AI Search Traffic: What Queries, Clicks and Pipeline Can Tell You

Gartner forecast a 25 percent decline in traditional search volume by 2026. Alphabet's reported usage and revenue do not establish that category-wide outcome, and a site's clicks are a different measure again.

On February 19, 2024, Gartner predicted that traditional search engine volume would drop 25 percent by 2026, with search marketing losing share to AI chatbots and virtual agents. Alan Antin, a Vice President Analyst there, framed generative AI as becoming a substitute answer engine, replacing queries that previously ran through traditional search. That prediction got quoted in roughly every deck about the future of marketing for the next two years, including some I sat through.

Compare the dated forecast with the dated observations, keeping their denominators separate.

The collapse has not come

In its April 29, 2026 Q1 earnings transcript, Alphabet said queries were at an all-time high and credited AI with driving Search usage. In Q2, Google Search and other revenue was $63.3 billion, up 17 percent year over year. The latter is revenue, not a query count.

Two limits apply to that comparison.

First, Gartner's prediction was about the search category, not about Google specifically, and Google is not the whole category. Alphabet's numbers are the best public proxy available at roughly ninety percent share, but they are a proxy. Second, the all-time-high query claim is Google's own, unaudited, and Google does not publish absolute query counts. It is also fair to argue that counting AI Mode sessions as searches is doing some work in that number, which is arguably the substitution Gartner described wearing a different label.

Alphabet's commentary does not establish a category-wide 25 percent decline. It also cannot settle Gartner's forecast: Google is one part of traditional search, absolute query counts are not disclosed, and the 2026 observation window was not complete at this article's original publication.

For planning, a category forecast is insufficient evidence that your own site will gain or lose visits.

Your clicks fell anyway

Search volume and website clicks use different denominators. A change in one does not establish the cause of a change in the other.

Some searches resolve on the results page without a website visit. Others still produce clicks. The mix depends on query intent and result format, so a site needs its own page and query evidence.

Google has been asked about this directly and its answer is worth reading closely, because it is carefully built. When independent research published in July 2025 found that clicks to external sites dropped substantially when an AI summary was present, Google called the methodology flawed and the query set unrepresentative, then said this: it has not observed significant drops in aggregate web traffic.

Aggregate. Across the entire web. That statement can be completely true while your specific site loses half its organic sessions, because aggregate stability is exactly what you would expect if traffic is being redistributed rather than destroyed. Some sites gain, some lose, the total holds. Nobody plans a budget against the total for the entire internet. You plan against your own line, and Google has never made a claim about your line.

Pew's July 22, 2025 study reports lower traditional-result click rates when an AI summary appeared. The dark AI traffic article retains its browsing-panel limits. Google's aggregate response cannot establish the outcome for your property.

Where the attention actually went

Similarweb's July 29, 2026 report estimates an average 9.5 billion monthly visits to generative AI platforms from June 2025 to May 2026, up roughly 70 percent year over year, with unique visitors up 57 percent to 655 million. These are assistant-platform visits and visitors, not search queries or outbound referrals.

The twelve-month average is not a current monthly run-rate. The figures are panel estimates with their own method and scope. They describe attention on the assistants and cannot be used as the number of visits available to your website.

ChatGPT's share of generative AI visits fell from roughly 76 percent in June 2025 to about 53 percent by May 2026. Gemini climbed from under 9 percent to roughly 27 percent. Claude went from around 2 percent to close to 9 percent, the largest proportional gain in the set. In twelve months the market went from one assistant with three quarters of the attention to a genuine three-way split.

That platform mix gives a reason to examine more than one assistant. It does not mean half of your previously reachable website referrals moved elsewhere. Measure your own observed referrals and citations by engine, with the period and method attached.

A warning about the statistics you are about to be sold

Check each AI-conversion benchmark's site sample, period, conversion event, and revenue definition before using it in a plan.

The specific figure I keep seeing is that AI referral traffic converts at around 7 percent, comfortably ahead of organic and just behind paid search. I went looking for its origin. It traces back to one agency reporting the result from a single website, not a multi-site benchmark, not a disclosed methodology, no sample size, no confidence interval. One site. That number then got cleaned up, given a decimal point, and put in a comparison table against channel averages built from completely different data.

Pre-click research might explain higher-intent referrals in some funnels, but that is a hypothesis until the funnel measures it. A one-site result cannot establish a universal AI conversion rate.

The agent-commerce article makes the same distinction between vendor telemetry, experiments, and completed-transaction evidence. Evaluate the method rather than the confidence of the headline.

What I actually did about it

For a domestic defense and advanced materials manufacturer, I helped rebuild the demand engine through a rebrand, a multi-brand and multi-domain search architecture, and content intended to be useful in both traditional search and AI answers. The case chart records 43 web leads in 2023, 119 in 2024 and 203 in 2025. Its 450-plus figure is a 2026 projection, not a completed result. These figures describe the combined commercial program; they do not isolate the effect of answer-engine optimization. The defense manufacturing case study gives the contribution context.

That program combined branding, search architecture, and content work. Its before/after results do not isolate the effect of answer-engine optimization or prove a guaranteed citation mechanism. Clear claims and checkable sources are useful publishing work; measure what they produce in the relevant funnel.

Preserve the source information a visit actually supplies and keep unknown origins visible. The attribution series covers those limits, including GA4's channel definition. The manufacturer results above describe the combined program, not recovered dark-AI traffic.

How to plan against this

Four things, in the order I would do them.

Plan against qualified visits and accepted outcomes at your own site. Total query volume provides context but cannot set a property's traffic budget.

Investigate ranking changes, result-format changes, query mix, and measurement changes separately. Several can occur at once; a fall in organic sessions alone cannot allocate the cause.

Publish useful, checkable answers and maintain the technical conditions for discovery. Citation and ranking can both matter. Neither automatically produces a click or sale.

Instrument before scaling. Keep observed referral sources, citation samples, and downstream outcomes separate, then compare equivalent periods. Do not give an unknown visit an assistant label to make the report complete.

What the planning evidence supports

The dated data does not settle Gartner's entire forecast. It does show why search volume, assistant-platform visits, and website clicks need separate treatment.

Technical SEO can still repair indexing, access, or ranking problems. It cannot force every answered query to produce a visit. Choose the fix against the observed cause at the property.

Use the pipeline handoff to evaluate the commercial outcome, including missing source evidence. A citation is one observed event in that chain.

Frequently asked questions

Did Gartner's prediction that search volume would drop 25 percent by 2026 come true?

The cited observations do not establish the forecast's category-wide result. Alphabet reported all-time-high queries in Q1 2026, while Q2 Search revenue grew 17 percent. Revenue is not query volume, Google is not the entire category, and the original article appeared before 2026 ended.

If search volume went up, why did my organic traffic go down?

They measure different events. Your decline could involve rankings, query mix, result formats, tracking, or several changes at once. Investigate your page/query evidence rather than assigning the cause from an aggregate trend.

How big is AI referral traffic compared to search?

The cited Similarweb figure of 9.5 billion monthly visits is a twelve-month average for visits to AI platforms, not referrals out to sites. It cannot establish AI-referral size relative to search. Use comparable outbound-referral data or your property's observed sources for that question.

What does Similarweb's assistant share measure?

It measures visits to generative AI platforms. The July 2026 report puts ChatGPT at roughly 53 percent in May 2026, Gemini near 27 percent, and Claude near 9 percent. Those platform shares do not establish outbound-referral shares.

Should I trust AI referral conversion rate statistics?

Check the cohort, date range, defined conversion event, attribution method, and sample. A single anonymous site or unrelated channel averages cannot establish the rate for your business.

What actually works to earn traffic from AI answers?

Publish useful answers with checkable evidence and maintain discoverability, then measure citations, observed referrals and accepted outcomes separately. The linked manufacturer case records web leads of 43 in 2023, 119 in 2024 and 203 in 2025, with 450-plus labelled as a 2026 projection. The program combined several interventions, so those figures do not establish AEO's separate causal effect.

---

Sources

The prediction: Gartner, "Search Engine Volume Will Drop 25% by 2026" (February 19, 2024; Alan Antin, Vice President Analyst). Alphabet's actual results: Alphabet Q2 2026 earnings release, filed with the SEC (July 2026), with Q1 2026 query-volume commentary from the Q1 2026 earnings transcript. Google's rebuttal language, including the aggregate framing: PPC Land and Search Engine Land (July 2025), reporting Google's on-record statement responding to Pew Research Center's July 22, 2025 study. Generative AI platform traffic and share shift: Similarweb, "AI Search Stats in 2026" (last updated July 29, 2026; panel-based estimates, a vendor's own data with limited disclosed methodology). The roughly 7 percent AI conversion figure traces to a single site reported via The Digital Bloom, surfaced in a June 2026 aggregation by Demand Local, and is characterized here as unsuitable for benchmarking rather than repeated as one. Client figures are from work I ran directly, de-identified by agreement.

---

About the author

Jeff Brokaw is a CMO and Certified Chief AI Officer. The de-identified manufacturer case linked above documents the commercial program and its reported results.