When people say "AI for marketing" - what do they actually mean? And what do they get wrong?

Most people hear it and immediately think: generate more, do more. There's a term going around right now - "AI slop" - and I think it captures the split we're seeing. People who are newer to this get excited about the volume of things they can produce. People who've been using it for a while start to notice when something feels off - when an output is technically fine but has no real substance behind it.

For me, there are two genuinely useful things AI does. The first is helping you think - exploring ideas you'd normally skip because the manual effort to test them wasn't worth it. Now you can actually go down that road and see if it leads somewhere. The second is the production side: creating content, analysing data, connecting systems that wouldn't otherwise talk to each other. Both matter. But the first one gets undersold, and the second gets oversold.

Where does the overselling happen most?

Vendors who say "one click and it's magic." You see it constantly in marketing tech right now. The pitch is that you just feed your data in and everything happens automatically - no decisions needed.

That's not how it works, and honestly I don't think it ever will be. There always needs to be someone in the room who had the original thought and is willing to own the decision. AI is good at surfacing what happened and mapping out options. It's not good at knowing whether a decision is right for your business, your audience, your moment. It's also always happy to please - and if you let it please you without questioning it, you'll never know if the answer was actually correct.

Does it make marketing more creative, or just faster?

Both, genuinely. It makes it faster in obvious ways - something that used to take a team two weeks to prototype can be done in a day now. But it also opens up creative territory that used to be too expensive to explore. If I have an idea for a campaign, I can search six months of customer and prospect conversations and find ten real examples of people talking about that exact problem. That changes how you write the brief, how you position the message. The creative thinking is still yours - but the material to build on is much more accessible.

Generic outputs are a real frustration for a lot of people. How do you avoid them?

Context. That's the whole answer.

I have a personal database of every task I've completed with AI. Every time I finish something, I have it write a session log - what we did, what worked, what I decided and why. The next time I do something similar, I pull those logs and start there. It's not drawing from the internet; it's drawing from my own work history. It's a self-improving system - the more you use it, the more it understands how you think.

The prompting frameworks people used to spend time on - give it a role, give it context, give it a goal - those are becoming less important as models get better at inferring. What still matters is giving it the specific context relevant to your work. That's where generic outputs come from: not bad prompts, but empty context.

Does AI help with strategy, or is it more of an execution tool?

I use it for strategy more than most people probably expect. I treat it like talking through something with a colleague. Here's the goal, here's where we are, here's the data - now help me map out a few scenarios.

That's mostly how I use it for strategic work: scenario planning. Not "what should I do," but "if I go this way versus that way, what are the likely consequences of each?" I make the call. AI helps me compare options more rigorously than I could alone, and faster than waiting for the perfect data set. Getting something to market and learning from real feedback is almost always better than spending months modelling the ideal strategy.

Two examples where it saved serious time?

The first is LinkedIn advertising reporting. Before, every Monday morning I'd export three reports, work through the data manually, build the analysis, decide what to change. Half a day, minimum. Now that process runs automatically on a schedule - it pulls the same data, follows a framework I built, and delivers a structured report covering spend, creative performance, engagement, and recommendations. It's ready before I've logged in. I still check it manually - you can't skip that - but it saves me around four hours a week just on that one task.

The second is post-webinar production. That's a grind: wait for the recording, clean up the transcript, build the landing page, write the blog, create sales talking points. Two or three people, most of a day. Now we have a workflow that takes the raw recording, cleans the transcript automatically, builds the HubSpot landing page with the right imagery, and produces a kit with blog angles, ad ideas, and outreach copy for the sales team. A marketer still reviews everything - that step is non-negotiable - but the volume of manual work is gone.

And where has it failed?

Most often in accuracy. The one that stung: I'd used AI to do the analysis for a presentation, checked that stage carefully, then had it build the slides. Assumed the numbers would carry over correctly. They didn't - it pulled a stat from the wrong data set. I only caught it after the meeting. The rule I've had since: if you can't point to exactly where a number came from, don't put it in front of anyone.

Code is the other area. Building that webinar workflow took 10 or 15 iterations before it ran reliably. It kept extracting the wrong content or formatting things incorrectly for HubSpot. You have to be patient with it and willing to test properly before you trust it at scale.

Have you ever killed an AI project entirely?

I have a graveyard. Some were too complex for what they actually needed to do. One I remember clearly: I built a shared marketing library where the team could upload and access HTML files. Solved a real problem in theory - Claude tends to generate HTML files that are hard to share otherwise. In practice it was just another place people had to go, and we already had Google Drive. Nobody used it, and maintaining it fell on me. I shut it down.

The lesson from that: build on top of tools people already use. Adding a new destination rarely works, however well you build it.

Where does AI create more work than it saves?

Maintenance, mostly. When you build something with AI, you own it - and keeping it current is a manual job. AI is good at creating the first version; it's less good at iterating on the same file over time. You end up with version two, version three, a folder of things that are slightly out of date.

The other problem is the fire hose. I can now pull together hundreds of sources and send someone a ten-page PDF. They have no idea if it's useful. Half the time they put it straight back into an AI tool to summarise it. We end up in a loop of producing more than any person can actually read or act on. In marketing especially, the discipline of sharing only what's most relevant matters more now, not less.

What actually separates useful AI from AI that's just fast?

The ability to take messy, disconnected data and make it readable. That's the real unlock. Before, I'd be switching between tabs, cross-referencing manually, trying to hold context across different data sets. Now I can bring LinkedIn performance and N.Rich ad data into the same view and analyse them together. The insight isn't new - the ability to get there quickly is.

A feature earns its place when it shows you something you couldn't have seen by just exporting a spreadsheet. Trends across time, patterns across cohorts, signals that only appear when data is read together. That's where the value actually is.

Every product claims to be "AI-powered" now. How do you tell real from noise?

I genuinely dislike the phrase. It's still just technology - input, processing, output - and it still requires a human to decide what question to ask and what to do with the answer. Calling something "AI-powered" tells you nothing.

The real question is: does this feature help someone see something they couldn't see before, faster than they could find it themselves? If yes, it's valuable. If it's just automating something that was already easy, it's a feature in search of a use case.

What skills matter most for marketers now?

Breadth, more than depth in one area. Companies are consolidating roles - they want one person who can move across content, performance, product marketing - and AI is what makes that possible. The specialist skills in SEO or demand generation don't disappear, but one generalist with AI can cover more ground than a specialist could before.

The risk is that everybody ends up doing the same thing. If you replace your entire marketing team with one person and an AI running 40 posts a month, none of that content has anything unique in it - and eventually you won't have anyone left who knows how to fix it when things go wrong. The human perspective is what makes work worth reading. That doesn't go away.

Where would you tell a team to start?

Write down every task you do in a typical week. For each one: how manual is it, how long does it take, how often does it happen. Rank them. Then ask honestly how much of that task AI could help with.

I'd rather save three hours a week across five small tasks than spend six months building one impressive thing that doesn't change much day to day. That's also how you learn - by doing, by seeing what works, by building up a sense of what AI is actually good at versus where it lets you down. Start with one thing. Scale from there.

What won't change, regardless of where AI goes?

People still want to work with other people. I've been remote for five years - if my only interaction all day was with AI tools, I'd go mad. The best work still comes from people sitting with an idea they're genuinely excited about and building on it together. AI creates more noise. It doesn't create that.

You can see it in how events are coming back. People want to be in the room. That won't change.

This conversation has been lightly edited for length and clarity.