Every few weeks I get on a discovery call with a CMO who is frustrated. They hired an SEO agency two years ago. The agency delivered: rankings improved, organic traffic held, the technical audit came back clean. And yet, when their prospects ask ChatGPT or Perplexity about the problems the company solves, the brand doesn't show up. So now they're trying to decide whether to fire the agency, double down, or try something else entirely.
My answer is usually: the agency probably did their job. The problem is that ranking and being cited by AI assistants are two different problems, governed by different mechanisms. AI search visibility isn't a SEO score with a better benchmark. It's a different measurement altogether.
The gap that most SEO audits miss
Traditional SEO operates on a specific model. A crawler indexes your page, an algorithm evaluates authority signals (backlinks, domain rating, title tags, structured data), and your page earns a position in a ranked list. You optimize for that list.
AI assistants work differently. They don't serve a ranked list. They synthesize an answer, and to do that, they retrieve passages, not pages. The unit of analysis isn't domain authority. It's whether a specific paragraph, in a specific piece of content, is extractable, stands alone, and directly answers the question the user asked.
For better or worse, those are different optimization targets. Most of the investments companies make in traditional SEO, which is not a bad investment on its own terms, don't transfer to AI search visibility in any direct way. That gap is what most SEO audits never even measure.
What actually governs AI search visibility
The way I see it, there are three things.
The first is passage extractability. AI assistants pull passages, not pages. If your content buries the answer after three paragraphs of preamble, that passage won't surface. The answer needs to be near the top of a section, complete enough to stand alone, and specific enough that the model can quote it without losing meaning.
The second is topical specificity. Generalist content that's loosely "about" a topic doesn't get cited as often as content that definitively answers a specific question. There's research out of Princeton on what's called generative engine optimization, or GEO, that's useful here: specific statistics with named sources boosted citation rates by about 37%, and expert quotes with name and title boosted them by about 30%. Keyword stuffing, by contrast, reduced citation rates by about 10% in the same research. So the old SEO reflex of layering in more target phrases can actually hurt AI search visibility, not help it.
The third is structural alignment. Content that mirrors how an AI would query a topic, with sections that map to real sub-questions, performs better than content organized around a company's internal messaging hierarchy. Write for the question being asked, not the narrative you want to tell.
Why your current SEO metrics don't capture this
Domain authority doesn't tell you whether your content is passage-extractable. Keyword ranking doesn't tell you whether your paragraphs answer the questions AI assistants are synthesizing. Backlink count doesn't tell you whether your content structure maps to how models think about a topic.
This is where most of the frustration comes from. Buyers I talk to describe the same pattern: the SEO program is delivering on its stated metrics, and yet nothing is changing in how AI assistants represent the brand. What they're measuring, and what their agency is being held to, doesn't capture AI search visibility in any meaningful way.
The market is beginning to respond. A HubSpot-focused agency recently launched a dedicated AEO service as their sixth core practice, built specifically around getting clients cited by AI assistants. An AI search agency published a detailed breakdown on their blog of how they restructured their entire practice around relevance engineering for the same reason. When specialist agencies are reorganizing to serve this problem, it's worth asking whether your current program is even pointed at the right target.
The levers that actually move AI search visibility
What I recommend is starting with content structure before anything else.
First, audit your most important pages for answer density. The question to ask is: if I pulled a single paragraph from this page and dropped it into an AI response, would it make sense on its own? If no, it probably won't be cited. That's not a technical SEO fix. It's a writing fix.
Second, run a question mapping exercise. For the five or six topics your prospects are searching, map out the ten to fifteen sub-questions an AI assistant would likely synthesize across. You'll find gaps: content that exists but answers the wrong question, or answers structured in a way that buries the point.
Third, prioritize specificity over comprehensiveness. A shorter section that cites a real number, a named example, and a named source outperforms a longer section that gestures at the same idea. For AI search visibility, one concrete claim is worth three general ones. This runs against most editorial instincts. We're trained to be thorough. AI assistants favor being direct.
The honest limit here is measurement. I don't have a great answer for tracking AI search visibility with the same precision as keyword rankings. The tooling is improving, but citation rates are noisier and harder to attribute than a SERP position. My recommendation is to measure directionally, check specific queries manually, and build the content structure correctly rather than waiting for perfect attribution before starting.
Where your SEO investment still earns its keep
This isn't an argument for walking away from SEO. The way I frame it: traditional search and AI search are two overlapping audiences with different content needs. A page that ranks well in Google still drives traffic. The goal is to make sure your content serves both, and most programs right now are built for only one.
The companies seeing strong AI search visibility aren't doing it by accident. They have clear content structures, specific claims, and content written to answer a question rather than to target a keyword. Those are operational choices, not just editorial ones, and you can make them without rebuilding your entire program.
For more on how we think about content strategy inside a broader GTM program, our writing is collected here.