Are You Optimizing for the AI Your Buyers Actually Use?

For months, I refused to download the Claude app. Not because I wasn’t using Claude - I was using it constantly - but because Chrome on my iPhone felt like enough. The result: my mobile AI usage looked nothing like my desktop usage, for no reason deeper than friction. One person, two devices, two completely different AI footprints.

Now multiply that by every buyer in your market. That’s the problem with how most GEO plans get built. They start from a leaderboard - ChatGPT is biggest, so optimize for ChatGPT - when they should start from the audience: who your buyers are, what device they’re holding, what task they’re trying to get done, and whether they’ve ever paid for an AI tool in their lives. Skip those four, and you’re optimizing for a market average that describes nobody.

AI usage is a behavior, not a personality trait.

Which AI someone uses is not a fixed trait you can look up once and file away. It changes with the job. A physician who lives in Perplexity at work because she needs cited, defensible answers may still plan the family’s Alaska cruise in ChatGPT, and check a flight detail in a Google AI Overview she never asked for. Same person, three AIs, one afternoon.

So the useful question is never “which AI does this person like?” It’s “which AI, for this job?” For travel brands, the job is trip planning - and that narrower question is one you can actually answer.

Half the AI your buyers use, they never chose.

There are two layers of AI exposure, and they behave nothing alike. The deliberate layer is the one we all picture: someone opens ChatGPT, Perplexity, Claude, or the Gemini app on purpose. The ambient layer is bigger and much quieter: Google’s AI Overviews sitting on top of searches people were already running (about 2.5 billion people meet those every month), Copilot preloaded on Windows machines, Meta AI living inside WhatsApp and Instagram, Siri riding along on every iPhone.

Nobody downloads the ambient layer. It arrives with the device. And most market share charts collapse both layers into one ranking, which is why the numbers disagree so wildly depending on who’s measuring - web traffic, app downloads, surveys. Separate exposure from choice, and the picture finally becomes useful.

The demographics will humble you.

Pew's latest data shows 61% of US adults under 30 have used ChatGPT. Among adults 65 and older? Nineteen percent. If your bookers skew 60+ - and for many of the tour operators we work with, they do - a ChatGPT-first plan is optimized for someone else's customer.

Older travelers aren’t avoiding AI, though. AARP’s latest Tech Trends survey shows generative AI use among adults 50+ nearly doubled in a year, from 18% to 30%. It’s just overwhelmingly ambient exposure: the AI answer sitting above the search results they were already reading.

Device assumptions deserve the same scrutiny. The common mental model says older Americans are iPhone people. The data says US adults 60+ are roughly evenly split and actually lean Android, which changes the default assistant they meet every single day. That one correction rewired how we score personas.

Free users live in the defaults. Paid users made a choice.

Roughly 3% of AI users worldwide pay for the tools they use, per Menlo Ventures. The other 97% live wherever the defaults put them - straight into the ambient layer. Travel sharpens the point: Phocuswright's research shows the travelers using AI to plan trips skew affluent, and about 40% pay for a monthly AI subscription, roughly double the general adult rate. If your audience is high-value travelers, the deliberate destinations matter more than raw market share implies.

What we don’t do is predict model tiers within a platform. Free model or paid changes constantly and is invisible from the outside. Willingness to pay is scoreable. Tier is guesswork dressed up as insight.

What we actually do about it.

Score, don’t assume. Every persona gets scored on age, market, device ecosystem, device type, discovery habits, task, and willingness to pay - and the output is a ranked list per persona, split into ambient and deliberate layers, because winning one does not win the other. And Google never gets merged into one line item: AI Overviews, AI Mode, and the Gemini app cite different sources for the same prompt. Three surfaces, three targets.

Here’s the thing: after all that scoring, ChatGPT might still land at the top of your list. Great. Now it’s a finding instead of a guess - and there’s a real chance the rest of the list surprises you. The leaderboards will keep reshuffling every quarter. Your audience is the stable thing. Build on them.

Sources: Pew Research Center (2026), AARP 2026 Tech Trends and Adults 50-Plus, Google I/O 2026, Menlo Ventures 2025 State of Consumer AI, Phocuswright The AI Surge (2026)