Somewhere right now, a person you’ve never met is describing your product in a thread you’ll never read. They might be recommending you. They might be warning people off. Either way, they’re doing something your marketing team can’t: they’re telling an AI model what your brand is actually like.
Here’s the uncomfortable mechanic behind it. When someone asks an AI assistant “what’s the best tool for X,” the model isn’t just consulting your homepage. It’s drawing on everything it has absorbed and retrieved — and a disproportionate amount of that material is third-party conversation. Reddit threads. Community forums. Review sites. Comparison posts written by people with no stake in your success. Your own website is one voice in that chorus, and it’s the voice the model trusts least, because the model knows exactly what your website is: an advertisement.
Think about how you’d answer the same question for a friend. If they asked which project management tool to use, you wouldn’t recite a vendor’s landing page. You’d say “I’ve heard good things about this one” or “everyone I know who tried that one gave up after a month.” You’d synthesize social proof. Language models do a statistical version of the same thing. First-party claims get discounted; third-party consensus gets weight.
This flips the traditional marketing hierarchy on its head. For two decades, the center of gravity was your owned property — your site, your blog, your funnel. Everything else was “earned media,” a nice-to-have you couldn’t control and therefore mostly ignored beyond a PR retainer. In AI search, the ratio inverts. The earned layer isn’t the garnish anymore. It’s the meal.
You can see this play out in how differently brands surface in AI answers. Companies with active, opinionated user communities — even small ones — get described with texture. The model can say what they’re good at, where they fall short, who they’re for. Companies with pristine websites and no community footprint get described generically, if at all. A thousand honest forum comments beat a hundred polished landing pages, because the comments teach the model things your copy never would: the edge cases, the comparisons, the “yes but,” the specific job someone hired your product to do.
The instinctive response from marketers is predictable, and wrong: go seed those conversations. Pay for placements dressed up as recommendations. Astroturf the threads. Set aside the ethics for a moment — it simply doesn’t work at the scale that matters. Models triangulate across thousands of sources and years of history. A burst of suspiciously glowing mentions in a handful of threads doesn’t move the consensus; the consensus is the accumulated sediment of every real interaction people have had with your brand. You can’t fake sediment. Communities are also ruthless at sniffing out shills, and a thread where you got caught astroturfing is itself durable training data — the worst kind.
So what does work? Three things, none of them glamorous.
First, be present where the conversations happen — as yourself. Founders and product people who answer questions in their niche communities, honestly and without a pitch, generate exactly the kind of material models learn from. Not because each comment is seen by millions, but because it’s specific, attributed, and sits in the middle of a relevant discussion. When your CTO explains in a forum thread precisely how your product handles a tricky use case, that explanation can outlive every blog post you published that quarter.
Second, give people something worth mentioning. Unprompted third-party mentions are downstream of product experiences that are notable enough to talk about. A feature that saves someone an afternoon gets a forum post. A support interaction that surprises someone gets a screenshot. This sounds like “just be good,” and partly it is — but the actionable version is: build and communicate the one or two things about you that are distinctly worth repeating. Vague betterness generates no conversation. Specific, describable difference does.
Third, listen to the record that already exists. Most companies have never actually read what the internet says about them in aggregate. They know their reviews average four stars and stop there. But the substance of those conversations — the recurring complaint, the comparison that keeps coming up, the feature people praise that you never advertise — is a preview of what AI models will say about you when asked. If the threads keep saying you’re powerful but painful to set up, that is your brand in AI search, no matter what your homepage claims. You can either fix the setup experience or keep arguing with the sediment.
There’s a deeper shift underneath all this. Marketing used to be the discipline of controlling the message. The channels were finite, the airtime was purchasable, and the loudest consistent voice won. What’s emerging now is closer to reputation in a small town: everyone talks, memory is long, and what’s said about you when you’re not in the room matters more than anything you say when you are. AI didn’t create that dynamic — it industrialized it. The models are the town’s collective memory, queryable in an instant, by every prospective customer, forever.
Which means the job description changes. Less broadcasting, more participating. Less campaign, more conduct. The brands that thrive in AI search will be the ones that figured out how to be worth talking about, showed up honestly in the talking, and treated the internet’s memory of them as an asset to be earned rather than a message to be managed.
Your website tells AI who you say you are. Strangers tell AI who you actually are. Give the strangers something good to say.