For most of my career the audience was a person. You wrote the ad or the product page for someone who would read it, feel something, and maybe buy. That person is still there. But this week's marketing news kept circling a second reader that now sits between the brand and the customer, and it isn't human.
Scrunch published a study of millions of AI citation events and found that when a sponsored YouTube video gets cited in an AI answer, the sponsor shows up in 40% of those answers, against 32.3% when the video isn't cited. On unbranded questions the lift reached about 1.35 times. It's one company's data, so I'd read the exact figures as directional rather than a benchmark. The useful part is what they point to: write the video description for the model, with the sponsor named in plain language, and then track which prompts actually cite you, because the same study found citation rates falling 43% within a month.
Time magazine is going further. Marketing Brew reported it's testing ads built specifically for AI agents, feeding them brand-verified facts before the agent reads the open web, partly to correct things AI systems get wrong about a brand. And Amazon, which spent months keeping shopping agents out of its own store, quietly became ChatGPT's largest retail advertiser between April and August, and is now letting other advertisers run inside ChatGPT conversations too.
So the copy and the description text are increasingly written for a machine that reads on the customer's behalf. This is the part of AI that has little to do with generating content and a lot to do with distribution. Benedict Evans made a related point this week. The real change isn't faster versions of the work we already do, it's noticing which work should exist at all.
I don't think the human audience goes away, and I don't think optimizing for the machine reader is beneath anyone. What I watch for is the quiet reallocation. If a team's hours drift toward feeding facts to agents and nudging citation rates, something has to give, and it's usually the slower work of deciding what's actually true and worth saying. The machine-facing work is easy to measure and the human-facing work often isn't, and measurable work tends to win the calendar.
The judgment call is which of your claims deserve to be optimized for a machine at all. A brand-verified fact is only worth feeding to an agent if it was worth standing behind in the first place. Optimize a weak claim and you've just taught the machine to repeat it faster.
I'm curious how much of this has reached actual marketing teams, and how much is still sitting in vendor decks. It's one of the things I want to ask the Vancouver practitioners I'll be sitting down with this fall, once the manuscript is locked. My guess is the real answer is messier, and more interesting, than the studies make it sound.
Sources
- Scrunch, YouTube influencers have an AI audience: scrunch.com
- Marketing Brew, on Time magazine's agent-targeted ads: marketingbrew.com
- CNBC, Amazon opens ChatGPT ads to its advertisers: cnbc.com
- Benedict Evans, AI, tools and transformation: ben-evans.com