GEO ROI: How to Measure If AI Visibility Is Actually Driving Revenue (Not Just Vanity Mentions)

TL;DR: A blended "AI visibility score" going up is not proof of business impact. Eli Schwartz, SEO/AEO consultant to companies including Zapier, Tinder, Coinbase, and LinkedIn, has publicly argued the industry is "measuring prompts instead of customers," and that visibility tracking is infrastructure, not leverage, unless it's tied to familiarity, branded search lift, and lower customer acquisition cost.^[1] He's right to push on this. Here's how to measure GEO in a way that actually survives that scrutiny.

---

Taking the strongest counter-argument seriously

Most content in this category (including plenty of ours) makes the case for why AI visibility matters. It's worth engaging the best version of the opposite argument directly, because ignoring it makes every visibility dashboard on the market look naive.

Eli Schwartz's critique, in his own words from a recent podcast conversation: "I would push back against the way the industry is approaching visibility right now. I think they're getting too granular on what visibility means, and it reeks of rank tracking from many, many years ago... We're back to that place where now we're measuring prompts instead of measuring customers."^[1]

His broader point, made separately: infrastructure isn't leverage. If someone has seen your brand cited in AI answers, surfaced in comparisons, reinforced in reviews, that builds familiarity. Familiarity is what drives branded search and lowers customer acquisition cost when that same person later encounters a paid ad or organic listing. But the visibility metric itself, the mention count, the citation count, is not the thing that pays the bills. It's a leading indicator, at best, of a downstream business outcome most GEO tools never actually measure.^[1]

This is a fair criticism, and it should change how you report on this work internally.

The vanity-metric trap, concretely

Here's what a vanity-metric report looks like: "Our AI Visibility Score went from 42 to 58 this quarter." That number, alone, tells a CFO nothing about revenue, and it's exactly the kind of isolated number that earned SEO's "rank tracking" reputation a decade ago, impressive-looking, disconnected from outcomes.

Here's the same underlying work, reported differently: "Our mentions on ChatGPT for [category] queries rose from 3 competitor comparisons out of 10 tested to 7 out of 10. In the same period, branded search volume for our name rose 22%, and AI-referral sessions in GA4 grew from near-zero to 340/month, at a similar or better conversion rate than our organic search traffic." That's a fundamentally different, and much more defensible, claim.

The measurement framework: leading vs. lagging indicators

Track both, always, never leading indicators in isolation.

Leading indicators (check every 2-4 weeks)

• Manual mention/citation checks against real buyer queries, per platform (see our mentions vs. citations explainer for why these must be separate numbers)
• Backlinks/referring domains to key pages
• Presence in your category's Citation Core (are the trusted third-party sites describing you accurately, see the full playbook)
• Organic impressions and click-through in Google Search Console

Lagging indicators (check monthly, tie directly to revenue)

• AI-referral sessions in GA4. Set up a segment for referrer traffic from chatgpt.com, perplexity.ai, gemini.google.com, and similar domains. This is the closest thing to a direct-click number this category currently offers.
• Branded search volume. Track "[your brand name]" search volume over time (Google Trends, Search Console). This is the metric Schwartz's critique points to directly, if AI visibility is building real familiarity, more people should be searching for you by name over time, independent of any specific click from an AI answer.^[1]
• Assisted conversions and CAC trend. If branded search and direct traffic are both growing, and paid acquisition cost is trending down even slightly, that's evidence the "familiarity" effect Schwartz describes is real for your brand specifically, not just theoretical.
• Trial/signup attribution by landing page, if you have UTM-tagged CTAs on your GEO-focused content specifically (this is squarely a leading-to-lagging bridge metric).

Why this framing makes GetSeenInAI's own reporting different

We built our dashboard around this exact tension. It's genuinely useful to know your mention and citation counts, they're the fastest, most controllable thing to act on week to week. But we explicitly do not claim that number, alone, is your ROI. Our reporting connects the leading indicators (mentions, citations, Citation Core presence) to the lagging ones you actually care about (AI-referral sessions, branded search trend) so you're never reporting a number you can't defend to a CFO.

A realistic reporting template you can use today

This quarter:

• Mentions on [platform]: [X] → [Y] (source: manual/tool check)
• Citations on [platform]: [X] → [Y]
• Citation Core presence: [list of sites you're now accurately represented on that you weren't before]
• Branded search volume: [X]% change
• AI-referral sessions (GA4): [X] → [Y]
• What we did to cause this: [specific, named actions, not "general GEO work"]
• What we're doing next quarter: [specific, named actions]

This format survives a skeptical stakeholder conversation because every line is either a verifiable number or a specific, named action, never a vague "visibility improved."

Where we agree with the critique, and where we'd push back

Where Schwartz is right: a mention/citation count in isolation is not revenue, and an industry that reports only that number is repeating a mistake SEO already made and corrected years ago. Tie everything back to familiarity, branded search, and CAC, or the number is decoration.^[1]

Where we'd add nuance: "infrastructure, not leverage" undersells one thing, in categories with real Citation Core dynamics (software, finance, health, per Semrush's data^[2]), being absent from AI answers entirely is now a genuine, measurable business risk, not a hypothetical one, because a growing share of comparison research increasingly happens inside an AI chat window before a user ever reaches a branded search or a paid ad. The right conclusion isn't "ignore this metric", it's "never report this metric without the revenue-adjacent numbers next to it," which is exactly the discipline this post is arguing for.

FAQ

Should I stop tracking mentions and citations altogether? No, they're still the fastest, most actionable leading indicators available, and the only way to diagnose why a lagging metric is moving. The mistake is reporting them alone, not tracking them at all.

How long before AI-referral traffic in GA4 becomes meaningful? For most mid-market brands, expect low, sometimes near-zero AI-referral sessions in month 1-2, with growth becoming visible around month 3-4 as citation-building work (see our timeline post) starts to compound.

What if branded search volume doesn't move even after mentions improve? That's a real, useful signal, it suggests the mentions you're earning aren't reaching the right audience with enough frequency or prominence to build actual familiarity yet, which should redirect effort toward higher-visibility platforms/placements rather than more of the same activity.

---

Track the numbers that actually connect to revenue, not just a score. GetSeenInAI reports mentions and citations alongside guidance on connecting them to your GA4 and branded-search data, start with the free ChatGPT Visibility Score check.

---

Sources

1. Previsible, "Jordan Koene and Eli Schwartz talk AI visibility in enterprise SEO," Voices of Search podcast recap, 2026.
2. Semrush & Adobe, AI Visibility Index 2026, Semrush, 2026.