智能體搜尋或比 GEO 更具機會，但衡量成效應先行

Generative Engine Optimization is useful shorthand for improving a site’s chances of being understood, cited, or recommended by AI systems. But it can become a distraction if teams treat AI referral traffic as a replacement for conventional search before the data supports that conclusion.

A Search Engine Journal analysis by Marie Haynes makes the case for a different priority: prepare for agentic search. That means focusing not only on whether an AI answer mentions a brand, but on whether AI-assisted experiences can discover reliable information, complete useful workflows, and lead to measurable business outcomes.

The distinction matters. AI search and chat referrals may show strong conversion rates in selected datasets, while Google Organic continues to provide the overwhelming majority of conversion volume. For most businesses, the sensible strategy is not to abandon SEO for GEO. It is to protect organic-search performance, improve data quality and conversion measurement, and make the site easier for emerging AI-led discovery and commerce flows to use.

What the analysis found about LLM referral traffic

Haynes analyzed 12 months of GA4 data from 100 websites representing more than 145 million sessions. In that convenience sample, ChatGPT traffic equaled 0.62% of Google Organic sessions. Including all LLMs increased the figure to 0.73% of Google Organic sessions.

That is a small share of traffic relative to Google, but the conversion data adds important nuance. Among the 55 sites in the sample that tracked conversions, Google Organic converted at 3.02%, compared with 13.8% for ChatGPT referrals. Higher conversion rates can indicate that people arriving from an AI tool are further along in a decision process, or that they have used the tool to narrow their options before visiting a site.

However, conversion rate is not conversion volume. Across the channels analyzed, Google Organic produced 96.5% of total conversions and roughly 27 times as many conversions as all LLMs combined. The analysis also observed a plateau in ChatGPT-driven visits beginning around August 2025, after GPT-5’s release.

These findings should not be read as a universal forecast for every industry. The sites were a convenience sample, only 55 reported conversions, and GA4 channel attribution can be affected by tracking configuration and changing referrer classifications. The figures demonstrate what occurred in this dataset, not a market-wide law.

Why agentic search is a broader strategic question

GEO often focuses on visibility in an AI-generated response: whether a system cites a page, mentions a brand, or recommends a product. Agentic search expands the question. The AI system may help a person research, compare, configure, purchase, or complete a task across several connected surfaces.

Google’s AI Overviews and AI Mode are part of that wider shift in search behavior. The potential implications extend beyond the search-results page to product feeds, checkout flows, and machine-readable ways for a website to expose actions or information.

The supporting technologies referenced in the analysis point in this direction. Google has announced Universal Cart, while the Universal Commerce Protocol describes a common approach for agentic commerce interactions. Chrome’s WebMCP documentation describes a way for sites to expose declarative tools to AI agents. Google has also discussed AI-enabled interaction through Android XR and intelligent eyewear.

These developments do not prove that every business needs an agent integration now, nor do they guarantee traffic or sales. They do show why the long-term opportunity may be less about chasing a citation in a chatbot and more about ensuring a business’s information, products, and actions can be reliably used in AI-assisted workflows.

A practical preparation plan for websites

The immediate work is largely disciplined SEO, ecommerce, and analytics work. It is not a separate replacement program called “agentic optimization.”

Keep the conversion baseline intact

First, protect the channel that currently drives scale. Monitor Google Organic sessions, leads, sales, and revenue alongside rankings and impressions. A decline in clicks caused by changing results-page layouts has a different business consequence from a decline in qualified conversions. Teams need both views.

Define GA4 key events that reflect meaningful progress, such as a qualified lead submission, account creation, completed purchase, booked appointment, or subscription. Avoid making page views or generic button clicks the primary evidence of AI-search success.

Then segment referral traffic where attribution permits. Compare Google Organic, ChatGPT, other identifiable LLM referrals, direct traffic, and paid channels using the same conversion definitions. Review conversion count, conversion rate, value where available, and the quality of downstream leads. This avoids overreacting to a high percentage produced from a tiny number of visits.

Make core information trustworthy and usable

Agentic discovery depends on the same basics that support search visibility and user satisfaction: crawlable pages, accurate content, clear site architecture, current product details, and pages that answer the user’s actual question.

For ecommerce businesses, product titles, descriptions, availability, pricing, variants, shipping information, returns, and product identifiers should be internally consistent. Aligning product data with Merchant Center is a practical way to reduce discrepancies across Google’s commerce ecosystem. For SaaS and service businesses, clear plans, capabilities, limitations, locations, contact routes, and conversion paths reduce ambiguity for both people and systems.

Structured data can help clarify entities and page meaning where it accurately reflects visible content. It is not a guarantee of inclusion in AI responses, but it supports a more machine-readable site.

Evaluate new agentic interfaces cautiously

New protocols and tools should be assessed against a concrete user problem. Ask whether an agent needs to retrieve information, compare options, build a cart, request an action, or pass structured details to a business workflow. Security, permissions, accuracy, and support requirements matter as much as implementation speed.

Do not deploy an endpoint merely because it is technically possible. A brittle or misleading integration can damage user trust. Start with well-maintained product or service data, validate the customer journey, and retain conventional website paths for users who prefer them.

For teams that want a structured way to identify where website clarity, entities, and content accessibility may need attention, create a GEO Search Leads account. It can help frame GEO readiness as part of a broader visibility review, so AI-discovery work supports the pages and information that already contribute to organic search and conversion performance.

Frequently Asked Questions

Is GEO no longer worth doing?

No. The analysis does not show that AI visibility is irrelevant. It shows that, in this dataset, LLM referral traffic was much smaller than Google Organic traffic and conversion volume. GEO work should complement strong SEO, clear site information, and outcome measurement rather than replace them.

What is agentic search?

Agentic search describes AI-assisted experiences that can help users move beyond finding information toward tasks such as researching, comparing products, and potentially completing commerce or service workflows. The exact capabilities and adoption of these experiences are still evolving.

Why can ChatGPT referrals have a higher conversion rate but fewer conversions?

Conversion rate measures the proportion of visitors who convert, while conversion volume depends on both the rate and the number of visitors. In the cited sample, ChatGPT referrals converted at a higher reported rate among sites tracking conversions, but Google Organic sent far more traffic and therefore generated far more total conversions.

What should businesses measure for AI-driven discovery?

Measure meaningful GA4 key events consistently across channels, then compare conversion volume, conversion rate, and outcome quality for identifiable AI referrals and Google Organic. Treat attribution carefully, because referrer classification and tracking setups can affect what is recorded.

Conclusion

The evidence does not support treating LLM traffic as a near-term substitute for Google Organic traffic. It does support watching AI referrals for high-intent behavior and preparing for a search environment where discovery, product data, and transactions may increasingly be mediated by agents. The strongest response is measured: maintain SEO fundamentals, improve trustworthy structured information, and judge new AI-search opportunities by verified business outcomes.