I asked Claude about a category I know well last week. Three brands came back. One was described as "the considered, founder-led option." Another was called "a polished mid-market alternative." The third got "an established but increasingly generic player."
All three brands paid for the citation in the sense that they've earned visibility. They're being recommended. The win, on paper, is even.
But if you're the buyer reading that answer, you're not picking randomly between three names. You're picking the considered, founder-led option. Of course you are.
The next layer of the problem
Most of what I've written across this series has been about visibility. Getting cited. Being the named answer. Showing up in the trusted-20 sources. That work matters and it isn't slowing down.
But it's the floor, not the ceiling.
Once visibility stops being scarce in your category, the differentiator moves up a level. The brands that get recommended will all be recommended. The question becomes how they get described when they're recommended.
That's the bit no agency is talking about yet.
How AI ends up with a description of you
When a model writes about your brand, it isn't reading your homepage in the moment. It's drawing on a compressed sense of your surface area. Wikipedia summaries, press write-ups, Reddit threads, your own marketing copy, podcast transcripts, the way founders describe themselves on LinkedIn, comparison posts written by people who don't work for you.
All of that gets blended into something the model can spit out in two clauses. "A boutique studio specialising in considered brand work." "A mid-tier provider competing on price." "A category leader." "A scrappy challenger."
Those phrases didn't come from nowhere. They came from patterns the model picked up across hundreds of mentions. If the language across your surface area is consistent, the description will be sharp and on-brand. If it's scattered, the model will land on whatever adjective shows up most often, and that's usually something you didn't choose.
I've watched brands get described as "well-known" when their founders would have killed for "best in class." I've watched brands described as "modern" when their actual positioning was "original." Those are not the same word, and one of them costs you the work.
Why the descriptions are sticky
Once a model picks up an adjective for you, it tends to repeat it.
Partly that's the math. Models reach for the highest-probability description. If three independent sources call you "boutique," you're boutique now, and the next answer reinforces it. The descriptor becomes self-citing.
Partly it's the way these systems propagate. Claude reads what ChatGPT generated. Perplexity cites a publication that quoted Gemini's summary. Once an adjective enters the bloodstream, it spreads laterally before anyone notices.
And partly it's that nobody on your team is looking. Most marketing dashboards are still measuring traffic and rankings. Almost none of them are measuring the words AI uses when it talks about you. The first time most founders find out their brand is described as "competent but uninspiring" is when a buyer mentions it on a call.
By that point you're not editing copy. You're trying to redirect a current.
What you can actually influence
I'm not going to pretend you can fully control how a model describes your brand. You can't. But you can stack the deck.
The first move is positioning vocabulary. A brand needs five or six words it owns and repeats. Not a tagline. Words. The adjectives, the verbs, the nouns it wants associated with itself across every surface. If your founder says one thing on a podcast, your website says another, and your case studies say a third, the model averages them out and gives you something none of you would have chosen. If they all say the same thing, the model has nothing to do but agree.
The second move is controlled mentions in trusted sources. Not press for press's sake. Coverage where the framing is right. A founder interview that lands the language you want. A case study in a publication an AI platform reads. A Reddit thread where the wording you'd choose is already in the top comment because you put it there. The trusted-20 sources do most of the citation work, which means they do most of the describing too.
The third move is narrative anchoring. A brand needs an origin story that's specific enough that AI can't paraphrase it into something generic. "Started by two ex-Stripe engineers who kept hitting the same problem on their own team" is anchorable. "Founded with a passion for design" is wallpaper. The specific story sticks. The generic one gets sanded down to the same description every other brand in your category is getting.
None of that is fast. All of it compounds.
Where I think this is going
I've spent the last year of studio work treating AI visibility as the problem. Getting cited. Being recommended. Earning the answer slot.
I think the next year is going to be about the layer above that. Not "are we cited," but "what is being said about us when we are." Not "are we the answer," but "are we the most flattering version of the answer."
That sounds soft until you watch a buyer pick the brand that got the better adjective. Then it stops being soft.
There's a discipline forming around this and it doesn't have a name yet. Part brand strategy, part editorial, part the kind of work copywriters used to do for press releases before press releases stopped mattering. It needs people who can think across surfaces, who understand how language compounds in machine-readable systems, and who care about the specific words a model reaches for when it summarises an entire company in nine words.
I'm building the studio around it. The work I want to be doing in twelve months is less "help us get cited" and more "help us get cited the way we'd describe ourselves." The first is a visibility problem. The second is a brand problem. They're not the same job, and the second one is harder.
Most agencies haven't even named the first problem yet. Almost none of them have spotted the second.
That's the gap I think Level44 fits into. Designing brands to be seen, yes. But designing brands to be seen as something specific. The adjective is the work now.
It's where I'm pointing the studio next.