AI is good at starting and bad at finishing. It generates ideas at volume, repurposes one piece of content into many formats, personalizes across segments, and finds patterns in data. It cannot set strategy: who to target, why they should care, or what offer would move them. Content pasted straight from a chatbot has a generic feel audiences detect, and it spends trust instead of building it. Use AI for candidates, never for finals, and keep human judgment on whether the output is worth multiplying.
Every client asks me about AI now, usually in the first meeting, usually with some mix of hope and guilt. The hope is that it replaces a marketing team they can’t afford. The guilt is that they’re already pasting ChatGPT output into their newsletter and suspect they shouldn’t be.
I use AI tools openly in my own work. My clients know, and it’s part of the stack rather than a secret. So this is not an argument against the tools. It’s an argument for knowing exactly which job they’re good at, because the companies getting burned are the ones asking AI to do the one thing it can’t.
What AI Does Well
AI is changing what’s possible for a small marketing team. Output that used to require a team of fifteen can, in some cases, be produced by two or three people with the right tools. The gains concentrate in a few specific places.
Idea generation at volume. Feed a model everything you know about your audience, from demographics to pain points to the specific language customers use, and ask for fifty angles. Most will be mediocre. Three will be ones you’d never have come up with, and picking from volume is exactly what the current tools are for.
Repurposing. Most of the cost of content sits in the original creation. Once a webinar or a long article exists, AI can turn it into a written summary, social posts, an email sequence, and a dozen segment-specific variations for a fraction of what that used to take. Repurposing is one of the few real economies of scale in content, and AI has made it dramatically cheaper.
Personalization across segments. The campaign that worked for VPs of Operations needs different hooks for IT Directors at the same companies. Drafting those variations is mechanical work AI handles well, once a human has decided what each segment needs to hear.
Pattern-finding in data, and routine communication that doesn’t require nuance. Surfacing what would take days to find by hand, and drafting the confirmation emails nobody should be writing from scratch.
Where It Burns You: Finishing Instead of Starting
AI gets you in trouble when you use it for finishing rather than starting. Output pasted straight from a chatbot has a recognizable shape and a generic feel, and your audience picks up on it, sometimes consciously and often not. It has no insight of its own, no specific experience behind it, and no point of view. If you ship it untouched, the content reads exactly like what it is.
This matters more than it seems, because generic content doesn’t just underperform. It spends trust. The whole job of content is to make a specific reader feel that you understand their specific problem better than anyone else does. Words that could belong to any company in any industry accomplish the opposite.
The model’s job is to generate candidates. Picking which candidate is worth developing, and then developing it with the specificity and perspective only you can bring, is still your work. The real story, the actual number, the opinion you’re willing to defend: none of that comes out of a prompt.
Want the full framework?
This article is adapted from How to Grow Any Organization by Tom Zandstra. The book covers all three pillars in depth, with real client examples and action steps for each chapter.
Download the free book →What It Can’t Do at All
AI can write a hundred emails. It cannot decide who those emails should go to, why those people should care, or what offer would move them. That work is strategy, and it sits underneath all the production. No amount of output fixes a program that’s pointed at the wrong audience with nothing worth saying.
The shorthand I use with clients: let AI multiply the output, and keep human judgment focused on whether the output is worth multiplying.
This is also why AI hasn’t changed my answer to what makes marketing work. The three things that separate companies that grow from companies that stall are knowing the audience, providing real value, and running a connected system. AI accelerates all three when they exist. It also accelerates their absence. A company with no defined audience can now produce ten times the content for nobody in particular, and the main effect of the tooling is that they reach nobody faster.
Four Rules for Using It Without Getting Burned
Feed it your audience research first. The quality of what comes out tracks the quality of what you put in. A model loaded with your customer interviews, review mining, and sales-team notes produces usable angles. A model prompted with "write a blog post about our industry" produces the same post as everyone else in your industry.
Use it for candidates, never for finals. Drafts, angles, variations, outlines: yes. Anything that ships with your name on it gets rewritten by a person, with a real example and a point of view added.
Give it the mechanical middle of the funnel. Segment variations, repurposed formats, routine sequences. That’s where the fifteen-person output from a three-person team actually comes from.
Keep strategy meetings human. Who we target, what we offer, what the numbers have to show: if those decisions are being made by autocomplete, no one is making them.
