The Companies Winning at AI Search Mostly Just Did the Boring SEO Work Years Ago

When buyers started asking ChatGPT and Perplexity which software to buy instead of opening ten browser tabs, a lot of marketing teams braced for a reckoning. So far, the reckoning has been uneven and not in the way many expected. The tools doing the recommending don't invent answers. ChatGPT with Browse, Perplexity, and Google's AI Overviews all sit on top of a retrieval layer that indexes and ranks web pages by roughly the same criteria search engines have used for two decades: authority, relevance, the quality of the content and its structure, and signals of endorsement from elsewhere on the web. The interface in front of the buyer is new. The machinery underneath is mostly not. That distinction matters, because it reframes a question companies have been treating as existential. The shift is not about your content being optimized for AI, but about whether you did the slower work of building credible, well-structured content in the first place, and a lot of organizations didn’t. What actually changed The behavioral change is real and worth taking seriously. A buyer evaluating enterprise software a few years ago typically moved through several pages across multiple sessions before forming a shortlist. Increasingly, that same buyer asks one conversational question, reads a synthesized answer, and treats it as a starting shortlist. Traditional search still works; it's just no longer the only route in. The consequence is that the threshold for being considered has moved earlier in the process and become less forgiving. If a model doesn't surface you in its summary, you may not enter the evaluation at all, and the buyer may never know you were an option. The brands showing up in those answers tend to be the ones that built content addressing the questions buyers ask. That was always supposed to be the point of content marketing. AI search has just raised the cost of having skipped it. The signals, and how much you actually know about them People who study this can name the signals AI retrieval appears to favor, though it's worth being honest that much of this is inference from observed behavior rather than disclosed mechanics. The major AI companies don't publish how their retrieval ranks sources, and the systems change frequently. What practitioners consistently point to: Content that answers a specific question directly. FAQ sections, explainer pages that state the question and resolve it up front, technical documentation organized around the questions an evaluator would actually ask. These were the same formats rewarded by featured snippets and conversational search well before generative AI. Attribution. Pages that cite primary research, link to authoritative sources, and reference verifiable data appear to be weighted more favorably than pages making bare assertions, and pages that are themselves cited by credible external sources more favorably still. This mirrors a much older principle from editorial and academic publishing: credibility comes partly from what you reference and what references you, not only from what you claim. Language that matches the buyer's problem, not the seller's product. The gap between how a company describes what it sells and how a customer describes what's wrong is where a lot of content fails to get found. Third-party and community validation. This is the most striking and the most cited finding, and also the one that deserves the most scrutiny. A June 2025 Semrush analysis of roughly 150,000 citations found that Reddit was the single most-referenced domain across the major engines, at about 40 percent, ahead of Wikipedia and YouTube. The 5W "State of AI Citations" synthesis , which rolls up more than 680 million citations from six separate studies, reached the same headline conclusion: Reddit first across every major engine. None of these are new recommendations. They were defensible content strategy before anyone was optimizing for language models. What's changed is that their absence now has a visible, measurable cost. The case for patience over spend There's a structural reason some teams are calmer about this than others. Paid distribution buys traffic only while it's funded; pause the budget and the traffic stops, with no residual. Content and the authority it accrues behave differently: they compound, and they don't reset when investment lapses. As more buying decisions begin inside AI answers, each citation a brand earns adds to its standing in the layer where shortlists are increasingly forming before any salesperson is involved. That's the optimistic framing, anyway, and it's worth flagging that it's a framing the SEO industry has an interest in promoting. The honest caveat is that "early positioning compounds" is a reasonable hypothesis, not a settled fact, and the maturity point everyone is positioning for is still being defined in real time. What this looks like in practice Stripped of the urgency, the practical advice is unglamorous. Audit your most important pages and ask whether each one directly answers the questions an evaluator would have before recommending a purchase; close the gaps where the answer is buried or missing. Build content for the comparison moment, the head-to-head, the "best option for X" query because that's where both AI retrieval and conversion concentrate. Check how your content handles claims, and whether anyone credible is referencing you. And track AI citation volume as a metric separate from search ranking, because a page can rank well and still be invisible in AI answers, or vice versa. The throughline is almost anticlimactic. The retrieval layer rewards much of what good content infrastructure always produced. Buyer behavior changed; the fundamentals largely didn't. For the organizations that did the work, answered real questions, built authority, earned outside references, the transition reads more like continuity than upheaval. For the ones that didn't, it reads like a bill coming due.