If you’re a marketer, I’m wondering whether you’re more excited, scared or frustrated by AI.
No matter the case, it’s coming, but maybe not in the way you expect.
Emily Kramer, who ran marketing at Asana and Carta and writes the MKT1 newsletter, had Claude Code pull and tag more than 1,100 open marketing job posts this fall. Only 47% ask for AI skills. 15% ask marketers to build anything with AI. 7% mention agents. Entry-level posts ask the least, at 25%. (MKT1)
The jobs are changing anyway. Two investors wrote in Fortune last week that they were about to hire several entry-level marketers at around $60,000 each, and hired one person for a lot more instead. Jason Lemkin runs SaaStr's sales with about 1.2 people and 20 AI agents.
I’m sure marketers would tell you all the issues with handing over more marketing work to agents—but at some point, they get better and better. What I’d argue isn’t that they won’t be able to do things marketers do now—it’s that we have blind spots about the full extent of a good marketing team’s work.
Rory Sutherland of Ogilvy calls this the Doorman Fallacy. You replace the doorman with an automatic door, book the savings, and nobody gets blamed for everything else the doorman was doing—security, checking in on the wellness of residents, talking to other doormen about neighborhood goings on.
A lot of AI in marketing is being sold as cost cutting, but what’s being cut that we might not be thinking about? And what does it mean marketers should be doubling down on to lean into where they can create the most non-commodity value:
1. Find out what your customers want
Human insight is a function of good data—without it, your impact is limited.
AI can empower humans with better data at scale.
A few examples:
Site search trained people to type in weird Google haikus, like "toddler plant based safe," which tells you almost nothing about what somebody is trying to do. A chat box gets the whole question, with the context around it. Dewey Labs builds chat for publishers like Emily Oster's ParentData and Gretchen Rubin, and reports back the most-asked questions, the ones the content can't answer, and how both shift by season. That's a list of what your audience wants that you haven't made yet, in their words. (Dewey)
AI can also do the interviewing. A study published this fall ran 139 AI-moderated customer interviews against 24 human-moderated ones and a batch of static surveys. On the same budget, the AI interviews surfaced more than twice as many distinct customer needs as the human ones, at about $19 per need versus $36, and covered more ground in most parts of the conversation. The researchers then used the interviews to build AI "digital twins" of those customers, and the twins predicted real responses better than personas built on demographics alone. (arXiv)
2. Build your own tools
The one marketer in that Fortune piece designs all the collateral and built software that benchmarks customers against their industry peers. The same company got a $60,000 agency quote for a new website. Three people from the business team who'd never built one did it in three days for a few hundred dollars, and along the way turned up customer testimonials and awards nobody knew they had. (Fortune)
Elaine Zelby, co-founder of Tofu, built an agent that checks HubSpot every week for newly closed deals and pulls the call notes on why each customer bought. It finds 10 lookalike companies and three to five contacts at each through Clay, drafts a four-email sequence and LinkedIn messages, and drops them in Slack for review. She says to build three skills first: your ICP, your personas, and your messaging. (MKT1)
Kamil Rextin of 42 Agency built one that pulls a company's competitors from G2 and TrustRadius, scrapes their ads from the free LinkedIn Ad Library, and produces a report on their messaging and ad volume in about five minutes, on a schedule. (MKT1)
3. Know what good looks like
The Nuremberg Institute for Market Decisions, GfK's parent, tested AI "synthetic respondents" against real consumers. They matched real choices about 79% of the time, but were consistently more positive about big brands and gave less varied answers. If you're the challenger, that bias works against you. (NIM) Mark Ritson, meanwhile, is bullish on synthetic research. (Incubeta) Someone on your team has to be able to tell the difference.
Aditya Vempaty, VP of Marketing at MoEngage, built a "humanizer" that scores drafts on how AI-written they sound, flags the giveaways, and rewrites them in his voice, using samples of his own writing. (MKT1)
4. Get your data right
Twilio CMO Chris Koehler wrote in Fortune last month that account research and personalized content for a campaign used to take his team two weeks and now takes under a day. That took them from three account-based campaigns a quarter to as many as 15. He said the model wasn't what made it work. Cleaning up their data and setting brand standards first was. When they let AI localize a French campaign without that, it put a beret on everyone. (Fortune)
5. Market to the machines, too
A boutique hotel in a Harvard Business Review piece this summer started checking how AI tools described it, then rewrote its public content so the descriptions were accurate and set it apart. (HBR)
Visa CMO Frank Cooper says the next customer is an AI agent shopping for someone. It won't be charmed by your copy. If your pricing is unclear or your inventory data is inconsistent, it moves on to a competitor whose data it can use. (Fortune)
6. Get good at what AI can't fake
Kramer says events are working because inbound is down, outbound is saturated, and people want to see each other, and that the best ones are small and built around specific target accounts. (MKT1) She's also written about companies hosting their own dinners instead of renting trade show booths. (MKT1) The Content Marketing Institute found events were B2B marketers' second-highest planned budget increase for 2026. (The State of Brand)
Ramp went all the way. In September it staged a one-night Broadway musical at the Broadhurst Theatre about its bill-pay product, with Billy Porter, Billy Zane, and Jessica Vosk, and invited CFOs. Anyone named Bill got a nametag. Reviews were mixed. (TheaterMania, InsideHook)
7. Show your work
In May, Ipsos and Syracuse's Newhouse School showed 3,000 people 20 ads, pairing human-made ads from before 2021 with AI versions built from the same briefs. Only 13% were confident an AI ad was AI, and the same share wrongly flagged the human-made ones. People can't tell. What they react to is whether it looks like anyone cared. (Grey Journal)
So brands are proving it. Prada's holiday campaign, out last week, was drawn by hand by a team led by veteran animator John Pomeroy, and Prada released a behind-the-scenes video of them doing it. Inc. called it effort signaling. Coca-Cola's AI-generated Christmas ad is getting hammered for the second year running. (Inc., Euronews)
That doesn't mean the clever stuff has no AI in it. Google won the Cannes Grand Prix for AI craft this year with Project Genie, which let regular people generate and walk around in interactive worlds using DeepMind's Genie 3. (Contagious)
What am I missing? If you've gotten really good at something AI can't do for you, I'd like to hear it.