AI Search Works, But Getting Buy-In Is the Real Challenge

At SMX Advanced in Boston earlier this month, two presentations on AI search and its impact on marketers revealed a critical disconnect: one speaker focused on what to optimize, while the other explained why most teams never actually ship those optimizations. The real insight, however, came from the gap between them. Following the conference, I spoke with both Crystal Carter, Head of AI Search and SEO Communications at Wix, and Jen Cornwell, Senior Director of AI SEO at Tinuiti, to ensure accuracy in representing their talks.

Carter’s framework centers on a key distinction between “memory” and “personalization” in AI assistants. Memory is what an AI infers passively from your tone, patterns, and complaints, while personalization is what you actively declare through profile settings and connected apps. Personalization carries enough weight to shape what an agent actually does, not just how it sounds. You cannot SEO your way into someone’s inferred memory the way you’d tune a meta description, but you can engineer the signals that shape both.

Her strongest evidence came from an iPullRank experiment using three accounts with identical prompts but different levels of connected personal data. The results produced visibly different AI Mode answers, including one that addressed a hypothetical child by name in a streaming recommendation. That controlled comparison should worry anyone still treating AI search results as a single, generic output.

Carter then moved into tactics, starting with denominal nouns like “actor” instead of “the person who acted,” because semantic models cluster identity-related queries that way. She also highlighted a critical gap: the average Google query runs three to four words, while the average ChatGPT opening prompt runs roughly 103 words. That gap argues for FAQ-style, narrowly specific content over broad landing pages. Users typing into an AI assistant are already further down the funnel than a search box ever made them.

Cornwell’s session contained almost no new SEO data, and that was the point. She named a different problem entirely: most search teams aren’t short on insight; they’re short on an organization willing to act on the insight it already has. That’s not a search problem. It’s a change management problem, and she offered two borrowed frameworks to solve it: Kotter’s eight-step change model and Everett Rogers’ diffusion of innovation curve.

She made the concept stick by re-skinning Kotter’s 2005 fable about a penguin colony on a melting iceberg with AI Overviews as the melting ice. She assigned five cast roles (Sponsor, Trust, Catalyst, Analyst, Skeptic) that every attendee was implicitly asked to map onto their own team. By the closing slide, you weren’t taking notes on an eight-step process anymore; you were running a casting call on your own org chart.

The research anchor worth keeping is Rogers’ tipping point math. Innovators make up 2.5% of any population, early adopters another 13.5%, and once a change reaches that combined 16%, adoption tends to become self-sustaining. Applied internally, that reframes “convince the whole company” into “find the findable minority,” a far less paralyzing target for an SEO arguing for budget in a room full of skeptics.

Here is the dissonance, and why it matters more than either talk alone. Carter’s framework assumes the bottleneck is knowing what to build: the right structured data, niche content, and MCP server configuration. Cornwell’s framework assumes you already know what to build, and the bottleneck is getting five other departments to let you ship it. Put them together, and they reveal why so many AI search initiatives stall: most teams only have tools for one half of the problem.

If your AI search strategy has a technical roadmap but no internal coalition, Carter’s tactics will sit in a deck nobody approves. If you have executive buy-in but no specific play to run, Cornwell’s framework will produce a motivated team with nothing concrete to do on Monday morning.

Three moves worth taking from both rooms:

Pick one niche content gap, not a full audit. Use Carter’s owned-channel framing, but resist building the comprehensive AI-visibility document nobody reads. Ship one piece of FAQ-style content that matches how people actually prompt AI assistants, then use it as your proof of concept internally. Find your 16% before you pitch the whole room. Identify the one or two people already directionally sold on AI search investment and build your brief with them first. You’re not trying to convince your most skeptical stakeholder on day one. Cast your own five roles before the next proposal. Name who on your team is the Sponsor, the Skeptic, the Catalyst. Walking into a budget conversation already knowing where resistance will come from is worth more than another slide of AI Mode screenshots.

Put Carter and Cornwell next to each other, and the lesson is hard to miss. Most teams treat AI search as two separate jobs: the people who figure out what to build, and the people who fight to get it shipped. Carter’s room assumed the hard part was knowing what to optimize for. Cornwell’s assumed you already knew, and the real work was getting everyone else to act on it. Both are right, which is exactly the problem.

A technical roadmap with no internal coalition stalls in a deck nobody approves. A motivated team with no specific play has nothing to do on Monday. The strategies that actually move are the ones run as a single job, not two. Optimization was never the hard part. Getting your organization to act on it is.