A while back I wrote about why most brands are invisible to AI. The short version: if you're not on the sources these models trust, you don't exist when someone asks the model to recommend something in your category.
That post talked about the four big platforms as a group. Useful for setting the scene. Less useful if you actually want to do something about it.
Because the four don't behave the same way. Not even close.
I've spent the last few months running brands through each of them, watching what gets cited and what doesn't, comparing the answers. The pattern that emerges is that each platform has its own diet. Optimising for one isn't optimising for the others. If you treat AI visibility as a single problem you'll get average results everywhere and great results nowhere.
So here's what I'd think about for each one.
ChatGPT
ChatGPT is picky. When you ask it to recommend brands in a category, it tends to name fewer of them per response than the others do. That sounds like a problem, but it's the opposite if you're one of the names. Less competition in the answer means more attention on your brand.
What does ChatGPT draw on? Reddit, heavily. Around 9.4% of its citations come from Reddit threads. Editorial publishers come up a lot too. Wikipedia plays a role. Its own training data shapes the baseline view of who you are, which means your historic press matters as much as last week's press.
If you want ChatGPT to cite you, the work is in two places. Reddit and editorial.
On Reddit, not the way most marketers approach it. ChatGPT seems to weight threads where there's actual discussion, not just a posted link. So someone has to be talking about you, not just mentioning you. The brands that show up in ChatGPT answers tend to be the ones that come up organically when a real person asks "has anyone tried X" in a relevant subreddit.
On editorial, not blog mentions. Real publications. The kind of place that has an editorial line and a publishing standard. If you're in the design world that's Creative Boom, It's Nice That, Fast Company. In SaaS that's TechCrunch, The Information, a16z's content engine. Get yourself written about there, and ChatGPT notices over time.
Claude
Claude looks similar to ChatGPT on the surface. Concentrated citations. Strong editorial bias. But Reddit matters even more here, around 9.8% of citations. And the bar for what counts as a useful Reddit signal feels higher.
Claude seems to read Reddit for substance rather than presence. A thread where someone explains why they switched from one tool to another, with reasoning, gets weight. A thread where someone just drops a link and walks off does not.
If you want to be cited by Claude, your customers and community have to talk about you with some depth. That's not something you can fake with a marketing team posting on burner accounts. Claude's model seems to read context well enough to spot promotional patterns and discount them. I've watched brands try the burner-account trick. It does not work. The signal that does work is real users explaining why your thing solved their problem, in their own words, in subreddits where that kind of explanation is welcome.
Claude also weights authoritative writing more than ChatGPT does. Long-form blog posts on credible domains. Substacks with a real audience. Industry analysis with named authors. If you have an expert on staff who writes well, get them publishing under their own name on a domain that has authority. That signal compounds over time.
Gemini
Gemini is the broad-coverage platform. For about 79% of brands, Gemini gives the highest visibility of any of the four. It cites more brands per response. The bar to get into a Gemini answer is lower.
That sounds like good news, and it is, but the leverage points are specific.
Gemini is wired into Google's grounding signals. That means structured data on your site, Schema.org markup, Knowledge Graph alignment, Wikipedia. If Google understands what your brand is and where it sits, Gemini understands too. If Google doesn't, Gemini won't either.
The biggest single move for Gemini visibility is Wikipedia, but Wikipedia is also the hardest to earn. You can't write your own entry. You earn one by being notable enough that someone else does, which means accumulated press coverage, third-party references, and time. The fastest realistic move for most brands is Knowledge Graph alignment, getting Google to recognise you as an entity with consistent attributes across the web. Sameness of name, address, description, and category across every place you appear online.
YouTube is the other Gemini lever. Google owns it. Gemini reads it. If you have product walkthroughs, founder talks, case study videos, anything with substance, hosted on YouTube with proper titles and descriptions, you create an extra surface for Gemini to draw on. YouTube alone makes up a significant share of citations across the trusted-domain set.
Perplexity
Perplexity is the editorial-first platform. It draws heavily on press and authoritative publications. Reddit is in the mix but it's not the dominant signal it is for Claude or ChatGPT.
If your brand has been written about in trusted publications in the last year or two, Perplexity is the platform most likely to reflect that quickly. If it hasn't, Perplexity is the platform most likely to ignore you.
The work for Perplexity is the work of being newsworthy. Real PR, not pay-for-play press releases. A point of view someone wants to write about. A product launch with substance. Research nobody else has done. The brands that get cited by Perplexity tend to be the ones a journalist would actually recommend to their editor.
This is also where Perplexity gets interesting for smaller brands. You don't need national press to score here. You need credible press. A well-written piece in a respected niche publication often counts for more than a buried mention in a tier-one outlet.
What ties it all together
Different platforms, different diets. Different tactics. The mistake most brands make is to optimise for whichever one feels closest to existing SEO habits, usually Gemini, and ignore the rest. That gets you a fraction of the visibility available.
But there's a pattern underneath it. Across all four platforms, a small set of trusted domains do most of the work. Wikipedia. Reddit. YouTube. Quora. Vertex AI's grounding sources. These domains account for around half of all citations, full stop. About 49%, in the data I've been looking at.
That's the leverage point. If you build presence on those domains, the work compounds across all four platforms. Wikipedia helps Gemini and ChatGPT and Claude and Perplexity, in different ways, but all of them. Reddit helps Claude and ChatGPT a lot, Gemini somewhat, Perplexity less. YouTube helps Gemini directly and gets quoted by the others. Quora is a steady drip across all of them.
You don't have to win every domain. You have to be visible on enough of them, with consistent information, to be the answer when the model is asked. The brands that figure this out are not the ones running the loudest paid campaigns. They're the ones whose presence on Wikipedia, Reddit, YouTube and a few editorial sources adds up to a clear, citable signal.
I think a lot of marketers will spend the next year trying to crack each platform individually and getting frustrated. The ones who pull ahead will work the trusted-twenty instead. Same effort, four times the reach.
That's the move.