You should not plan next year's content budget the same way you planned last year's, because search itself did not hold still this year. People are typing longer questions, AI summaries are answering a real share of searches before anyone clicks a link, Google is rolling AI search out to more countries and languages, and AI platforms are becoming a measurable source of referral traffic in their own right. This is not a forecast that organic search is collapsing, but it is a claim worth taking seriously on its own terms. The environment your content budget planning depends on is moving, and a plan built only on last year's numbers assumes it is not.
Evidence grade: strongly supported that you should plan for continued change, mixed on how big or which direction any single future change will run.
Several independent studies, observed browsing data, and Google's own product announcements point the same way. None of them tell you exactly what next year looks like. That gap between "things are changing" and "here is what will happen" is the whole reason a search behavior change budget matters more than a static one.
What the claim actually says
The claim here is not that last year's search data is wrong. It is that last year's data describes last year's search environment, and that environment has already moved more than once within a single planning cycle. Getting the content marketing budget AI search question right starts with separating those two ideas, because conflating them is what leads a planning conversation astray. A content budget planning process built entirely on last year's keyword volumes, rankings, and click-through rates treats the path from a query to a visit as fixed. The evidence below shows that path is not fixed. It is worth separating from the start: this is a case for building room into the plan, not a case for predicting a specific percentage decline in traffic or a specific rise in AI referrals. Nobody in this dossier makes that prediction responsibly, and neither should you.
The evidence that search behavior is changing
The strongest piece of evidence in this review is Pew Research Center's July 2025 analysis, which watched what people actually did rather than asking them to recall it. Pew studied browsing data from 900 U.S. adults across March 2025, covering 68,879 unique Google searches. An AI-generated summary appeared on 18% of them.
The behavior split when a summary showed up. Users clicked a traditional search result on just 8% of visits when an AI summary appeared, compared with 15% when it did not. Only 1% of visits to a page with a summary produced a click on a link inside the summary itself. And 26% of visits to a summary page ended the browsing session there, against 16% for pages with only traditional results. This is observational data, not a controlled experiment, so it shows association rather than proof of a single cause. But it is a direct look at real behavior on real search pages, which makes it harder to dismiss than a survey about what people say they do.
The query itself is changing shape too. Pew found that one- or two-word searches produced an AI summary only 8% of the time, while searches of ten words or more produced one 53% of the time. Searches that opened with a question word triggered a summary 60% of the time. A keyword plan anchored to short, isolated search terms describes a narrowing slice of how people actually search now. BrightEdge, a vendor with its own analysis, reported a similar pattern from a different angle. AI Overview presence in longer queries rose by up to 100% between September and December 2024, and AI Overviews showed up in 25% of searches using eight or more words. That is proprietary vendor research, not independent measurement, so treat it as directional rather than definitive, but it points the same way as Pew's data.
Google has also been explicit that this is a deliberate product direction, not a side effect. Google's own materials describe AI Overviews as generative summaries built on a customized Gemini model, layered alongside existing Search ranking systems, with links to supporting pages. By May 2025, Google said AI Overviews had reached more than 200 countries and territories and more than 40 languages, and reported that AI Overviews drove more than a 10% increase in Google usage for the query types where they appeared, based on an internal cohort comparison running from September 2024 through April 2025. That is a company reporting on its own product, so it should be read as a self-reported metric, not outside confirmation. Google's AI Mode goes further still, using what it calls query fan-out to split one question into several subtopic searches run on the user's behalf, and Google says people are already asking longer, more complex, more conversational questions inside it.
None of this describes a search environment that stopped changing after one big update. A July 1, 2024 SparkToro analysis of Datos clickstream data, covering September 2022 through May 2024, found that 58.5% of U.S. Google searches and 59.7% of EU Google searches ended without any click at all. For every 1,000 U.S. searches, only 360 clicks reached a site that was not Google's own property, and the share of clicks reaching the open web sat at a historic low even before AI Overviews rolled out broadly. Zero-click search was already a real constraint on organic discovery before the current wave of AI summaries. That is worth remembering, because it means this is a continuation of a longer shift in how a search maps to a visit, not a single event.
Search discovery has not simply gone quiet either. It has partly moved somewhere new. Adobe's May 2025 analysis, drawn from its own Digital Insights data across July 2024 through February 2025, found that AI-driven referral traffic to U.S. sites grew more than tenfold over that period, with retail traffic up 12 times, travel up 17 times, and banking-site visits up 12 times. Adobe also reported that by February 2025, AI-referred retail visits carried a 23% lower bounce rate and lasted 41% longer than other traffic, and that travel referrals from AI sources generated 80% more revenue per visit than non-AI traffic. Similarweb's later report, published in December 2025, estimated more than 1.1 billion AI-platform referral visits in June 2025 alone, up 357% year over year, with roughly a 7% conversion rate on transactional sites in its own measurement system. Both figures come from vendors measuring their own data with their own methods, so read them as evidence of rapid, real growth rather than a universal number you can apply to your own funnel.
The evidence against a simplistic conclusion
Eighteen percent is a meaningful share of searches, but it is not most of them. Four out of five Google searches in Pew's March 2025 sample showed no AI summary at all. Writing as though traditional search has already been replaced would overstate what a single study, however well designed, can support.
AI referral quality is uneven, not uniformly better. Adobe found that AI traffic converted 43% worse than other traffic in July 2024, and while that gap had narrowed to 9% by February 2025, it had not closed. Performance varied by category too: Adobe reported stronger results in research-heavy categories like consumer electronics and jewelry, and weaker results in apparel and grocery. Similarweb's 7% conversion figure applies to transactional sites inside its own measurement system, which may look nothing like a B2B SaaS funnel.
Trust in AI summaries is mixed as well. A separate Pew survey from October 2025, covering 5,153 U.S. adults surveyed in August 2025, found that 65% had encountered AI summaries at least sometimes, and 45% saw them often or extremely often. But among people who had seen them, only 53% expressed even some trust, and just 6% trusted them a lot. People are using these summaries more than they are trusting them, and that gap matters for anyone assuming AI answers will simply substitute for a site visit at the same rate people once clicked an organic result.
It is also worth being precise about what kind of evidence each source actually is. Google's statements about how AI Overviews and AI Mode work are confirmed product mechanics. Google's usage-increase figure is a self-reported metric from an internal experiment, not independent proof. Pew's click and session data are direct observations of real behavior, but they are correlational, not causal. Adobe, BrightEdge, and Similarweb are vendors reporting on their own proprietary datasets, useful as directional signals but not neutral industry benchmarks. None of these sources cover the same dates, countries, devices, or query sets, so stacking them into one combined market-size number would misrepresent what any of them actually measured.
What the primary sources actually establish
Strip this down to what is confirmed versus what is repeated as received wisdom, and the confirmed list is shorter but still substantial. Google has confirmed that AI Overviews generate summaries with supporting links, that they run on a customized Gemini model alongside existing ranking systems, that AI Mode breaks a question into subtopics through query fan-out, and that the feature had reached more than 200 countries and 40-plus languages by May 2025. Those are rollout and mechanics facts, not predictions about your traffic.
Pew's metered browsing study is the strongest observational evidence of changed behavior, because it watched real users on real search pages instead of surveying them, and it showed lower traditional-click rates and more session endings when a summary appears. That is stronger footing than an opinion poll, though it still cannot prove that AI summaries alone caused every difference observed.
Everything from Adobe, BrightEdge, and Similarweb sits a step below that. These are real measurements from real proprietary datasets, useful for showing scale and direction, but bound by each vendor's own definitions and coverage. Cite them by name and date if you use them in your own planning conversations, and resist the temptation to average them into a single tidy statistic.
Verdict: what content budget planning should assume
The defensible conclusion is not a specific forecast. It is a planning posture. Last year's search data is still a useful historical baseline, it is just not sufficient as the entire model for next year's demand or content return. A content budget should assume the query mix, click-through rate, and discovery path that worked last year will not necessarily hold, because AI summaries, AI Mode, and AI-platform referrals are all still moving. Visibility can now happen through a traditional result, an AI summary citation, or an AI-platform referral, and a plan that only tracks the first of those is measuring a shrinking part of the picture.
This is really the content marketing budget AI search question in practice, and it matters more for a lean team than a large one. A small SaaS company often plans content around a short list of keyword volumes and an expected ranking outcome, because that is what a spreadsheet and a couple of hours a week can support. That approach carries real planning risk right now, because the same query can now trigger a traditional result page or an AI-generated answer, a page can earn visibility as an AI source without ever producing a traditional click, and AI referrals bring different engagement and conversion patterns than organic search did. None of that means SEO stopped working. It means the return on a content budget planning cycle depends on more surfaces than it used to, and a plan that only measures one of them will misread its own results.
The practical shift is to build in the capacity to observe and adjust, not to lock a full year of spend to one channel's numbers from twelve months ago. That means keeping some budget and attention free to track how your own pages show up across search surfaces during the year, not just at the planning meeting. Watching where a brand actually appears in AI answers, not just where it ranks in a results page, is part of what DeepSmith's AI search analytics tracks alongside its content production tools, precisely because content planning and AI visibility measurement now depend on the same underlying data.
What would change this verdict
This verdict would weaken if repeated, independent, cross-engine studies over several planning cycles showed that query formulation, result-page composition, click behavior, and referral patterns had settled into a stable pattern. It would strengthen if future independent research kept finding continued movement in AI-summary prevalence, query length, click behavior, and AI-platform referrals across more markets and device types. Right now, every data point in this review points toward continued movement, not settlement, which is the whole basis for planning around change rather than around a fixed baseline.
If you're weighing how much of your own content plan should assume next year looks like this year, treating that as an open, testable question rather than a settled one is itself the safer bet.
