boss, client, whoever you report to.
All right, so here’s your to-do list. I’ve added time periods in now, [00:12:00] so adding a channel, five minutes, create an audience that’s gonna take you two minutes. Uncover AIO visits that takes you 10 minutes, create a reporter exploration, 10 minutes, maybe more if you haven’t created a reporter exploration before, and then creating an audience, also two minutes. This is, something real quick you can do. We’re talking about less than an hour of work now. So once you’ve done that, now let’s turn to doing AI analysis with GA4 itself. Can AI tools do analysis for you? Technically yes you can, but I don’t know if you’re gonna have the good time you have if you just turn over AI tools to GA4’s entire dataset and say, Hey, figure it out.
Because we’ve heard talk about these mcps or Google GA4 as an MCP, and then you can attach it to something and just put in a question like, what was interesting that happened last week? Not necessarily the best use of that tool. And why is that? The problem is that most of these AI tools are large language models, and what that means is they’re built for language, not statistical reasoning.
So if you wanna learn more about why this [00:13:00] is, there’s a great post by Juliana Jackson called Large Language Models, can’t Handle all your Data Problems. And I really recommend it as a good read because it just talks about where the limitations are with LLMs. They’re great at specific things, they’re not great at other things.
And I would also recommend reading her newsletter too, because it’s really good. But definitely one of the problems with LLMs is it isn’t purpose-built statistical analysis tool. So you can’t export 20 reports from GA4, slap it into Chat GPT, and expect some sort of magic insights to be bubbled up, because your LLM will be like oh, absolutely no problem, I can work with this. It loves it. But you end up with low volume noise, formatting issues, the AI will find patterns in garbage data, because it can’t really say to you, Hey, I think this is garbage. It assumes you’re giving it great data. It is thrilled with what you give.
AIs never say no. They are the person who is overconfident for no good reason. They are the person who is, and you’ve heard of the Dunning Kruger effect. Where the incompetent people are too incompetent to realize they’re [00:14:00] incompetent. That is LLMs in a nutshell. They don’t know that they don’t know.
So because of that, what you end up with is a lot of weird data as a result and weird insights that you’re like, that doesn’t make any sense. LLMs also do not understand the concept of correlation versus causation. They’re not smart enough to do that. I’m sure you learned about this in high school, but LLM skipped that day and they really struggle with correlation versus causation.
For example, is there a left-handed epidemic that worked out, cut off, but pretend it’s an epidemic. I’m a left-handed person. Those fellow left-handers know life is weird. As a left-handed person, of course there’s not an epidemic. It’s just socially acceptable, now, to be left-handed, it didn’t always used to be.
It doesn’t mean that there’s more left-handers, it’s just so people are comfortable being left-handed, now. This is the kind of context information that would make sense to us, but an LLM might miss that if it was not in their learning data set and they would think, oh, there’s a epidemic of left-handed people.
You really need to provide the bare minimum data required for the LLM defined patterns and nothing else. [00:15:00] And notice too that I sAId patterns not analysis. This is what we’re doing with an LLM. We aren’t doing analysis. It’s pattern recognition because an LLM does pattern recognition. So what I recommend in GA4 is to have some explorations ready to go. Create explorations in particular, because I find with explorations in GA4, you get some data like entrances and exits that are only avAIlable in explorations and not in standard reports. And then once you’ve created the exploration, you can use them agAIn and agAIn to export them and put them into Chat GPT, Claude, whichever lLM, you really like.
Personally I’m a Claude person. Lots of people use Chat GPT, we call it Chad here at Kick Point. Pick the tool you like. I don’t care. So if we’re thinking about a content performance exploration, so how good is your content performing? There’s these eight pieces of information I would typically include.
For dimensions, I would look at landing page title, because that gives some nice context to what the landing page is. The session channel group, how do people get there? The session campAIgn, if there was a specific campAIgn that got people there. And then for [00:16:00] metrics, we be looking at sessions, entrances, exits, and key events.
And then really, that’s a pretty simple exploration in terms of the information that you wanna have, but you also wanna make sure you’re including a filter. So for example, I would filter out. Less than 50 sessions, for example. I would also filter out if people are doing translations of your pages. So for example, I only speak English.
I speak French in the sense that people who’ve lived in Canada for a long time speak French and it’s just from reading off of cereal boxes. It’s not real French. I do not dare write in French, but I find that a lot of people come to my site and they wanna read stuff in Spanish, which is great.
I don’t speak Spanish, I’m not gonna translate in Spanish because I hope people can’t reach out to me in that language. I can’t answer it. So one of the things that I have in my filter is that the page title is not contAIned “Como”, which I believe means how in Spanish. And as a result I’m filtering out those Spanish language pages, whatever you know is your filter for making sure that you’re filtering out translated pages. You wanna do that as [00:17:00] well. because most of my posts start with how to.
You might also need to clean it up after you export the data, after you export any exploration to CSV, open it up in Excel or Google Sheets, whatever you’re gonna use. There’s always these extra lines at the top of the exploration that’s been exported.
Make sure to remove those or else the LLM is not gonna be able to understand what’s going on here. It’s gonna be like, this is not a proper CSV, I dunno how to format it. Those five, six rows at the top really screw an LLM up, so make sure to delete those. Save it before you go ahead and export it.
Now what you’re not gonna do is you’re not gonna then upload that and say, analyze this GA4 data. Because you’re not telling it what you’re looking for. You’re not telling it what the patterns are. And you might be thinking yeah, I want it to tell me what’s interesting. You’re still asking it a question and that question is not, is this data, it’s basically what you’re asking it.
Instead, you’re gonna ask it a question. ” Based on this content performance data showing engagement rates and traffic sources for our top 25 landing pages during the last 90 days, what patterns suggest [00:18:00] opportunities for improving our organic search strategy?” I know that people are also fans of context setting.
When you do prompts, for example, you’re an SEO expert, skilled at analyzing data, et cetera, but I find that used to change the type of analysis you get. It doesn’t really change it so much anymore. LMS have figured it out now. Your experience may vary. Test that for yourself, but I find that I personally don’t need to include that anymore.
And you can see here as well, I’ve sAId patterns, not analysis, right? What you’re looking at is, here’s the data I’ve given you. You’re explaining the context of the data, you’re explaining the time period of the data. It’s almost like a reverse smart goal, right? Where you’re thinking about time-based, you’re thinking about realistic, how much data you’re providing, and I’m only giving it 25 landing pages.
I’m not overwhelming it. You’re being specific on what you wanted to think about, which is organic search, and then it can take that data and give you an analysis. Based on the patterns that it finds, and only you can provide the business context that explains these patterns. This is your job. Why are you asking this?
What are you hoping to learn? Make sure to give that context to the [00:19:00] AI at that point.
And let’s not also forget about privacy. Anything you upload to an AI tool, we all remember when people are like, oh, no Chat GPT is spewing our stuff all over the place. People don’t read. This stuff is public all of a sudden, when you share something that makes it public, even if you remove it later, that doesn’t mean it’s not public anymore.
Claude, a couple weeks ago, changed by default that they would use what you uploaded it for trAIning. You had to go in and change that manually in your settings. They communicated that, but if you’re just like yeah, whatever, click, you wouldn’t have seen that information. And just remember anything you upload to an AI tool that could be out there in the world anywhere.
So don’t share sensitive information, please and thank you. And I also want you to consider the risk level of the data that you’re uploading. So when you’re deciding what information to provide, what risk level is it at? Low. I’m only talking about low risk data here. Aggregated metrics like sessions or page views, page paths, device categories, broad channel groups.
I’m not getting too in the weeds here. [00:20:00] Medium would be something like onsite search queries that could contAIn personal information. People selecting different things in form fields or location data that’s iffy. Then high, you never wanna give personal information. Or user IDs, custom dimensions based on user data.
For example, someone’s membership level or page titles that exist behind a user login. because that could contAIn personal information. Make sure to really be looking at your data before you upload it anywhere else. You’ll also wanna confirm that your privacy policy also includes sharing data with third party processors, because there’s a lot to dig in here if you are in the European Union or if you’re working with people in the European Union, you need to make sure you have a data processing agreement with whatever AI tools you’re using.
Most people don’t do that, so also make sure that’s something you understand. From what I can see here, mostly everyone is in, canada, United States. So we don’t really think about privacy as much as people in Europe do. You should, if you’re working with European clients at all, or if you are agency siding, you work with clients who have European clients, at Kick Point Playbook, people buy our [00:21:00] courses from Europe. So I always have to be thinking about GDPR and those privacy boundaries.
And then you also wanna communicate those privacy boundaries to everyone on your team. This is not just a you thing. Anyone who has access to GA4 can potentially download this stuff and toss it into an AI system.
So it’s really important to trAIn everyone on what you can and can’t do when it comes to AI systems in your GA4 data, especially because unfortunately, there’s just no audit trAIl of downloading data from GA4. GA4 does not let you know when people have downloaded it, so it’s really up to you to make sure to communicate to everybody else not to do this or only use these explorations or don’t include this kind of information so that everyone on the team is aware of what you should not be uploading to AI.
Last thing you want is to get a big fat fine because somebody decided to upload customer data to Chat GPT.
So here is your improved to-do list. So one of the things I do recommend is checking with legal or your privacy officer about whether or not you can use AI for GA4 data. I [00:22:00] don’t know how long that’s gonna take.
Good luck talking to lawyers always takes forever, but it is something where I do recommend starting that process just to make sure that you’re allowed.
Edward had a great question. What sensitive information do you recommend not to use AI for?
Yeah, I think one of the things that people make a lot of mistakes on are things like onsite searches, especially if it’s for a site where people are looking for health information.
You wanna avoid that. And then if you do have a user login, for example, Jane App is a really popular app for booking massage appointments and physiotherapist appointments and stuff like that. In their title tags, they actually have the name, your name in the title tags. So if you have your website tracking on Jane, for example, your client’s website, then if you just grabbed all the title tags and uploaded it to AI, you’d be including everyone’s personal information.
So that is one of the things to watch out for. You also wanna check things like page parameters to make sure that there isn’t any personal information contAIned in page parameters because sometimes that could be there. So really just do a scrub of your data first to make sure there’s [00:23:00] nothing personal in there.
This is especially important for financial or health, but obviously all industries have personal information to just make sure it’s all scrubbed before you go ahead and upload it to AI.
So now that we have all these pieces in place, what I wanna go through now is your practical AI analysis framework.
So step one, of course, clean and prepare your data. You’ll end up with a much better result. And I’m gonna go through the checklist of how to clean and prepare your data. So first thing is to make sure to use an exploration that only has the dimensions and metrics that you need. As I mentioned in that example with content analysis, those four dimensions and four metrics, that’s probably enough for most content analysis, but only give it what is needed for the analysis.
Sometimes to, when I’m starting an analysis or pattern matching with AI, I might ask it: what data would you like in order to answer this question and then based on that it will give me some information or I [00:24:00] say to it: if you feel like you don’t have enough information, tell me. And then it will tell me if it thinks it’s missing some information.
Sometimes it doesn’t. But at least you’re asking it to make sure that you’re getting all the information possible from it or to it rather. You’ll also wanna make sure that you add a filter to exclude low volume data. For example, less than 50 sessions or views. 50 is just a number that I picked.
It will depend on your own site volume. What, when things start to, degrade into noise. If you have a really low volume site, it might be something like 10 sessions. If you have a very high volume site, it could be something like a thousand or 5,000 sessions. So what that number is up to you.
But really look at your own. Data to make to pick the right value for when things start to degrade into noise. Also, as I mentioned, check for translated page titles or other artifacts, and make sure to remove those from your dataset, and then export and remove that metadata at the very top of the report, or if you use Looker Studio to export these instead of explorations in GA4.
That won’t include the metadata when you export it from Looker Studio at all, so you could also use that as an [00:25:00] option, but then you’re missing out on some metrics which are only in explorations such as entrances and exits, which I do find really useful. I don’t know why they’re only available in explorations.
I wish they would fix that. I wish a lot of things about GA4. People ask me sometimes, do you like using GA4? And I just, I dunno if you know that kombucha girl gif, where she’s maybe; that’s how I feel about GA4. Sometimes I use it every single day. My job is to keep up on GA4 and sometimes I’m just so annoyed at it.
So it’s not just you if that makes you feel any better.
All right, next step we’re gonna be using context setting. So again, taking that time to set the context. Don’t ask it, just analyze this. Make sure to provide the business context and the prompt. So tell it what kind of company you’re working with.
This is data for GA4. Sorry, my cat might make a special appearance. So he’s left. Okay. He just jumped out on my desk. I know he is. He’s mad about dinner. He wants to get fed soon. What’s his name? His name is Rufus. And he is enormous, fuzzy Maine Coon cat. He’s just like a giant fuzz ball. But he’s wandered off now.
[00:26:00] If he comes back I’ll have him do a cameo. Thank you. So provide the business context and the prompt, explain what type of business it is. because that does also really help the AI. You don’t have to say you are an SEO person skilled at blah, blah, blah, blah, blah. But do say this is a B2B SaaS company that, sells this thing to developers to help them be better at their jobs or whatever it might be.
Include important considerations. For example, we have a six month sales cycle. That sort of thing is important for the LLM to know. And then also add data specifics. What exactly are you giving it? This is content performance for our top 25 pages. For example. Include time constraints as well. Make sure to say that this is from Q4 2024 and we’re excluding holiday periods.
We’re not dealing with the we’re not dealing with the Christmas period, for example. And then also ask specific questions, what pattern content patterns suggest lead quality, for example. So really trying to be as specific as possible in the information that you provide. And then the last, the first step for you do [00:27:00] anything you’ll wanna end up with this. Clean up that data, you’ll end up with that better result. So quality control. Make sure to cross reference the output that you get with that business knowledge. Is what you got from the LLM. Does it make sense based on what you know, do not act on small sample sizes?
And this is one thing that I find that is really something that leadership can really get obsessed with. Especially, I was working with an extremely large client talking like a Fortune 100 company. Huge organization. We were talking about their website and they just got obsessed about this one page that represented something like 0.25% of their overall visits.
It’s this doesn’t matter. I only care about this other thing that’s actually gonna realistically move the needle for you. But they could not get over this one tiny page. So we just stopped including it in reports. because it’s such a small sample size. It does not matter. It is a random page. It didn’t have anything to do with anything.
If anything, I could have deleted it from the website, but just don’t include those small sample sizes. Don’t let the LLM get confused or distracted. It’s like a, [00:28:00] it’s like a puppy sometimes. It’s don’t show it anything shiny or else it’ll just wander off on its own.
Emanuel: Whoever work with clients can relate.
Dana: Yeah, absolutely.
And then look for what are called confounding variables. Can something else explain this? Is there a different explanation for what it is that you’re seeing as opposed to what the LLM might be finding? And then make sure to test any recommendations before widescale adoption. So if it says to you, Hey, I think that you should change this hero image, or You should change these headings to be this, or you should create something on this.
Only make a change and say 10% of your pages. See if that helps. Before you go ahead with wide scale adoptions, and of course everyone’s bad at this. Document your processes and outcomes. because what you don’t wanna have happen is six months later you’re like, did we ever try this? What happened here? And then you’re like, oh, it didn’t work.
That’s because you tried it six months ago and it didn’t work back then either. So make sure to document those processes and outcomes. It doesn’t have to be some sort of formal process. It could just be you have a Notion and in the notion you document the stuff that you worked on. It doesn’t have to be anything fancy.
Just write [00:29:00] something down for somebody else. Especially if you’re on vacation and you wanna explain what’s happening, so someone isn’t oh, this is weird, and then they undo all your hard work while you’re away. Avoid that from happening too.
So now we have your giant to-do list. So make sure to, you’re adding your channels, you’re creating those visitor, you’re creating those audiences, you’re adding AIO visits.
You’re creating your report or exploration, you’re creating that audience of visitors from AIO, you’re checking with legal, I dunno how long it’s gonna take. And then you’re starting with one focused analysis based on this AI analysis framework. Just spend an hour doing that. Play around with it.
See what you think of it. Don’t overdo it. Just start with that one single analysis. Instead of timers, you’d end up with some sort of wild goose chase in your data or that happens to me all the time. Maybe you’re just fine. Sorry if you can hear my cat in the background. He’s decided it’s food time.
It’s not food time, so he’s gonna have to wait.
One thing I wanna remind everyone too, AI is an assistant, not a replacement for your expertise. It’s not gonna replace this anytime soon. As long as we’re bringing expertise and we’re [00:30:00] not just being checklist marketers, checking stuff off a list. We’re actually bringing insights to what it is that we’re looking at.
It can help. AI can help you find patterns that are very difficult for human brains to uncover, but that does not mean that it is a replacement for what it is and what you do. Because what you’re bringing to it is quality questions and quality data, and as a result of that, you’re gonna be getting quality insights.
The LLM is not necessarily gonna know this is the business context or this is the data you should pull. And if you give it the entirety of GA4 and say, figure it out, it’s not gonna do a great job. Your business context is what turns these patterns into those actionable strategies as well. It’s gonna give you some information.
You can act on that, but you decide what the actions are gonna be. Of course, keep privacy top of mind at all times. And if you’re ready to work on your to-do list, then I think you should definitely get started.
Emanuel: Amazing. Where should I start essentially and do like everyone to ask their questions in the chat. [00:31:00] But I do wanna make a couple of comments because that’s what I do as a host. I’m adding I’m adding my notes in my personal experience in all of these things.
First of all, I would like to thank you Dana from the entire digital marketing community for all the information that you’ve. Put out there throughout years for free at no cost, because I was laughing most of the time when you are saying these things because I learned the hard way. Many of these things I learned the hard way.
And especially GA4. GA4 is an incredible, great tool, but I wasn’t able to understand it correctly until I put on my data analyst hat. As opposed to my digital marketing or SEO background hat, they don’t work well together. But once I understand it from that perspective, things started to show up.
And obviously Dana has been tremendous help. I’m a digital marketer mostly, so I work with the paid, I work with social, with pAId social and different departments I show them. Build some custom reports. I show them the Omni Bug Chrome extension that I’ve learned from you recently, and, has simplified [00:32:00] many other Chrome extensions that I was using.
I was pulling from many places, so that helped me a lot. In terms of data privacy. I can’t stress how important data privacy is not just. Probably, I think California has some similar laws similar to Europe.
Dana: The CPA, which is focused on people not selling data. More so than, it’s not quite GDPR-esque yet, although I think it is gonna get there.
It is mostly focused on making sure that people aren’t selling data to third parties.
Emanuel: Yes. And a couple of years ago we were talking about Cambridge Analytica and the data leaks and all those things, yet. Since Chat GPT, I can’t how many people willingly share personal information about them themselves, upload their own documents, and I can totally see, because you said about the legal com compliance department, I can totally see sending them an email and then they go to charge GPT and say, Hey, somebody asked me [00:33:00] this question, and they upload that data. We have a question from Delia Can GA4 tell us about user behavior at all, or that’s what you mean, that’s what you meant by patterns?
Dana: Yeah. It depends on how you’ve got things set up.
I would say when I’m thinking about user behavior, I’m thinking about how people are engaging with your website. So GA4 by default does give us some, obviously, you know what pages people viewed. It records 90% Scroll depth. It records. If people have gotten to the end of a video, for example, but it doesn’t necessarily tell us anything beyond that.
So a lot of the extra things that I would record when I’m setting up a Google Tag Manager instance for to push data off to GA4, I’d be looking at say call to action button clicks, accordion collapses and expands. You know, how people are navigating around the website. Are they using body links?
Are they using header navigation or footer navigation? How long are they spending on each page? Some of that’s captured by GA4, but not enough of it. And so I find that really thinking about what are the things that I need to know about my site to be able [00:34:00] to tell if it’s doing the thing it’s supposed to be doing?
And then most importantly, if it didn’t do the thing it was supposed to be doing, how can I tell where people got hung up? And so in that case, what I’d be looking at is say, could they even see the call to action button? Did the form even enter the viewport? And those are the kinds of things that tell us like, oh, they couldn’t convert.
Of course they couldn’t fill out the form. because the form, they never saw the form, for example, right? Seems unfair to be like, why didn’t you convert when they couldn’t even see the thing that they were supposed to do? So those are the kinds of things that I would record through Google Tag Manager and put in GA4.
Emanuel: I believe it’s fair to say that the user will never behave as you expected, and there’s a lot of things that you can uncover and learn about that then that’s something you need to I do have a question, but let’s take Yasmin’s question. If I just open a brand new GA4 account, what are the things I should set up to get the whole picture of a 25-year-old all in one website?
Dana: Yeah, I definitely just covered some of them for [00:35:00] sure. I would say it does depend on what’s the most important thing for you to know about your website? Without knowing what the website is, it’s hard to tell exactly. But that’s when I go through the site and think about what are the business goals and then how can I attach those metrics that I’m recording to those business goals.
So if one of the goals of the website, sorry, I’m just going to yell at my cat one second here.
This is Rufuss. He’s a big fuzzy guy and he really wants dinner, and so that’s why he’s annoyed at me picking him up. But you stay up here and stop yelling at me. All right. We won’t keep you long. I know. His dinner isn’t even till six. It’s four 30 now, so this is when the parade of our dying starts.
Oh yeah. No, it’s great. Love it. Anyway. Yeah, I would say definitely looking at it in terms of how can you connect the actions that people are taking to those business metrics. So [00:36:00] what are the actions that people could take that, what, how do you make money, basically? So is that selling stuff, is that contact information?
Is it, like what is that process? And then how can you measure those events to tell, like basically is the website doing what you want it to do?
Emanuel: Okay. Hope that answer the question because you mentioned, I do have a question. I’ve, because I work in agency environment with many departments, I at one point gave up on trying to make the argument or explaining why Google Tag Manager and not simply taking the GA4 and just, put the code there as all the popular platforms have a place for GA4 and that’s it.
We’re talking the Wix and the other sites.
Dana: Yeah, I mean with GTM you can measure so much more than you can with just GA4. If you’re just getting started, you just have GA4. That’s great. Put it on there. Something is better than nothing, but you really do need Google Tag Manager in order to measure [00:37:00] extra stuff.
And so one of the courses that I have, analytics for agencies, which is on my website @kpplaybook.com in that course I have a default Google Tag Manager contAIner that contAIns everything that I typically measure by default on a website. Along with analytics plans and stuff, and that really helps people figure out like, oh, these are the things that I could, in terms of table stakes of measuring things on websites, and we always add more to that.
Thanks for sharing that link. But yeah it’s something where I find that just at least measuring these things can also help you think, oh, you know what? I’m now measuring this; now I wanna know if this is happening too. And that can start you on the path of getting beyond the basics when it comes to GA4 analysis.
Emanuel: One of my frustration is that unfortunately, I’ve seen some big spenders not measuring and not capturing the entire the entire picture. And, when you spend money, you like to get the results and, data, data is important and getting the right data is actually what’s important.
because unfortunately, from what I’ve seen from my own experience, especially [00:38:00] since the emergence of what we call AI, but I like that you highlighted there, LLMs path recognitions. We won’t go into that debate now, but since the emergence of AIs in many reports with data that is wrong or even doesn’t exist.
So we know the hallucination, especially with GPT five. I personally noticed that it tends to hallucinate more than the others did right now. The right data is important. Can paint the right picture, and you can take the actions that you need at the end of the day to bring more value, to bring more business to a business because that’s why you are there.
That’s why they hire you. That’s why they pay you money to do if nobody asks any other questions, I do have couple of more topics to note here.
What are some weird things that you have seen and what are some basic stuff that one can do to prevent capturing, you already noticed the mentioned the unwanted referrals and some [00:39:00] other aspects, but for example, recently I’ve seen some Elementor URLs from the backend in the analytics they were captured.
So what are some weird things that you can share? Of course. And what are some steps that one can take to filter out even more the data?
Dana: Yeah, I think the weirdest, I could do a whole presentation about weird stuff i’ve seen, the thing that I’m spending the most time fixing right now are consent implementation. So people have those like, Hey, are you cool with cookies?
Notifications that come up And most of the time I find that if you say yes or no, nothing happens. Literally nothing changes. On the worst, most egregious version I saw was that it actually sent off a Facebook conversion when you clicked no to Don’t track me, it’s just that just having the banner doesn’t actually do it.
You have to connect the banner with your tag manager or whatever it is that you’re using so that it can ingest the selection by the user for consent, and then make changes on what tags, as in pixels are delivered on the website. As a result of that consent. So absolutely, please do that. [00:40:00] And actually, you know what, I will, I’m gonna move around a couple windows here.
I’m actually gonna show you what I mean by that. So let me I’m gonna share my screen again if you could just pop that up in a second here.
Emanuel: Theoretically, you shouldn’t see anything unless the user click Yes.
Dana: Yes, exactly. So one of the things that I find, so this is, I use a plugin for Chrome right here, and it’s called Consent Mode Inspector.
And so I’m gonna go to one of our client’s websites, so camps and crew. And so I’m not gonna click on the ad. Okay. So for this particular client, they have this website uses cookies. So right now I have not accepted anything. So what this means is that the default state is denied. And then I have security storage granted, because that’s what stores the acceptance.
If I click on Allow All now at this point, I can see that the GCS parameters 1 0 1, which means that Google has activated and my update state is granted for everything. This is a really simple check that you can do just to make [00:41:00] sure that things are set up, that the way that they should be, and I also have a post that goes through this on my site.
I’ll just pull up the link and I’ll toss it in. Right here, how to verify consent signals are being received correctly. So I’ve put that in there if you wanna put that on the other networks that you’re including as well. Just to make sure that these things are being received correctly in GA4 for modeling and it does walk through I have a video, I always have a video on exactly what the default state should look like. I see people where we don’t have a default state. What would happen is I load up the site and then I have an update state, even though I haven’t chosen anything yet. I see that quite a lot. I see where people say no to being tracked and then literally nothing changes.
There’s no parameters showing up in here, even though data has gone off to GA4. So definitely you wanna be checking these things as well for consent. So if you have a consent dialogue, just make sure it’s actually doing something. That’s the number one thing I’m fixing right now.
Emanuel: And I can tell you from my experience, the big platforms that we talk about didn’t do a great job of helping [00:42:00] users implement these things as well either. I’m European myself, so I was I had to learn. It’s not as. Scary and as difficult as no one might think.
Dana: But I have to say OneTrust is the most frustrating. I think out of all the ones I work with, I mostly work with Cookie Bot. They’re my favorite, but OneTrust is so difficult to set up, so if you’re struggling with OneTrust, it’s not just you, it’s OneTrust. I think we have a question from Emit. Does AI traffic? Yes. Okay. AI agents do not trigger JavaScript, and if you’re using GA4, what’s called client side, which is what most people are using, you would know if you’re using server side.
Then that means that because it, that’s JavaScript code. AI agents do not execute JavaScript, at least as of right now. I don’t know, tomorrow, maybe they will. This field is changing so fast, but usually you do not see visits, and the easiest way to check is to do something like go into Chat GPT or Claude and ask it to fetch a URL.
Then watch your real time view in GA4 and see if that URL that you [00:43:00] asked it to fetch comes up in the real time view. If it didn’t, they’re not actually executing JavaScript at all.
Emanuel: Okay. Good question. That being sAId, we covered a lot. If every anyone has any other question, you now is the time to you said Dana, that, linkedIn is the best place for people to reach out and connect. So I just with your permission, I’m just dropping the URL over there where people can reach out to.
There’s the website as well. I’ll include everything in the show notes links to the articles linked to the courses, linked to the LinkedIn learning, which is an incredible resource that I recommend everyone in marketing overall, you don’t need to be necessarily the expert, but you’ll become once, once you learn and you’ll discover how powerful this tool is, and it’ll help you make better decisions because what gets measured gets managed.
Dana: I would also say too, for LinkedIn [00:44:00] learning if you have a Toronto or really, like most of you’re Toronto Public Library, Burlington Public Library, different public libraries have completely free access to LinkedIn learning with your library card.
So try that. Don’t, you don’t have to pay for LinkedIn. I do, but that’s because I do courses on there. So don’t feel like you have to pay for LinkedIn just to get access to LinkedIn Learning. Check your local library.
Emanuel: Pro I’ve taken advantage when I came to Canada, actually there was a, a very helpful resource.
So Donna, thank you so much. It’s been a pleasure. AgAIn, thank you so much for all the information you’ve, were out there. I’m following you closely in terms of the information that you’re sharing. I’m really honored and privileged to have you as a guest. Hope that we have the opportunity to do this some other time.
If you didn’t manage to join from the beginning or wanna come back and watch the recording. How about some marketing.com you’ll see probably a popup. You’ll see probably a notification that will encourage you to sign up to the newsletter. [00:45:00] Without anything else to add. Thank you everyone for being here.
And see you next time.
Dana: Thanks so much.
