AI is a New Audience For Brands

Over the past year, buyers have been turning to AI instead of Google. The result is that organic search traffic is declining across the board, roughly 30% on average. At the same time, a new line item is showing up in analytics: `utm_source=chatgpt`. Referrals from AI models. And this traffic converts at 3-6x the rate of organic search.

If you're a marketer looking at those numbers, the natural reaction is to see AI referrals as an emerging channel. The data shows that traffic is migrating from one source to another, and the imperative is to show up in the new one. That framing is reasonable, and for the near term it's even useful.

But that perspective only captures a tiny slice of AI's total impact.

Direct referral traffic from AI represents the cases where an AI model sent a buyer directly to your website. But it doesn't represent the case where a lead clicked your Google ad, and then asked ChatGPT "is this product actually good for my use case?" Or the one who got a cold email from your SDR and asked Claude's opinion.

In these increasingly common cases, AI shaped the outcome, but it never shows up in your attribution. The buyer arrived through paid search or email or direct, and your analytics have no idea that an AI model weighed in along the way. Or, worse, ChatGPT turned your perfectly qualified lead to one of your competitors instead, and that lead never made it to your site. Direct referrals are the exhaust trail of something much larger: AI is influencing decisions across all of your channels, because prospects are consulting it regardless of how they found you.

This is why the channel framing is limiting. A channel is infrastructure: a pipe that delivers your audience to you. Organic search, paid social, email — these are mechanisms for reaching buyers with your messaging. If all you see when you look at AI is a new version of that, you're measuring the smallest slice of its effect on your business.

Companies need to see AI as a new audience. reading everything on the internet and forming opinions about you and your competitors; they are advising your buyers on what to purchase; and they are increasingly making purchasing decisions _on their own_. That's audience behavior. AI is an audience you need to understand, market to, and win over.

Right now, while AI mostly functions as a research tool for human decision-makers, the channel framing and the audience framing look nearly identical. The difference will become obvious as AI models move further along the arc from research assistant to autonomous buyer, which is already happening faster than most people realize.

The arc

There's a pretty clear progression to how people are using AI to make purchasing decisions, and most of the conversation is stuck on step one.

Research assistant. This is where most people are today. You ask ChatGPT to compare options, summarize reviews, outline tradeoffs. The AI does the legwork and you make the decision. Faster and more conversational than Googling, but fundamentally the same relationship: you're the decision maker, it's the tool.

Advisor. You tell the AI about your specific situation, your constraints, your preferences, and it comes back with a recommendation. It pushes back when you make a bad call. I [wrote about this recently](https://www.unusual.ai/blog/ai-search-is-already-dying): I had a 20-message conversation with ChatGPT about running shoes where it overruled my budget preference because it remembered a knee issue from a previous conversation. We reached an agreement before I bought anything. That's advisor behavior.

Decision maker. You tell your AI assistant what you need and it handles the rest: research, evaluation, selection. You review its recommendation and approve the transaction. The AI is making the call. You're ratifying it.

Autonomous purchaser. You don't tell the AI anything. It already knows what you need based on everything it knows about you, your patterns, your preferences, your schedule. It buys the thing. You might not even notice.

Each step requires more trust. And trust is the only thing between where we are and where we're going.

Already happening

Step four sounds far off until you look at developer tools, where we're basically already there.

Engineers using Claude Code and other coding agents are trusting AI to make their tech stack decisions. The agent evaluates options, picks a service, and tells the engineer to come back with an API key. The engineer didn't research alternatives. Didn't read docs. The model understood their codebase and requirements better than they had time to, so they went with its judgment.

This pattern is spreading fastest in categories where the product is complex, the decision is more about meeting requirements than expressing taste, and the cost of researching every option yourself is high. Enterprise software, financial products, infrastructure, insurance. B2B purchasing broadly.

The common thread: the more a decision looks like "does this meet my needs" rather than "does this reflect who I am," the faster AI takes it over. And that describes a very large share of the purchasing decisions that happen in business every day.

The trust curve

Right now, AI models have this odd trust profile. They can solve unsolved math problems, but they sometimes get confused about whether you should walk or drive to the car wash. Seeing those moments where AI is obviously _dumb_ makes people cautious.

But models improve on a curve that's steeper than anything we've seen in technology. Every few months, the dumb mistakes get rarer, the advice gets more tailored, the reasoning gets more transparent. And each improvement quietly expands the set of decisions people are comfortable delegating.

Two years ago, nobody would have trusted an AI to pick their cloud infrastructure provider. Today, engineers do it routinely without thinking twice. The set of decisions we delegate two years from now will be much larger than most people currently expect, and the "channel" framing will have nothing useful to say about it.

Different ways of thinking = different outcomes

With a channel, you think about visibility and funnel efficiency: how do I show up, and how do I convert the traffic it delivers? With an audience, you think about belief: what does this audience currently think about me, is that accurate, and what would it take to change their mind?

Channel thinking tells you to create more content so you show up more often. Audience thinking tells you to figure out what AI already believes about you, whether that's working for or against you, and invest in the positioning and evidence that shifts their opinion. AI models have already read everything you and your competitors have ever published. They need better reasons to believe you're the right choice rather than more content to make them notice you.

What this means

If what AI models believe about you is influencing buyers across every channel you run, then your AI perception is effectively an invisible multiplier on everything else you do. Your paid campaigns, your outbound, your content, your partnerships — all of it performs better or worse depending on what happens when a buyer asks an AI model about you along the way. Today, most companies have zero visibility into that.

AI models are already the most well-informed audience your brand has ever encountered. They've synthesized everything on the internet about you and your competitors, formed opinions, and they update those opinions continuously. Visibility is the easy part: this audience already knows you exist. The question is what they believe about you, and whether that's helping or hurting every time a buyer asks.