GEO Attribution Is Hard. It Is Also Worth Getting Right.

In traditional search, the attribution model is relatively clean. A user types a query into Google, clicks a result, lands on your site, and Google Analytics records the referral source, the landing page, and the conversion path. You can trace a revenue event back to a keyword.

AI search breaks almost every step in that chain. When someone asks ChatGPT "What is the best CRM for a 50-person team?" and gets a synthesized answer that recommends your product, there is often no click, no referral header, and no analytics event. The buyer may take that recommendation, go directly to your website by typing your URL, and convert. In your analytics, that shows up as direct traffic. The AI recommendation that caused the visit is invisible.

This is what Rand Fishkin at SparkToro calls "dark traffic". Between Apple cookie changes, ad blocker adoption above 40%, multi-device journeys, and AI-synthesized answers, a growing share of your most valuable traffic has no traceable origin. Google now answers nearly two-thirds of all queries without generating a click to external sites. AI chatbots take this further by synthesizing answers from multiple sources, often without linking to any of them.

The result is that traditional attribution models undercount AI-driven demand. If you wait for referral data to show up in Google Analytics before investing in AI visibility, you will be late. The signal is real. The measurement just requires a different approach.

Attribution in AI search cannot rely on a single signal. No individual metric tells the full story. What works is a layered approach where each signal reinforces the others, building a picture of causation that no single data point can provide alone.

We think about GEO attribution as three distinct layers, each with different latency, reliability, and actionability.

There is a reasonable skepticism about GEO metrics. If you cannot tie a specific revenue event to a specific AI mention, is the metric meaningful? We think the answer is yes, and the reasoning comes from how marketers already treat other high-impact channels that lack clean attribution.

Brand advertising has always had an attribution gap. You cannot trace a specific billboard view to a specific purchase. But marketers have spent decades investing in brand because the evidence shows it works: brand awareness correlates with market share, purchase intent, and pricing power. The Binet and Field research showed that brands with Share of Voice exceeding their Share of Market tend to grow, and vice versa. SOV is a leading indicator of market share change. Nobody calls that a vanity metric.

AI Share of Voice works on the same principle, applied to a new channel. The data supporting its importance is mounting rapidly.

When AI-referred visitors convert at 4.4x the rate of traditional organic, and AI chatbots are now the primary influence on B2B vendor shortlists, the channel is too large to ignore because the attribution is imperfect. Perfect attribution has never been a prerequisite for smart investment. Evidence-based estimation has always been enough.

The question is whether you are estimating with rigor or flying blind. That distinction is what separates evidence-based GEO measurement from vanity metrics.

Attribution difficulty does not mean attribution impossibility. Several concrete signals are available right now, and combining them produces a reasonable picture of AI-driven business impact.

1. Server Log Analysis for AI Crawler Activity

Your server logs capture 100% of bot requests. Google Analytics cannot see AI crawlers, but your access logs can. By filtering for known user agent strings, you can track exactly which AI systems are crawling your content, how often, and which pages they prioritize.

A spike in PerplexityBot crawl frequency on your pricing page, followed by an increase in direct traffic to that same page, is a strong correlational signal. It does not prove causation on its own. But combined with Layer 1 SOV data showing increased Perplexity mentions, the evidence becomes compelling.

2. Referral Traffic from AI Platforms

Some AI platforms do generate trackable referral traffic. Perplexity links to source websites when it answers queries. ChatGPT search (via OAI-SearchBot) produces citations. Google AI Overviews link to source pages. You can set up GA4 regex filters to capture traffic from these sources and track it as a distinct channel.

The catch is that this captures only a fraction of AI-influenced visits. A buyer who reads a ChatGPT recommendation and then types your brand name into Google will show up as organic branded search, not AI referral. A buyer who goes directly to your URL shows up as direct. The referral data is real but incomplete.

3. Branded Search Volume as a Proxy

One of the most underused attribution signals is branded search volume. If your AI SOV increases and your branded search volume increases in the same period, that correlation is meaningful. Buyers who hear about you through AI often validate the recommendation by searching for your brand name on Google. That behavior creates a measurable signal in Google Search Console.

This is analogous to how marketers have always measured TV and podcast advertising impact: you cannot trace a specific ad view to a specific purchase, but you can measure the branded search lift that follows a campaign. AI visibility works the same way.

4. Switchback Testing

The strongest attribution approach we have seen is switchback testing. Optimize your content for GEO in one product category while leaving a comparable category untouched. Track SOV, crawler activity, branded search, and conversion rates for both. If the optimized category shows meaningful improvement across all layers while the control does not, the causal inference is strong. This is the same experimental design used in offline advertising measurement and in ride-sharing pricing experiments. It is not perfect, but it is rigorous enough to support budget decisions.

The reason attribution matters is that real money is at stake. Companies are making budget allocation decisions right now about whether and how much to invest in AI visibility. Those decisions should be grounded in evidence, and the evidence is increasingly clear.

Gartner predicted that traditional search engine volume would drop 25% by 2026 as users shift to AI chatbots. That migration is already underway. SparkToro and Datos research shows AI tool usage has nearly tripled in the past year, and the growth rate is accelerating.

If 25% of your search traffic migrates to AI channels over the next two years, and you have no visibility in those channels, you are not just missing a new opportunity. You are losing ground you already held. The cost of inaction is measurable: it is 25% of your current search-driven pipeline, compounding over time as AI adoption grows.

Meanwhile, companies that invest in AI visibility now are capturing a channel where only 23% of marketers are currently active. The competition is thin. The conversion rates are high. And the brands that build SOV early will have a compounding advantage as the channel matures, just as early SEO adopters did in the 2000s.

We hear the same concerns from marketing leaders evaluating GEO investment. Each one is reasonable on its surface. Each one has a clear counterargument.

The pattern is consistent across marketing history. New channels always start with imperfect measurement. The companies that invest anyway, using the best available evidence, gain the first-mover advantage. The companies that wait for perfect attribution miss the window.

We built Sill to provide the leading-indicator layer that GEO attribution requires. Our monitoring dashboard queries the actual chat interfaces of ChatGPT, Perplexity, Gemini, Google AI Overviews, Copilot, and Grok daily, with web search enabled and citations included. This means we measure what real buyers see, not an API approximation that skips retrieval-augmented generation.

We track three core metrics that together form the foundation of GEO attribution:

• AI Share of Voice: How often AI platforms mention your brand when buyers ask purchase-intent questions. Tracked daily across six platforms. This is the GEO equivalent of traditional SOV, and the same relationship between SOV and market share applies.
• Sentiment and Positioning: Whether AI mentions are positive, negative, or neutral, and where your brand ranks when mentioned. A 30% SOV with positive sentiment is worth more than 60% SOV with mixed or negative framing.
• Semantic Positioning: How AI models perceive your brand relative to competitors on dimensions that matter to buyers. The Semantic Map plots this on configurable axes, so you can test whether changes in your content and off-site presence actually shift how AI positions you.

The attribution loop works like this: you make a content change (adding statistics, restructuring for answer capsules, increasing off-site mentions). You watch the SOV and Semantic Map for movement. If your SOV increases and your position shifts in the direction you intended, the intervention worked. If it did not, you iterate. This is the same test-and-learn cycle that drives performance marketing, adapted for a channel where direct click attribution is not available.

We are also actively running experiments and case studies to strengthen the causal link between GEO interventions and business outcomes. Our goal is to build the most rigorous attribution framework in the GEO space, grounded in the same experimental methodology that peer-reviewed GEO research uses.

We are transparent about the state of GEO attribution: it is early. The measurement infrastructure does not yet exist at the level of maturity that SEO enjoys. But we are actively working to close that gap through structured experiments with real brands.

Our current experiments focus on several questions:

• How quickly do on-site content changes (adding statistics, restructuring headings, implementing answer capsules) translate into SOV improvements?
• What is the lag time between an off-site mention (a Reddit thread, a YouTube review, a press mention) and a change in AI recommendation behavior?
• Do SOV improvements in Layer 1 correlate with increases in branded search volume (Layer 3) on a consistent timeline?
• Can switchback testing reliably isolate GEO impact from other marketing activities?

Across 748 GEO recommendations we generated for 62 brands, 87% were on-site fixes that brands can implement today. That gives us a large sample of interventions to track. As brands implement those recommendations and we monitor their SOV over time, we are building the dataset that will make GEO attribution more precise.

We will publish our findings as the data matures. If you want to be part of these case studies, reach out below. We are looking for brands willing to implement GEO recommendations and share anonymized results.

If you are evaluating GEO investment and need an attribution framework, here is where to start:

1. Establish a baseline. Measure your current AI SOV across major platforms. You need to know where you stand before you can measure change. This is Layer 1.
2. Set up server log monitoring. Start tracking AI crawler activity on your site. Note which pages get crawled most frequently and track changes after content updates. This is Layer 2.
3. Configure GA4 for AI referral traffic. Set up regex filters to capture traffic from Perplexity, ChatGPT search, and other AI sources. Track branded search volume trends in Google Search Console. This is Layer 3.
4. Implement GEO optimizations. Start with the highest-evidence tactics: add statistics to content (30-40% visibility improvement), restructure for answer capsules (87% of cited posts use them), update stale content (67% citation increase within 90 days).
5. Correlate across layers. After 30-60 days, compare SOV trends, crawler activity trends, and referral/branded search trends. Consistent directional improvement across all three layers is strong evidence of GEO impact.

Attribution does not need to be perfect to be useful. It needs to be rigorous enough to support the decisions it informs.