Ninety-four percent of B2B buyers now use AI search in their purchasing process. That number, from Ahrefs' recent analysis, should make every CMO sit up straight. It means your prospects are asking ChatGPT, Perplexity, and Claude about your category before they ever hit your website. And here's the kicker: you probably have no idea if any of that visibility is actually working.
Welcome to the newest headache in marketing measurement. AI search visibility is the channel everyone wants to crack, but almost nobody can prove ROI on. It's like trying to measure the impact of word-of-mouth, except the mouth belongs to a large language model that doesn't leave referral data behind.
The Attribution Black Hole
Let me be blunt about why this is so hard. Traditional search gives you clicks, sessions, conversions. You can trace a path. AI search? A user might get a complete answer mentioning your brand and never visit your site. When they do click through, the referrer data often gets stripped, making it look like direct traffic. And even when someone acts on an AI recommendation, they rarely tell you that's what happened.
IBM's research found that only 29% of companies can effectively measure AI ROI, while 79% report seeing productivity gains. That gap tells you everything: we know something's working, we just can't prove what.
But here's what I tell my team: executives don't need perfect numbers. They need reasonable estimates they can track over time. And between what you can measure and what your customers tell you, there's enough signal to build something useful.
The Four-Step Framework That Actually Works
If my CFO asked me tomorrow to justify our AI search investment, here's the process I'd follow. It's not bulletproof, but it's repeatable, and that's what matters.
Step one: Calculate your actual AI costs. This sounds obvious, but most teams undercount. You need the full picture: tools for monitoring AI mentions, content optimization for AI visibility, any specialized talent or agency fees, and the internal hours your team spends on this. As one analysis of 200 AI deployments noted, internal time is the most systematically omitted cost. Skip it, and your ROI looks artificially inflated.
Step two: Estimate your AI-attributed revenue. This is where it gets creative. Start with what you can track: referral traffic from AI platforms that does show up in analytics, form submissions where prospects mention AI tools in "how did you hear about us" fields, and sales conversations where buyers reference AI recommendations. Then layer in survey data. Ask new customers directly. You'd be surprised how many will tell you "I asked ChatGPT for recommendations" if you simply ask.
Step three: Turn those numbers into a percentage. The formula is straightforward: (AI-attributed revenue minus AI costs) divided by AI costs, times 100. McKinsey reports that 87% of companies see positive ROI within six months on well-executed AI projects. Your mileage will vary, but having a baseline number, even an imperfect one, beats flying blind.
Step four (optional but valuable): Calculate equivalent ad value. If you're appearing in AI responses for queries that would cost you $50 per click in Google Ads, that visibility has a shadow value. It's not revenue, but it's a useful way to contextualize your presence to stakeholders who think in media spend terms.
The Baseline Problem Nobody Talks About
Here's where most measurement efforts fall apart: they start measuring after the project launches. Without a baseline measurement before deployment, any claim of improvement is just a feeling dressed up as data.
Spend four to eight weeks before any AI visibility initiative documenting your current state: brand mention frequency in AI responses, referral traffic from AI platforms, conversion rates from that traffic, and qualitative feedback from sales about buyer behavior. That's your point zero. Everything after gets compared to it.
One company I know discovered their "massive time savings" from an AI tool actually amounted to 7 minutes and 24 seconds per task when they measured it properly. The perceived problem was enormous; the actual problem was tiny. Baselines keep you honest.
Hard ROI vs. Soft ROI: Pick Your Battles
IBM distinguishes between "Hard ROI" and "Soft ROI" for AI projects, and the distinction matters for how you report results.
Hard ROI is the stuff finance loves: reduced cost per lead, increased conversion rates, shorter sales cycles directly attributable to AI visibility. These are the numbers that get budgets approved.
Soft ROI is trickier but often more significant: improved brand perception, faster access to information for buyers, better positioning against competitors in AI recommendations. A marketing director who spots trends faster because AI surfaces your content creates value. A buyer who trusts your brand more because AI consistently recommends you creates value. But try putting that in a spreadsheet.
My advice: lead with hard metrics, but don't abandon soft ones. Frame soft ROI as leading indicators that predict future hard ROI. Brand visibility in AI responses today becomes pipeline tomorrow.
The Uncomfortable Truth About Precision
Deloitte's 2025 survey indicates most AI use cases need two to four years to reach satisfactory ROI. That's a long time to wait for certainty, and it means you'll be making decisions based on incomplete data for a while.
Get comfortable with that. The alternative, waiting for perfect attribution before investing in AI visibility, means ceding ground to competitors who are willing to act on reasonable estimates.
The searches for "AI attribution" and "AI ROI" have grown 212% over the past 18 months, according to Ahrefs' keyword data. Everyone's trying to figure this out. The marketers who win won't be the ones with perfect measurement. They'll be the ones who built a consistent, repeatable framework early and refined it as the data got better.
What I'm Telling My Team
AI search is like dating: you don't propose on the first ad impression, and you can't always trace the moment someone fell for your brand. But you can track patterns, ask good questions, and build a picture over time.
Set your baseline now. Measure what you can. Survey what you can't. And stop waiting for the attribution gods to deliver perfect data. They're not coming. The CMOs who figure out "good enough" measurement first will be the ones explaining their success to the rest of us at next year's conferences.
