AI Citations Without Clicks: Does GEO Have Business Value?

It is tempting to call a citation a win. It is equally tempting to dismiss it when sessions remain flat. Both shortcuts are weak. A citation is evidence of answer-level exposure; a click is an observed visit; a sale is a business result. They may be related, but one cannot be substituted for another.

An August 2026 preregistered field experiment with 1,100 participants makes the click concern hard to wave away: removing Google's AI Overviews and AI Mode increased click-through to publishers, while an AI Mode-only condition reduced it and worsened reported user experience and trust. The study concerns Google Search behavior in its experiment, not every e-commerce category or a measured effect of GEO services. It is a reason to set a higher bar for business-value claims.

Methodology & sources

Editorial review for factual claims (as of 2026-09-28).

We checked the study abstract and preprint status, Google's generative-AI Search Console reporting and Bing's definition of Citation Share. The experiment identifies effects in its Google Search conditions; the measurement framework below is our recommendation. It does not turn AI citations into proven incremental sales.

The four-column scorecard

Choose a category and a fixed panel of buyer questions. Agree on the review window before changing content. Keep these outcomes separate:

Microsoft explicitly calls Citation Share an observational measure of displayed citations for a particular grounding query—not traffic share or a ranking. Google reports URL impressions in its own AI features. Neither metric should be multiplied by an assumed conversion rate to create a made-up revenue estimate.

Diagnose the mismatch before calling it success or failure

Case A: citations rise, qualified visits do not. Check whether the cited material answers the full question without a reason to visit. For a publisher, that may be a real substitution risk. For a retailer, the answer may still put the product in consideration, but you need independent evidence—such as later branded search, first-party customer research or a controlled test—before claiming commercial influence.

Case B: referrals rise, leads do not. Audit landing-page relevance, stock, pricing and the next step. A citation from a broad informational query may bring low-intent traffic. The problem may be offer fit rather than visibility.

Case C: recommendations improve, but revenue remains flat. Review sample size, seasonality, margin and sales cycle. If buyers cannot buy the product, a stronger recommendation is not the priority. If the change concerns a long B2B cycle, measure qualified pipeline rather than same-week revenue.

A more honest test of value

Select one narrow question cluster, document the baseline and make one meaningful change to the cited page or underlying product facts. Keep other campaigns visible in the change log. Re-measure the same questions, search exposure and downstream outcomes over a window appropriate to the buying cycle. Where possible, compare with an untouched category or market. This is still not perfect causal attribution; it is a stronger decision design than choosing a favorable before-and-after screenshot.

Define the decision threshold in advance: for example, “continue if the answer becomes accurate on most sampled observations and qualified product-page visits or sales conversations move in the right direction.” If the answer improves but business outcomes do not, report that plainly and decide whether the learning justifies another iteration.

GEO is useful when it improves a buyer's ability to find and evaluate a real offer—and when the organization can show what changed. A citation by itself is an observation, not the invoice.

Frequently asked questions

The same Q&A pairs ship as FAQPage structured data so AI engines can quote them verbatim.

Is an AI citation proof that GEO increased revenue?

+

No. A citation proves that a source was displayed under observed conditions. It is not a visit, qualified lead or sale. Report answer exposure separately from first-party website and commercial outcomes, then evaluate a documented change on comparable buyer questions over a suitable buying-cycle window.

What does the 2026 experiment say about AI search clicks?

+

An August 2026 preregistered Google Search field experiment with 1,100 participants found that removing AI Overviews and AI Mode increased publisher click-through, while an AI Mode-only condition reduced it. The finding applies to the study’s conditions; it is not a measured conversion effect for every retailer or GEO campaign.

How should a business evaluate citations when traffic stays flat?

+

Diagnose the buyer intent and landing page, then look for independent evidence such as qualified visits, customer research, leads or orders. A citation may substitute for a publisher visit or precede later product consideration. Neither explanation can be confirmed by citation count alone, so document the uncertainty.

Primary evidence and limits

• Wang et al., “AI in Search Reduces Publisher Referrals Without Improving User Experience” (arXiv preprint, August 2026) — preregistered Google Search field experiment, N=1,100.
• Google Search Central: generative-AI performance reports — URL-impression definition.
• Microsoft: Citation Share definition and limits — citations are not traffic or ranking share.

Reviewed September 28, 2026. The paper is a preprint and its experimental click effect is not a GEO Tracker AI customer outcome or a universal e-commerce conversion estimate.