- Rand Fishkin
- Jul 30
- 7 min read
Updated: Aug 13
Does your Linkedin feed look like mine? So, filled with talk of AI replacing marketing jobs, being essential to your marketing career, and simultaneously so hard to figure out that you need to immediately subscribe to 600 newsletters, take 12 intensive courses, and follow 42,118 top AI influencers?
Yeah…It’s a little much for me, too.
Depending on the survey, somewhere between 55% and 99% of Americans used AI (in some form or fashion) in 2025. But how much they use specific AI tools (like ChatGPT, Gemini, Claude, Deepseek, or Perplexity) varies dramatically based on their industry, job role, and interests.
Unfortunately, there’s an overwhelming amount of hype around AI adoption, AI killing jobs (or not), and early adopters of AI tech outperforming their luddite peers. In every previous hype cycle (the PC revolution, the rise of the public Internet, the iPhone-spurred mobile craze, the social media bubble, blockchain and web3, etc.) the dust eventually settled into a new normal pattern of how people adopted and used these technologies, but not every company that made early, heavy investment benefitted equally.
Think back to the social media heyday of 2009-2012, when every marketing blog post, conference, and personality (shamefully, yours truly included) touted the value of building your own social network. Or the mobile phone adoption panic of a few years earlier, when those same blogs, conferences, and personalities screamed about how critical it was for every business to have their own mobile app.
In this short piece, we’d like to help you do two things:
Avoid getting caught in a hype cycle and overinvesting in a space that might not be a match for your customers
Determine how and whether your audience is using AI, how much, and which platforms
At SparkToro, we focus on the first point because we believe in following audiences, not leading them. Companies that overinvest in platforms or technologies that don’t get adopted by their potential customers waste resources and miss out on more pressing marketing opportunities. Don’t let that be you. Audience research is a long-practiced technique for uncovering a group’s real sources of influence. It’s meant to be a practice we invest in regularly (we’ve found that quarterly or annually are both reasonable cadences, because audience behaviors at scale don’t change more frequently than that).
And we care deeply about the second point because knowing your audience’s behaviors and preferences is, as the PSAs from our childhood taught us, half the battle.
When you have provable data on what your audience consumes before they make a purchase decision, you can:
Invest in the right types of PR and comms
Be present in the social media conversations they consume
Compare your brand’s presence to your competitors
And, most relevant to this conversation, discover which search and AI tools they’re using to research topics in your space
I.E. Do Better Marketing.
How can marketers uncover which search and AI tools their audience uses?
The field of audience research has a conveniently simple four options:
Surveys
Interviews & focus groups
Analytics
Panel data
Which one should you use? TL;DR - Number 4.
Surveys, interviews, and focus groups rely on human beings possessing perfect recall about all the steps they took in a research and purchase process—they don’t. My favorite story about this might help convince you:
In the early years of SparkToro, I was able to convince one of our customers who loved and relied on post-consumer purchase surveys (a pop-up that asked people who made an e-commerce purchase on their site how they found them) to insert two “false,” famous influencers as options in the drop-down. These two people are household names in their field (TV chefs in the food world), but had never talked about their product.
Fast forward a month, and 30% of their customers said that one of those two celebrity chefs were responsible for their purchase. The e-commerce firm discarded the survey, and switched to higher accuracy forms of audience research.
People aren’t intentionally lying to you. They’re not trying to make your data messy. They just can’t remember all the dozens of nudges and influences that had an impact on the discovery of your brand or their purchase decision. And yes, some of them just want to get through the checkout process as fast as possible and will simply choose a random radio button in a survey to make the darn thing go away.
This holds true in interviews and focus groups as well. Memory is fallible, and no one records their multi-step buyer’s journey with precision. To do that, you need machines.
What about analytics? Can your GA4 setup tell you whether your audience is using a particular AI tool or shifting away from traditional search engines?
Nope.
Some AI tools have started providing referral data, meaning you can view your logs or look at GA’s referring domains to see traffic from a source. But this referral data doesn’t say:
What the prompt/chat contained
Whether the referral was relevant to your products/services
When your competitors are getting more AI traffic than you
The biggest issue is that no matter how much or little referral traffic AI engines send (and according to the numbers, ChatGPT refers a visit from <2% of sessions compared to Google’s ~40% per search), the numbers in your analytics can’t answer the real questions:
Is my audience using AI tools?
How much more/less than average?
How much more/less than other platforms I might invest in?
That leaves us with panel data—a solid, though imperfect option, it’s the best one we’ve got.
What do we mean by “panel data?”
Way back in the 1950s, Nielsen pioneered a device that sat on top of television sets and sent information about what people watched back to their corporate offices for compilation and analysis. If you were a “Nielsen TV family,” your viewing habits informed what networks created and how they measured the success of their TV shows compared to their competition’s.
Fast forward to the Internet age and a similar service—clickstream panels—provides similar data about what people visit on the web. Clickstream offerings from companies like Nielsen, Comscore, SimilarWeb, and Datos enable us to see how popular a particular website is, how often it might be visited, and what else people who visit that site do on the internet.
At SparkToro, we use Datos’ clickstream panel to power a significant portion of the information our software shows about a given audience. We can’t know what every person on the web visits, or every site’s traffic precisely; but, thanks to this methodology, we have an excellent idea of what’s more or less popular than average with almost any describable group of online users.
Panel statistics use a similar methodology to election polling: by assembling a sizable, diverse group of people, you can make predictions about how they’ll vote (or, in our case, which websites they visit and don’t). The numbers may be off by +/- 3-5%, but while that matters a huge amount in elections, it’s rarely as worrisome in the world of web traffic analysis.
Uncovering your audience’s search & AI tool usage
To sum up: Don’t ask your audience via surveys or interviews whether they use AI tools. Don’t try to back into it via your analytics. And don’t make assumptions based on stereotypes or preconceived notions (or worse, the highest paid person’s opinion). Get real, panel-observed data from a clickstream source that sees what millions of devices do on the web, and narrow to the group of people you care about reaching.
Here’s how that works in SparkToro:
Above, I’ve searched for people whose profile contains the words “HVAC contractor.” That data comes (mostly) from public LinkedIn accounts who have those words somewhere in their job history, profile description, or professional interests.
The software then connects those profiles up to behaviors from the clickstream panel and produces an analysis of which traditional search engines (e.g. Google and Bing), e-commerce and content search platforms (Pinterest, Reddit, Amazon, etc.), and AI tools (ChatGPT, Perplexity, Claude, etc.) are likely to be visited, how much, and how much more/less than average American web users.
The graph looks like this:
From left to right, we see the “Audience Rank” of each different search tool. Google is the most-used by HVAC contractors, Amazon second, and Wikipedia third. ChatGPT is 8th, but it’s used ~60% more frequently than average by this group. DuckDuckGo, the privacy-focused search engine, is 11th and is also +40% more popular than average. Pinterest, on the other hand, is in 7th place, but is used ~7% less by this group than by the average American web user.
Compare that against another, more B2B-centric audience in a completely different field:
Balsamiq.com is a wireframing tool for founders and product designers of software companies. It’s especially popular with indie startups and folks at smaller software businesses. And that audience’s usage of search and AI tools is dramatically different from our HVAC contractors (no surprise).
Amazon has moved from position 2 to 6. It’s actually less-used than average by this group (no surprise, as software folks have far less need for the multitude of physical parts and components that HVAC pros are frequently buying online). ChatGPT is the 5th most popular tool for this group, with +142% more use than average by this group. Perplexity and Claude are now in the top 12, too! Clearly, we’ve hit a very AI-centric audience here, but even among this group, Google and Bing are first and second. Traditional search is hard to unseat.
What next?
Knowing your audience uses a particular platform, how much, and in comparison to others gives you an excellent starting point for where to invest your future marketing efforts. Our recommendation has long been that large companies with big budgets invest everywhere their audiences participate heavily (and relevantly). On the other hand, if you’re an SMB, focus on just a few channels at the intersection of your organization’s aptitude, where you can provide unique value, and where you know your audience pays attention.
With contributions from Amanda Natividad, VP of Marketing at SparkToro
