ChatGPT started recommending brands, so you figure it’s your turn to get ready. But the more you look at it, the more it just feels like SEO again. Push your pages up the rankings, only now for the AI instead of Google, and you’re done. Right?
I thought that too, honestly. The AI reads the web, so the pages that win at search should be the ones it picks up. It seemed obvious enough that I never really questioned it.
But some data showed up that pokes a hole in the whole “AI optimization is just SEO with extra steps” idea. And it’s a big pile of data, not a hunch.
”AI Optimization Is Just SEO, Right?”
The people who threw a lot of data at that exact question are OppAlerts, who published an LLM Ranking Factors analysis in March 2026.
The scale is the interesting part. They modeled 145 industries and 1,595 buyer personas, fired more than 100,000 prompts at AIs to see which brands got recommended, and then cross-referenced all that against a crawl of 1.1 billion-plus pages and 5 billion-plus Reddit posts.
Boiled down, they were asking one thing: the brands the AI keeps recommending, what do they have going on out in the wider world? Search visibility, backlinks, a Wikidata entry, Reddit chatter — they lined each signal up against how likely a brand was to get named, and looked at what actually moved together.
One caveat up front: OppAlerts sells AI-visibility tooling, so this is a study with a house to win. Worth keeping in mind. But let’s look at the numbers anyway.
What Beat Search Rank Was “The Facts Other People Hold On You”
So here’s what they found, and it’s not what I expected.
Start with the thing SEO people obsess over — search visibility. It did help, a little. But it explained only 5.8% of the variation in which brands got recommended. Backlinks landed in the same neighborhood, a few percent and no more. The weapon everyone’s been sharpening for years mostly swings through empty air here.
What did the heavy lifting, in some industries, was a signal most marketers never think about: Wikidata.
- A Wikidata entry: correlated 0.65 to 0.71 in hotels, ERP, and furniture — far ahead of everything else
- Search visibility: correlated 0.24 (explaining 5.8%)
- Backlink count: correlated 0.20 (explaining 4.2%)
Wikidata is basically a giant machine-readable database of plain facts about the world. When a company was founded, where it’s based, what it actually does — that kind of structured “who is this, really” record, all sitting in one place the machines can read.
So the AI leans less on where your own site ranks and more on how complete and clean the facts other people hold about your brand are. The prime real estate turned out to be nowhere near the map SEO taught us to read.
Why “Who You Are” Beats “Where You Rank”
Here’s the part I find genuinely interesting: this doesn’t look like an OppAlerts-only quirk.
Ahrefs ran a separate analysis of 75,000 brands, and the thing that tracked most closely with AI visibility wasn’t backlinks either — it was plain brand mentions across the web, roughly three times stronger. How much other people talk about you, how well your facts line up. The stuff sitting outside your own site kept coming out on top.
And when you think about the mechanics, it sort of clicks. A search engine picks which page to put on top. An AI picks which brand to say out loud. The first is a page-by-page contest; the second is a contest over whether the model recognizes you as a real, coherent thing at all. So it reaches for brands whose identity is clear and told the same way in a lot of places — the ones it feels safe naming.
Sit down at a new board with your old winning strategy and you have to relearn how to win (it’s a bit like someone quietly swapped the rules of the game while you were setting up the pieces). Stare only at the SEO rankings and you miss that the board even changed.
Two caveats, though, and they matter. This is correlation, not proof — nobody showed that adding yourself to Wikidata causes more recommendations. And the effect swings hard by industry; it won’t hit every brand the same. I think that’s worth holding onto honestly rather than waving away.
Conclusion: Tidy Up The Facts Others Hold, Then Measure Your AI Mentions
So here’s what I think this analysis actually hands you.
The biggest takeaway is that the battleground for getting recommended by AI has moved — off your own site’s ranking, onto how complete the brand information sitting outside your walls is. Climbing the search results still isn’t a waste. But on its own, it walks right past most of what the AI is actually reading.
If you want a concrete first move, start with your brand’s factual record. Is the founding date, the location, the line of business accurate and current in a fact database like Wikidata? Do third-party pages describe you the same way, over and over? None of it is flashy. But it’s the groundwork for the AI deciding you’re a brand it can name without flinching.
The second thing, and don’t skip it, is measuring. You can tidy up every fact and still not know whether it moved the needle, because a search-rank report won’t show you. Watch how often the AI actually brings your brand up, as a gauge that sits apart from your keyword rankings. Only when the move and the effect connect does any of this start to compound.
Sort out the facts the wider world holds on you, then measure how you show up in AI. The prime real estate of this new era is a step off the search-results map — and, I suspect, still oddly wide open.
Sources
- OppAlerts, “LLM Ranking Factors”, March 2026, oppalerts.com (145 industries, 1,595 buyer personas, 100,000-plus prompts analyzed; study by an AI-visibility vendor)
- Ahrefs, “Brand mentions vs. backlinks as predictors of AI visibility” (75,000 brands analyzed; brand mentions correlated roughly 3x more strongly than backlinks), ahrefs.com
