Search is no longer a list of ten blue links. It's a conversation, a summary, a shopping carousel, a voice response, and sometimes still a list of ten blue links. For brands trying to grow in 2024 and 2025, that fragmentation has forced a real rethink of what marketing teams actually do every week. AI-powered digital marketing strategies sit in the middle of that shift, not because every agency wants to slap "AI" on a deck, but because the underlying mechanics of discovery have changed.
Google's AI Overviews started rolling out broadly in May 2024. By the end of that year, Gartner was predicting a roughly 25 percent drop in traditional search engine volume by 2026 as users moved to generative answers. Whether that exact number lands is beside the point, and frankly Gartner predictions have a mixed track record on this kind of thing. But the directional trend is real, and the marketing playbook built around blue-link clicks is showing its age.
What changed when generative search arrived
The old SEO model assumed a fairly stable equation: publish a useful page, earn links and authority, rank for a query, capture the click. Each step was measurable, each step had a known set of levers, and a competent team could forecast results with reasonable confidence.
Generative search blows up most of that. A user asking ChatGPT or Perplexity for "the best CRM for a 12-person sales team" may never see a ranked page at all. They see a synthesized answer, possibly with three citations, possibly with none. The brand that gets named wins. The brand that ranked third on Google for the same query may not even register, which is a strange and slightly demoralizing thing if you spent the last two years building those rankings.
Search ranking improvement now has to be measured across surfaces, not just on google.com. A BrightEdge analysis from late 2024 found that AI Overviews were appearing on the majority of informational queries in industries like healthcare and finance, and that the brands cited inside those overviews often were not the top organic results. Ranking number one is no longer a guarantee of being quoted.
The practical implication: content that wins citations in generative answers tends to be specific, structured, and source-rich. Vague thought leadership loses. Pages with clear definitions, comparison tables, original data, and dated updates win.
The data layer that makes AI marketing actually work
Most of the noise around AI in marketing focuses on output. Write a blog post in 30 seconds. Generate 50 ad variants. Spin up a landing page from a prompt. That's the visible layer, and it's the least interesting one.
The layer that drives real results sits underneath. Data-driven growth marketing in 2025 looks less like A/B testing headlines and more like building a feedback loop where customer signals, search behavior, ad performance, and content engagement all feed the same model. Teams that have invested in clean first-party data, properly tagged events, and a usable customer data platform are running circles around teams that are just prompting ChatGPT for blog topics.
McKinsey's 2024 State of AI work found that companies they categorized as "AI high performers" in marketing were several times more likely to have unified their customer data across channels than peers. That is not a coincidence. AI models, including the ones inside ad platforms like Meta Advantage+ and Google Performance Max, perform dramatically better when fed structured, deduplicated, consented data. Garbage in, expensive garbage out.
The agencies that have adapted fastest tend to treat data infrastructure as the first deliverable, not the last. Firms like ELK Marketing have built practices around aligning analytics, content, and paid media against shared performance signals, which is increasingly the table stakes for organic search optimization that actually compounds. Without that foundation, AI features just generate noise faster.
SEO and content strategy in an answer-engine world
SEO and content strategy used to operate on roughly month-long cycles. Identify a keyword, brief a writer, publish, wait for rankings, iterate. That cadence is too slow now, and the metrics it produced are getting harder to trust. Organic click-through rates on informational queries have been falling since AI Overviews launched, and the Ahrefs data circulating in mid-2024 suggested meaningful CTR drops on queries where an AI summary appeared above the fold.
The response is not to abandon content. The response is to publish content that does different work, which mostly means writing for two audiences at once: a language model that needs verifiable specifics to feel safe citing you, and a human reader who showed up because the model already mentioned your brand by name.
Earning model citations is its own discipline. Clear factual claims, original statistics, named sources, and structured data markup all help. Pages that read like edited editorial with verifiable specifics get pulled into generated answers more often than pages full of hedged generalities, which is probably good news for journalism and bad news for the "ultimate guide" industrial complex.
Brand searches matter more than they did a year ago. When a model recommends three options, the user often follows up with a brand-name query on Google or directly on the brand's site. Brand awareness through PR, podcasts, YouTube, and social is now a direct input to organic conversion, not a separate funnel stage. Demand creation feeds demand capture in a tighter loop than most attribution models can handle.
And refresh cycles have to be shorter. Pages dated 2022 are routinely passed over for pages updated in the current year. A quarterly content audit with a real budget for updates, not just net-new publishing, has become non-negotiable for sites that want to defend rankings.
Paid channels are getting smarter and less transparent
Paid media has absorbed AI faster than any other marketing function, partly because the platforms gave teams no choice. Performance Max, Advantage+ Shopping, and similar products hand most of the targeting and creative optimization decisions to the platform's models. Advertisers set goals and feed signals. The black box does the rest.
This works remarkably well when the inputs are good. Tinuiti's quarterly benchmark reports through 2024 consistently showed that Performance Max campaigns with strong first-party audience signals outperformed those relying on platform defaults, often by double-digit ROAS margins. It works poorly, sometimes disastrously, when the inputs are weak or the conversion tracking is broken. There is no middle ground anymore, which is one of the less-discussed consequences of handing creative and targeting to a model at the same time.
The uncomfortable trade-off is transparency. Marketers know less about which keyword, audience, or placement drove a sale than they did five years ago. The platforms have decided that aggregate performance matters more than line-item visibility, and they are probably right from a pure performance standpoint. But it makes attribution arguments harder and puts more weight on incrementality testing, media mix modeling, and clean experimental design.
Teams that lean into that reality, building out proper geo-holdout tests and lift studies instead of arguing about last-click reports, get to actual answers faster.
What the next 12 months probably look like
A few things seem reasonably safe to predict. Generative search will keep expanding, and the share of zero-click queries will keep rising. Brands with strong topical authority and consistent citations will benefit. Brands relying on thin content and aggressive link building will struggle, and probably blame Google for it.
The gap between agencies that understand the data and infrastructure side and agencies that just resell AI-generated content will widen quickly. Clients are already getting better at telling the difference, partly because the cheap version is so easy to spot. Measurement frameworks will get messier before they get cleaner. Expect more investment in MMM, more incrementality testing, and continued frustration with attribution dashboards that promise certainty they cannot deliver.
And brand, the unfashionable word that everyone in performance marketing used to roll their eyes at, keeps coming back into the conversation. Generative engines reward known entities. Customers ask language models for recommendations using brand names they already trust. The flywheel between brand and performance is tightening, not loosening, which is going to be uncomfortable for any team that built its career on the opposite assumption.
AI-powered digital marketing is not a product anyone can buy off a shelf. It is a working discipline, and the brands getting it right in 2025 are the ones treating the integration as the actual job, with the AI tooling as accelerant rather than strategy.
