Where AI quietly makes marketing worse. And the line I won't cross.

Where do you see AI quietly making marketing worse?

In engagement. Comments on posts, and direct outreach, where the whole point is that a real person is paying attention, and AI slop turns up instead.

You can spot it in half a second. An instant reply, or a fresh comment on your post, that is clearly the output of an AI nobody bothered to give a voice. The concept behind it is fine. Engaging with people at scale is reasonable. The execution is the problem: people put almost no effort into their AI's tone, so it produces generic, business-Deloitte fluff with em dashes all through it, and that's the opposite of what social is for.

So much of B2B marketing now runs through real people's accounts. Backing your thought leaders with robotic slop makes them look worse, not busier.

So how do you use AI in engagement without sounding like a robot?

Give it your actual voice, and keep it light. Two things do most of the work.

Feed it how you really talk. When I brief a blog I record a full voice note in my own words. That gives the AI my tone, my personality, the things I'd notice. Copy that habit and your engagement stops reading like a template, because it's trained on you, not on a corporate style guide. It's the same principle behind our content work.

Respond to the human bit, not the whole article. The common mistake is setting AI to read the first line, go and research the topic elsewhere, and return a generic take. What people want is a human response to the human part of what they wrote. If someone shares a funny story from a conference, the good reply is "ha, that reminds me of the time...", not "a very interesting insight, in fact 3.5% of...". And please retire "what are your thoughts on this subject?" It's the most AI thing you can end on.

What's your rule of thumb for what's safe to give AI versus what must stay human?

Whenever it's a human interaction, the safe thing to give AI is the ability to respond like a human. The thing that must stay human is the deep insight. You don't want AI answering someone with a half-remembered stat from an article you've never read. Save the real insight and the impressive numbers for your own posts, where you control them.

The other half is sales. If you've finally got someone in your ideal audience engaging with you, that's the moment you've worked for. Automating the reply there isn't a time save, it's throwing the lead away. Knowing which parts to automate and which to protect is the whole game, and it's closely tied to lifecycle and how you decide what to automate in GTM.

What about regulated industries?

If you're in finance or anything regulated, put a gate in that has one job. Not to write the post, not to think about it, not to send it. Just to sit in the middle, take whatever the AI produced, strip out the context and reasoning, and check it cold against a fixed list. Does it touch any known issues? Does it clear the required points? Anything that fails, or anything outside the pre-approved set, stops and waits for a human. That's the pattern that lets you use AI at scale without it wandering into something it should never have said.

If you want help drawing those lines for your own team, book a free Growth Audit.