Search has changed more in the past 24 months than in the decade before it. Google’s AI Overviews now appear on roughly 13.14% of queries according to BrightEdge’s June 2024 data, ChatGPT Search rolled out to all users in late 2024, and Perplexity reported handling over 250 million queries per month by mid-2024. The old playbook of ranking a blue link and winning a click describes maybe a third of what’s actually happening on a results page now. Everything else is some flavor of AI-mediated answer, and brands are scrambling to figure out what that means for the work.
When an AI assistant summarizes three sources into a single answer, two of those sources get cited and one gets ignored entirely. The cited ones are quietly compounding traffic while their competitors watch click-through rates erode and blame the latest algorithm update. It’s a slow-motion redistribution, and most marketing dashboards aren’t built to see it happening.
What changed in the search stack
Classical SEO was a contest of relevance and authority signals, fought with keywords, backlinks, on-page structure, and technical health. All of that still matters. It just now sits underneath a generative layer that decides what gets surfaced in answer boxes, chat responses, and AI Overviews, which means a page can do everything right by 2018 standards and still lose to a competitor whose content the model happens to find more quotable.
Authoritas, in a study published in March 2024, found that when AI Overviews appear, the traditional number one organic result gets displaced 62% of the time. A Pew Research report from July 2024 added that users who encountered AI summaries clicked through to source pages at roughly half the rate of users who only saw standard results. So the traffic isn’t disappearing evenly across the web. It’s concentrating among the small number of sources the models trust enough to cite, and the criteria for that trust overlap with traditional SEO without being identical to it. Structured data helps. Consistent entity references across the web help. Clear authorship helps. What seems to matter most, though, is whether a page actually answers a question instead of dancing around it for 800 words before getting to the point.
Data-driven growth strategies in an answer-engine world
The metrics that mattered in 2019 are not the metrics that matter now. Sessions and bounce rate and average keyword position still tell a story, but increasingly it’s a story about the wrong thing. Teams getting ahead are tracking citation share in AI responses, branded query volume after AI exposure, and assisted conversions from zero-click impressions, none of which show up cleanly in a default GA4 view.
Semrush released a dataset in late 2024 showing that brands cited in ChatGPT responses saw a 23% lift in direct and branded search traffic within 30 days of first citation, even when no click was recorded from the chat interface itself. Which is a strange and somewhat thrilling finding if you sit with it for a minute. Being mentioned by an AI is starting to function the way being mentioned in a major publication once did, a reputation event with a measurable downstream effect, except the publication is a model and the byline doesn’t exist. A piece of content might generate no clicks for a month and then drive a wave of branded searches when it finally gets ingested into a model’s retrieval index. Attribution windows built for paid social do not capture this at all, and marketing teams reporting only on last-click conversions are missing the most interesting thing happening in the channel.
The new content brief
Writing for AI retrieval is not the same as writing for human readers, but it overlaps more than skeptics expect. Models tend to extract and cite content that is factually dense, clearly attributed, and structured in a way that maps to common question formats. Long preambles get skipped. Vague claims get ignored. Specific numbers and named sources and direct statements get pulled into responses where vaguer copy doesn’t survive the summarization step.
That’s part of why publications leaning into reported data, original research, and named expert commentary have gained citation share even as their raw traffic has dipped. Stratechery, The Pragmatic Engineer, and a handful of trade publications in niches like logistics and biotech now show up disproportionately in AI Overview citations relative to their domain authority. For marketing teams trying to win an organic search traffic increase in this environment, the brief looks different than it did three years ago. Less SEO-template content built around keyword stuffing, more original data, more concrete examples, more willingness to take a position the writer might have to defend in public. Agencies that understand this distinction, including teams like ELK Digital Marketing, have been rebuilding content programs around citation-worthy assets rather than volume plays. The trade-off is fewer pieces published per quarter, but each one tends to earn more durable visibility, which is the kind of math that only works if leadership has the patience to wait out the lag.
Brand awareness through search is now upstream of the click
One of the stranger effects of generative search is that brand awareness through search has decoupled from clicks. A user can encounter a brand name three times in AI responses, never click a link, and then type the brand directly into Google a week later. From a classical analytics view, that journey starts at the branded search. In reality it started in the chat window, and there’s no clean way to draw the line between them in a report.
This changes how marketing budgets get allocated, or at least how they should. Top-of-funnel content that historically struggled to justify itself on a direct-response spreadsheet is becoming easier to defend, because it’s doing work that shows up later in branded demand rather than immediate sessions. Mid-funnel content built around commercial intent keywords is getting squeezed in the other direction, because AI assistants are increasingly answering those queries directly and the user never makes it to the page that was optimized to capture them. The brands adapting fastest are treating their content libraries the way a PR team treats a media list, asking what is being said about the brand in the places where decisions get made, including AI surfaces, instead of counting indexed pages.
What to ask of an agency partner
The market for digital marketing agency services has fragmented into roughly three camps: traditional SEO shops still selling 2018 deliverables, generative-AI-first newcomers with strong theory and thin track records, and a smaller group that has spent the last 18 months rebuilding methodology around how search actually works now. The first two are easier to spot than the third, which is part of the problem.
A few questions separate the useful agencies from the rest. How do they measure citation presence in AI Overviews and chat responses? What is their process for entity optimization across third-party sources, not just the client’s domain? Can they show a case where a piece of content drove branded demand without driving direct organic clicks, and can they explain how they knew? How do they handle the gap between what ranks and what gets cited? Vague answers to any of those should worry a buyer. The discipline has gotten technical again in a way it hasn’t been since the early Panda and Penguin years, and slide decks don’t carry the same weight they used to.
The honest assessment
Traditional SEO isn’t dead. Backlinks still matter, site speed still matters, and schema markup matters more than it ever has. What’s changed is the layer sitting on top of those fundamentals, and the uncomfortable fact that ranking and visibility are no longer the same thing, which is a hard sentence to say to a client who has been paying for ranking reports for ten years.
The brands winning right now accepted this earlier than their competitors and rebuilt their measurement, content, and partnership decisions around it. The ones still arguing about whether AI search is real are losing share they probably won’t get back, at least not without spending a lot more to recover it than it would have cost to adapt in the first place. AI-powered digital marketing, stripped of the marketing language wrapped around it, is mostly just the recognition that the inputs to visibility have multiplied. The teams treating that seriously are pulling ahead, and the ones still pretending it’s 2019 are going to spend the next year wondering why their traffic charts look like that.
