We spent 2026 running AI visibility programs for a range of brands, working to get them named, cited, and recommended inside the answers that ChatGPT, Gemini, Google’s AI Overviews, and others now hand people before they ever click. A year in, the clearest lesson is also the least convenient one: this is foundational work, not a campaign you switch on and off. Here are the four things that have stuck with us.
1. The sources are concentrated, and the bar to qualify is high
AI answers don’t pull evenly from the whole web. A small set of high-authority sources currently carries an outsized share of what the models say about a brand and its category: Google Business Profile, Wikidata, Wikipedia, and Reddit, where community rules are strict and gaming them is hard by design. That concentration is the opportunity and the obstacle. Being present and accurate in those places moves the needle more than almost anything else, but they are built to be difficult to earn.
Wikipedia is the clearest example. Qualifying for a page takes real notability and well-sourced, independent coverage, and plenty of businesses simply won’t be eligible. The standard is high on purpose. The upside is that once you clear a bar like that, it’s a bar your competitors have to clear too.
2. It’s foundational work, not a quick win
Generative engine optimization (GEO) is genuinely multi-dimensional, and it’s proving out only when you work every angle together: structured data, Google, Wikipedia and Wikidata, PR, and the reputation signals underneath all of them. There’s no single lever. The foundational pieces are more involved than they are in traditional SEO, with more hoops to clear and more dependencies between them, and none of it pays off on a short timeline.
That reframes how to budget and how to set expectations. A GEO program isn’t a sprint toward a ranking. It’s the slow construction of an evidence base the models trust. Teams that treat it like a quarterly campaign tend to quit right before the work starts to compound.
3. The ground keeps moving
Search behavior is changing fast, but search isn’t going away. It’s expanding into new shapes and new surfaces, and the platforms are shifting in real time. ChatGPT’s business hub is the latest addition, and there will be more by the time you read this. The algorithms aren’t settled: Reddit’s citation share was high earlier this year and has already come down. What earns a mention this quarter may not next quarter.
The practical consequence is that you’re optimizing for several platforms at once, each with its own behavior, and you’re doing it against a target that moves. A program built around one platform’s rules is fragile. The durable approach spreads the foundational work across sources so that no single algorithm change knocks you out of the answer.
4. Measuring it is its own discipline
Reading results is harder than it looks, and the nuances matter. There’s a real difference between a mention and a citation. You can be named in an answer with no link back to your site, and consumers are getting more aware of the links that do appear, which changes what each one is worth. Counting “mentions” and “citations” as the same thing hides what’s actually happening.
Two more measurement realities shaped how we work. Training data matters less than it used to, because the major platforms increasingly default to live web search, which means fresh, crawlable signals carry more weight than they once did. And personalization heavily skews what any individual sees, so checking visibility from your own logged-in account tells you very little. A clean, controlled testing sandbox turned out to be essential just to see the truth of how a brand shows up.
The strategic read
Put the four together and the shape of the work is clear. AI visibility isn’t a campaign you run, it’s a foundation you build, across a concentrated set of hard-to-earn sources, on platforms that keep changing, measured with more care than a dashboard of clicks requires. That’s more demanding than it sounds in a pitch deck. It’s also why the advantage is real: the brands willing to do the foundational work now, earning the authoritative mentions and getting the structured data and reputation signals right, will be the ones the models reach for when asked. The ones waiting for it to settle will find the answer already written.
Want to go deeper?
Explore Luckie’s AI visibility program or download our AI readiness white paper.
Article Contributor: Mark Unrein, Chief AI Officer



