// THE 2027 CMO / AI SEARCH TRAFFIC
Your Search Traffic Is Not Coming Back
In early 2024 Gartner put a number and a date on the end of search. We are inside the deadline year now. The number is nowhere in sight.
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.
We are eight months into the deadline year. Here is what the scoreboard says.
The collapse has not come
Alphabet reported query volume at an all-time high in the first quarter of 2026, crediting its own AI features for driving usage rather than cannibalizing it. In the second quarter, Google Search and other revenue came in at $63.3 billion, up 17 percent year over year.
Two honest caveats on that, because I would rather you hear them from me than from whoever pushes back on this in your next planning meeting.
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.
Even granting all of that, nothing in the public record looks like a category losing a quarter of its volume. Search advertising revenue growing 17 percent is not the financial signature of collapsing demand. The year is not over, so the prediction has not technically failed yet. It would now need a historically abrupt reversal in four months to land, and no serious observer is forecasting one.
If you built a 2026 plan around search volume falling off a cliff, the cliff is not there. Now the part that should worry you.
Your clicks fell anyway
Search volume going up and your organic traffic going down are not contradictory findings. They are the same finding described from two different altitudes, and the gap between them is the entire planning problem for the next two years.
More people are searching than ever. Fewer of those searches end at a website. The query gets asked, the answer gets assembled on the results page, and the session ends there. The demand did not disappear. The visit did.
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.
I covered that click-behavior research and the fight over its methodology in the dark AI traffic piece, where I also took apart a separate and much weaker statistic that gets repeated alongside it. I am not relitigating either here. The point for a CMO is narrower: the one number Google chose to defend is the one number that has nothing to do with you.
Where the attention actually went
The demand moved to a place most marketing plans still treat as a rounding error. Similarweb data, last updated July 29, 2026, put generative AI platforms at an average of 9.5 billion monthly visits across the twelve months from June 2025 to May 2026, up roughly 70 percent year over year, with unique visitors up 57 percent to 655 million.
Three things about that number matter more than its size. It is a twelve-month average rather than a current run-rate, so the exit velocity is higher than the headline. It measures visits to the assistants themselves, not referrals out to your website, which is a much smaller figure again, and anyone quoting 9.5 billion as their AI traffic opportunity is quoting the wrong denominator. And it is a vendor's own panel estimate with limited published methodology, which puts it in the same category of evidence I spend the next section warning you about. Underneath all that, the composition is churning violently.
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.
If you built an answer-engine strategy in early 2025 optimized around how ChatGPT specifically retrieves and cites, roughly half of the audience you were optimizing for is now somewhere else. That is not a reason to skip the work. It is a reason to build it in a way that does not depend on one vendor's retrieval quirks, and to re-check the split every quarter instead of every planning cycle.
A warning about the statistics you are about to be sold
Every consultancy in this space now has a conversion statistic proving AI traffic is the best traffic you can get. Check them before you repeat them, because the sourcing is frequently worse than the confidence.
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.
The directional claim underneath it is probably right, and I would defend it: a visitor arriving from an assistant has usually had their question pre-qualified before they ever reach you, so they should convert better than a cold organic click. That direction is worth planning around. The decimal is not. If you put a specific conversion percentage for AI referrals in front of your board and someone asks where it came from, you need a better answer than a vendor blog post citing one anonymous site.
This is the same discipline problem I wrote about in the agentic commerce piece, where the two most-quoted statistics in the category both fall apart under a source check. The pattern is consistent enough now to be a rule: in an emerging channel, the confidence of a statistic is inversely correlated with the quality of its sourcing.
What I actually did about it
I run this work rather than describe it, so here is the version with receipts attached.
For a domestic defense and advanced materials manufacturer, I rebuilt the demand engine around being the answer rather than being a result. That meant a full rebrand, a multi-brand and multi-domain search architecture, and content engineered specifically to be extracted and cited by answer engines, not just ranked by a crawler. Organic traffic went to roughly twenty times where it started. Inbound web leads moved from about 43 per year before the work to a range of 25 to 40 per month after the launch. The full build is in that case study.
The mechanism is not exotic. Answer engines cite sources that state a specific claim cleanly, back it with something checkable, and belong to an entity the model can identify with confidence. Most corporate content fails all three tests at once: it hedges the claim, sources nothing, and comes from a brand with no clear entity footprint. Fixing that is unglamorous and it works.
What it is not is a replacement for measurement. Every recovered visit needs to land somewhere you can see it, which is a harder problem than it sounds, because a large share of assistant-referred visits arrive carrying no referrer at all. That machinery is its own subject, covered across the AI attribution series, and the specific limits of Google's native tooling are in the GA4 channel breakdown.
How to plan against this
Four things, in the order I would do them.
Stop planning against search volume. It is going up and it is telling you nothing. The metric that matters is your qualified sessions and what they do, not the size of the pool they were drawn from. A plan built on total query volume was always measuring the ocean instead of the fish.
Separate the two declines. Organic sessions falling because you lost rankings is a different problem with a different fix from organic sessions falling because the answer now resolves on the results page. Most teams have both happening simultaneously and treat them as one number. Split them before you spend against either.
Build for citation, not just position. Ranking first for a query whose answer gets summarized above you is a decreasingly valuable asset. Being the source that summary cites is the durable one. Those require overlapping but genuinely different content decisions, and only one of them is what your agency is currently optimizing.
Instrument before you scale. If you cannot see which assistant sent a visit and what it did afterward, you will not be able to defend the budget line twelve months from now when someone asks what answer-engine work returned. The measurement is the harder half and it is the half everyone skips.
The honest version
Gartner got the mechanism right and the number wrong. Generative AI did become a substitute answer engine. It just did not take volume away from search, it took the click away from the publisher, and those look identical in a strategy deck while requiring completely different responses in an actual plan.
Search traffic as a category is not coming back to the shape it had in 2022, and no amount of technical SEO restores it, because nothing is broken. The system is working as designed. It just stopped routing through your site on the way to the answer.
The companies that will be fine are the ones that stop trying to recover a click and start earning a citation, then build the measurement to prove it happened. That is the whole job now, and it is a commercial job before it is a technical one, which is the throughline of AI Commercialization: the complete guide.
Frequently asked questions
Did Gartner's prediction that search volume would drop 25 percent by 2026 come true?
No. Gartner predicted on February 19, 2024 that traditional search engine volume would fall 25 percent by 2026. The opposite happened at the volume level. Alphabet reported query volume at an all-time high in Q1 2026, and in Q2 2026 posted Google Search and other revenue of $63.3 billion, up 17 percent year over year. One fair caveat: Gartner's prediction covered the search category, not Google alone, and Google is a proxy for it rather than the whole of it. With four months left in the deadline year, the prediction would need a historically abrupt reversal to land.
If search volume went up, why did my organic traffic go down?
Because volume and clicks are different metrics and they have decoupled. More searches are happening than ever, and a growing share of them resolve inside the results page without anyone visiting a website. Google's own rebuttal to critical research is precise on this point: it says it has not observed significant drops in aggregate web traffic. Aggregate across the entire web says nothing about any individual site, which is the number a CMO actually has to plan against.
How big is AI referral traffic compared to search?
Still small in absolute terms, and growing fast. Similarweb data, last updated July 29, 2026, put generative AI platforms at an average of 9.5 billion monthly visits across June 2025 to May 2026, up roughly 70 percent year over year, with unique visitors up 57 percent to 655 million. Two caveats: that is a twelve-month average rather than a current run-rate, and it measures traffic to the AI platforms themselves, not referrals out to your site, which is a smaller number again. Treat AI referral volume as a fast-growing minority channel, not a replacement for search.
Which AI assistant sends the most referral traffic in 2026?
ChatGPT still leads, but its dominance has eroded sharply. Similarweb's July 2026 data shows ChatGPT falling from roughly 76 percent of generative AI visits in June 2025 to about 53 percent in May 2026, while Gemini climbed from under 9 percent to roughly 27 percent and Claude grew from about 2 percent to close to 9 percent. Any strategy built around a single assistant is being repriced every quarter.
Should I trust AI referral conversion rate statistics?
Verify each one before you put it in a board deck. A widely repeated figure claiming AI referral traffic converts around 7 percent traces back to a single site's result reported by one agency, not a multi-site benchmark with disclosed methodology. Directionally, AI referrals do appear to arrive with higher intent because the assistant pre-qualified the question. That direction is defensible. A specific decimal is not, unless the source publishes its sample and method.
What actually works to earn traffic from AI answers?
Publish content that answers the specific question a buyer asks, structure it so a machine can extract the answer cleanly, and build entity clarity so the assistant knows who you are and what you are authoritative on. Then measure citations and referrals directly rather than inferring them from a drop in Direct traffic. On a defense and advanced materials manufacturer I worked with, that approach took organic traffic to roughly 20 times its starting point and moved inbound web leads from about 43 per year to a range of 25 to 40 per month after launch.
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 coverage. 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 sitting CMO and Certified Chief AI Officer who ships AI in production, not slideware. He built the answer-engine demand program described above for a domestic defense manufacturer, built the precision-measurement layer that scores visitor-identification accuracy rather than coverage, and built the commercial engine behind $185M in new-business revenue. He has been building AI systems commercially since 2016.