

For length reasons, we've edited and shortened parts of this transcript. You can listen to the full episode on Spotify and Apple Podcasts.
Host: Luke Kenna, Market Growth Lead, ICS-digital · Guest: Tom Capper, Director of Search Product Strategy, Moz · Listen on Spotify and Apple Podcasts
People haven't stopped buying, but AI search has made their journeys much harder to track. Moz's Tom Capper on why zero-click is an analytics problem, why your brand matters more than your domain, and how citations tell you what to do next.
On the latest episode of the Everything Digital Podcast, Luke Kenna, Market Growth Lead at ICS-digital, sits down with Tom Capper, Director of Search Product Strategy at Moz. Listen to the episode on Spotify and Apple Podcasts.
Tom spent eight years at Distilled, ending up running its consulting team in London, before joining Moz in 2021. An econometrician by training, Tom has always come at search from a measurement and data angle.
Building on Tom's recent brightonSEO talk, they discuss why zero-click is an analytics problem, who should own AI measurement, why you should track your brand rather than your domain, how to use citations and the most common prompt-tracking mistakes. Below is an edited version of the interview.
Luke: Can you tell us about your background, from Distilled to your role at Moz today?
Tom: I worked at Distilled, a fairly technically focused agency, from 2012 to 2020, when it merged into Brainlabs, and ended up running the consulting team in London. I've always come at search from a measurement and data background. I'm an econometrician by training, and when I started at Moz in early 2021, I was excited about the data Moz had and its focus on metrics and tracking.
I started in a research-heavy role and have since moved into more product-adjacent work: visual SERP tracking metrics, how we approach AI tracking and measurement, which is a very difficult topic, and our brand authority metrics. It's a super interesting space to be working in right now.
Luke: In your brightonSEO talk, you said zero-click isn't an SEO problem, it's an analytics problem. Can you unpack that?
Tom: It's a provocative statement, but SEO feels like a slightly fragile place to work right now, because a lot of clients and stakeholders are nervous about the channel. The traffic and revenue they can directly attribute to SEO is, in many cases, going down. But there's no sign of consumers actually buying less stuff. We're not in a recession, people aren't moving back to offline channels.
So we're all making the same kind of purchase decisions, but our journey looks very different now, and it's a lot harder to track. A buying journey used to go through top-of-funnel and middle-of-funnel landing pages, so you had those touch points. Now there's no touch point. Vanishingly few people click on citations, so you have no idea how someone got recommended your product when they suddenly appear by direct on a bottom-of-funnel page and make a purchase. A lot of people are seeing their direct conversions, or their branded PPC, going up, and it's not a coincidence.
It's an ironic thing for me to be saying, but we need to be more willing to accept that our measurement is flawed and has holes in it and that we're telling a more complex story. It's more like the story traditional marketers have told for a hundred years: you put up a billboard, and you have no idea whether the person who walks into the store saw it. Sure, you can do regional testing, incrementality testing or correlation analysis to find a stronger link, but it's fuzzy. It's not like it used to be and we're not going back.
Luke: Should brand teams and SEO teams share ownership of AI measurement?
Tom: You should at least talk to the brand team about it. The KPIs might be more at home there, but the work you're going to be doing still looks more like SEO work to me. I think we should, as an industry, take ownership of this and say, "This is the value of this channel. This is how we're going to measure it, and this is the work we're going to do."
We're in a position to talk about that whole value chain. We did something that looks a bit like an SEO activity, such as off-site PR. It caused the articles we influenced to get cited. The citations are associated with us being endorsed, which we know from our prompt tracking. And when we're endorsed, people convert, which we know from our post-purchase surveys.
That doesn't look like last-click conversion in Google Analytics. It's a much more complex pitch, but you can trace the steps on an individual campaign, which in traditional SEO you often couldn't. We understand the moving parts in ways that the traditional brand team, the one that works on your billboards, might not.
And as SEOs, we undersell ourselves a bit in terms of our track record on this. Think of a random client: which team owns their Google Analytics installation? It should be the PPC team, but it's usually the SEO team, because we've always been the people who take ownership of slightly technical things that need a multifunctional understanding. I don't think we're far away from where we need to be. It's just a slightly different mindset.
Luke: For someone setting up AI measurement for the first time, what's a sensible starting point?
Tom: Traditionally in SEO, the unit of work is a domain: how well is this domain ranking, which pages is it losing to its competitors? But AI responses are a brand channel. What you care about is whether people are going to pick out your brand in a list.
Say I ask ChatGPT or Gemini to recommend a pair of walking boots for hiking the GR20 in April. Often the boot that tops the list will be a Salomon something, and the citations will be various third-party hiking websites, but probably not the Salomon website. As Salomon, you have to look at that as a win, and as Salomon's competitor, you have to look at that as a loss, even if you're cited.
I see a lot of people who are obsessed with getting their website cited, and sure, it's not a bad thing for your website to be cited, but I really don't think it should be your end goal. The unit is probably your brand, including its products and sub-brands, a bit like separating branded and non-branded traffic in Google Search Console. The difference is that you now need to do the same for your competitors. You can't say whether you're winning a comparison if you're not tracking your competitors in the same way you're tracking yourself.
Luke: You made a distinction between mentions and citations. How can teams use citations to work out which tactics have an impact?
Tom: The mention is which brands, sub-brands or products come up in the body of the response. Citations are the links at the bottom: the web pages the tool used to form that response. I don't think getting cited should be your KPI, because if you make it your KPI, you're going to miss the biggest opportunities.
Oversimplifying, what ChatGPT is doing in the background is googling it. It looks at a large number of results about your product, your industry and your competitors, not just yours. In some ways your own pages are the least credible source on whether your product is good. Depending on which study you look at, you've missed 80 to 90 percent of your opportunity if you only optimise your own website here.
So you're trying to appear in all these other places, and that's the value of citations. Whatever it is that you are least pleased about in these responses, the citations are your instruction of what to do next. Sometimes it's an article you can directly influence, or your own social media profile. Often it's more nuanced: you can't appear in one industry publication because they've got an advertorial deal with your competitor, but you could appear in a different one.
That also gives you your proof of value. You can say: ChatGPT used to recommend our competitor, and these were the articles cited. We got ourselves mentioned in this article, now it's cited, and now we're recommended. That's really concrete, and it's something the brand team isn't used to having.
Luke: So they show you what you're doing well, or where you need to do something differently.
Tom: They're your instructions. Here's what you should go and do next.
Luke: What are the most common mistakes you see?
Tom: One is treating it like SERPs, with metrics like traffic and units like URLs. Probably the next biggest is people who just take their keyword list and copy it into a prompt tracker. No one goes into Gemini or whatever, types "laptops" and hits enter.
You need some way of generating a realistic list of prompts that matches what users actually do. I'm very wary of prompt volume as a concept. We try to think of it as topic volume rather than prompt volume. The vast majority of prompts have a volume of one. If your tool doesn't help with this, People Also Ask questions are a good starting point: you probably have them already from your rank tracker, and they're real, long-tail questions.
The last thing is the number of prompts you track. Prompt responses are famously not deterministic, so you want a large enough sample to see through the noise. If the line in your prompt tracking data is all over the place, that's just noise. If it's reasonably smooth, with occasional longer-term movement, you've got enough data to make a sensible decision. The cost of collecting this data isn't vastly different from the cost of collecting a Google SERP, so there's sort of no excuse to be doing like 10,000 top of funnel keywords and 50 prompts. It doesn't make sense.
Luke: Is daily or weekly sampling enough to smooth out the noise?
Tom: You could track a large number of prompts weekly, or a smaller number daily and then aggregate them. Either is perfectly reasonable. What you won't get is the instant response some businesses are used to, where you spot a ranking drop in the morning and fix it by the next day. You can't really influence these platforms that quickly.
I saw a great case study at brightonSEO about traditional SEO migrations that were also tracked in AI platforms. Even perfectly executed migrations that showed up instantly in traditional search took months to show up there. These platforms do just move more slowly. There is stuff you can affect in the short term, but it's not going to be any quicker than in Google search.
Luke: ICS's research on the human-machine memory gap found brands with strong human-facing distinctiveness but under 1% share of voice in AI retrieval. Does that gap show up in your data?
Tom: You often see a mismatch: very distinctive brands, loved by the people who use them, that are maybe niche and don't come up. That's probably more of a problem in AI search than in traditional channels, because these AI responses do tend towards the mean. Businesses that didn't care too much about mass recognition probably have to care a little bit more in this channel.
But we used to look at a brand with a big market share and a low share of voice in traditional search as an opportunity, and I think it's the same here. It means they've got an opportunity to show up in a channel that has been underserving them.
Luke: Looking ahead to MozCon London on 13 November, what's one thing listeners could do tomorrow to start measuring their AI performance?
Tom: Reprioritise around what's actually working in this space, which is probably things that aren't your own site. This is the hardest thing for SEOs to get over because we've had so long where really you could get away with just thinking about your own site. For the last 10 years or so, the most valuable work was probably on your own website. That's swinging back again.
So what I challenge people to go and do is find easy things they can do to improve their presence on other people's websites and then also see where those turn up in their citations. Then, in a month's time, because it won't be tomorrow, they can tell their boss a story: "I did this thing and now we're the recommendation for this question."
Hear the full conversation between Luke and Tom on the Everything Digital Podcast, available on Spotify and Apple Podcasts.
Find out more about MozCon on the Moz website.



