AI referral traffic is up 527% year-over-year. One marketing firm now attributes 80% of its leads to AI platforms. And most enterprise demand gen teams are measuring their AI visibility completely wrong.
AI referral traffic is up 527% year-over-year. One marketing firm — Innovaxis — now attributes 80% of its leads to AI platforms, driven by structured data, schema optimization, and systematic prompt monitoring. And most enterprise demand gen teams are measuring their AI visibility completely wrong.
That gap between the opportunity and the measurement reality is where pipeline goes to die.
For demand generation leaders who’ve spent the last two years watching AI eat into traditional search traffic, prompt tracking has become the new keyword ranking — the metric everyone talks about, and almost everyone gets wrong. Tom Capper, who heads the Search Science team at Moz, has catalogued four mistakes that show up repeatedly in how organizations approach AI visibility measurement. They’re not obscure edge cases. They’re the default behaviors of teams trained on SEO instincts that simply don’t transfer.
Here’s what those mistakes look like — and why fixing them matters more than most marketing leaders currently appreciate.
## Mistake #1: Obsessing Over Citations Instead of Mentions
This is the most common error, and it’s a logical one given where most demand gen teams came from.
In traditional SEO, the citation is the prize. You want the backlink. You want your domain referenced. So when teams start tracking AI responses, they naturally look for whether their site is the cited source in a given answer. If PCMag gets credited and you don’t, that reads as a loss.
It isn’t.
Consider how AI responses actually function. When a user asks an AI tool which phone to buy, the response might mention Apple, cite PCMag as the source, and send zero clicks anywhere. The citation is an artifact of how the AI constructed its answer — it’s not the equivalent of a ranking position. The brand mention is what matters. Apple appearing in the response is the win, regardless of which domain gets the footnote.
This distinction has direct implications for how teams should respond to their tracking data. If your brand is being mentioned but a third-party site is being cited, that’s not a failure to fix — it’s an outreach opportunity. Getting your product featured in the roundup articles that AI systems pull from is a different play than traditional link building, but it’s tractable. If your brand isn’t being mentioned at all, that’s the actual problem.
The practical audit: pull your current AI tracking reports and check whether you’re measuring domain citations or brand and product mentions. If it’s the former, you’re optimizing for the wrong signal.
## Mistake #2: Applying Ranking Logic to a Non-Ranking Environment
The second mistake follows directly from the first. Teams that migrate from SEO to AI visibility tracking tend to bring the ranking mindset with them — and the ranking mindset is almost entirely the wrong frame.
In search, position one means something concrete. You appear first, you get disproportionate click share, you win. So when AI tracking tools show that a competitor is mentioned first in a response, the instinct is to treat that as a rankings deficit to close.
But AI responses don’t work like search results pages. A brand mentioned third in a conversational response isn’t necessarily losing to the brands mentioned first. The relevant question is: across a large sample of prompts, what percentage of responses mention the brand at all?
That percentage — not position — is the baseline metric. Once you have it, you can layer in context: how the brand is described, what adjectives appear around it, whether the mentions are positive or qualified. But those nuances only mean something after you’ve established whether you’re in the conversation to begin with.
The benchmark for category leaders, based on current best practices, is a 30% or higher inclusion rate across core prompts. That’s the target — not first position, not citation frequency. Inclusion rate across a representative prompt set.
For B2B demand gen specifically, this shift matters enormously. With 20–30% conversion rates now attributed to AI recommendations in some categories, being in the conversation is a pipeline question, not just a brand awareness question. Missing from AI responses means missing from consideration — earlier in the funnel than most attribution models currently capture.
## Mistake #3: Tracking at the Wrong Scale for Your Business
The third mistake is more mechanical but equally consequential: tracking too few prompts, or the wrong number for the organization’s size and complexity.
Fifty prompts is a reasonable starting point for a local business, a niche-specific product, or a small organization with a narrow product line. For an enterprise brand operating across multiple products, markets, and customer segments, 50 prompts produces data that’s statistically unreliable at best and actively misleading at worst.
Here’s why. Prompts are fundamentally higher-variance than keywords. A keyword like
