A practical guide for marketing directors and CMOs who built their inbound on Google—and are watching the traffic curves flatten.
How search worked, and what's changed
For most of the past decade, the path to being found online ran through Google's ten blue links. A company published useful content, earned backlinks over time, optimized its pages for the right queries, and climbed the rankings. Traffic came, leads followed. The relationship between content investment and inbound traffic was imperfect but legible—you could look at what ranked and understand roughly why.
Search has quietly split in two. The familiar version still exists—a list of links, ranked by relevance. But alongside it, a different system now handles the research-heavy, high-consideration questions your buyers are actually asking. They get a paragraph answer at the top of the page, assembled by AI from dozens of sources and presented as a single authoritative response. Google calls this AI Overviews. ChatGPT, Perplexity, and Claude each have their own version. According to BrightEdge, AI Overviews now appear in 48% of Google searches—up 58% year-over-year.
This new reality is impacting the value of traditional rankings. According to 5W PR's GEO research, the overlap between what ranks in Google and what gets cited in AI-generated answers has dropped from 70% to below 20%. Two years ago, a page-one ranking came with a roughly 70% chance of also appearing when AI answered a related question. Today that chance is less than one in five.
Why SEO disciplines don't transfer cleanly
Traditional SEO was built around relevance—the idea that the most useful page for a query should rank highest. The discipline developed to signal relevance: the right keywords in the right places, enough authoritative backlinks, page structure that search crawlers could read. It worked because Google's job was to find the best page and surface it.
AI search systems are doing something different. While traditional search surfaced a page and let you read it, AI search surfaces an answer—assembled from multiple sources, presented as a single authoritative response. The sources they draw from are chosen based on whether they can be quoted with confidence.
A page can rank because it covers a topic thoroughly, uses the right terms, and has accumulated authority over time. None of that guarantees it will be cited, because citation requires something ranking doesn't: a clear, specific, attributable claim that the AI can incorporate into a coherent answer without distorting it.
The content that gets cited tends to make arguments, not summaries. It uses specific data rather than general assertions. It's written for a reader who has a real decision in front of them, not for an algorithm assessing topical coverage. And it holds a position—the kind of position a named person or organization would be willing to stand behind—rather than presenting balanced information from multiple sides. AI systems are quite good at identifying content that sounds authoritative but takes no real position, and they pass over it.
This is why some companies that invested heavily in AI-assisted content production in 2024 and 2025 are now puzzled by their traffic results. The volume went up and the rankings held or even improved. But the AI-driven traffic didn't follow, because the content produced was optimized for the old system and remains invisible to the new one. The content may be fluent and comprehensive, but it lacks a point of view.
What the companies holding their visibility are doing differently
The organizations that are maintaining or even growing their share of AI-driven traffic have generally made four adjustments that their competitors haven't.
- Establish a point of view before producing content at volume. Most companies have a brand voice guide that describes tone and style. What they're missing is a set of actual positions—specific beliefs about how their industry works, what buyers in their market consistently misunderstand, what the smart approach looks like and why. Without this, every piece of content sounds interchangeable with every other piece of content in the space, and AI systems treat it that way.
- Write for decision-stage questions rather than informational ones. When someone asks an AI a question, they're often in the middle of evaluating something—which means content that addresses specific, decision-stage questions tends to serve the moment more directly than broad informational coverage. AI systems cite what's useful to the query at hand, and late-stage questions tend to be more specific, which plays to the strengths of content with a clear position.
- Make specific, verifiable claims. AI systems favor content that cites data, references named studies, and makes assertions that can be checked. A statement like "many companies are struggling with this" gives an AI nothing to work with. A statement like "only 19% of content teams track AI-specific KPIs, according to Digital Applied's 2026 benchmarks" is quotable—it's specific, it has a source, and it can be incorporated into an answer without distortion.
- Publish consistently to build a picture of your expertise over time. A single well-written article doesn't establish topic authority for AI search. A body of work that covers a niche from multiple angles, updated as the topic evolves, is what gets a brand recognized as a source worth citing. AI citation sources are volatile in a way that traditional search rankings never were—a brand that appears consistently in AI answers one month can drop significantly the next—which makes this an ongoing discipline, not a one-time investment.
What this means for a content program in practice
The practical implication is that ranking content and citable content are different things, and teams that have optimized for one are not automatically producing the other. Ranking content covers a topic thoroughly, targets high-search-volume keywords, and is structured for crawlers. Citable content takes a position, uses specific evidence, and is written to satisfy a skeptical reader who wants to know whether this source actually understands the problem—not just whether it mentions the right terms.
For most marketing teams, the adjustment isn't about producing more or producing less. It's about being more deliberate at the point where content strategy gets made. The question to ask before a piece of content is commissioned is this: "What do we actually think about this, and can we say it specifically enough that an AI system could quote us?"—rather than the usual "what keyword are we targeting?" If the honest answer is that the company doesn't have a clear position on the topic, that's worth knowing before the content is produced.
Where to start
The most useful starting point is a conversation with your leadership team, your product people, and your most experienced customer-facing staff about what your company believes that your competitors aren't saying clearly.
What does your team know about your buyers' problems that most vendors in your space are getting wrong? What advice do you give clients that consistently surprises them? What position would you be willing to defend in public, with your name on it? The answers to those questions are the raw material for content that gets cited. The work of turning those positions into a consistent body of content is where a disciplined content operation earns its keep.
What Saltbox AI does
I help mid-size companies build content programs structured for the way AI search works—content with a clear point of view, anchored in your organization's expertise, written to be cited rather than just indexed, and structured to be found and recommended by the AI systems your buyers are already using to shortlist vendors.
If your traffic curve is flattening and you're not sure what's driving it, I'd like to show you what a different approach looks like.
← Back to Blog