Last Updated on October 20, 2026
Pretty much every marketer uses AI now. That’s no longer the differentiator. The real question is whether you’re using it in a way that actually changes your results, or whether you’re doing the same work you always did, just a little faster.
This is a practitioner’s guide for thinking about where AI actually makes a difference, where the traps are, and how to move from AI-curious to genuinely advanced.
Everyone in marketing is using AI
The floor has risen across the field. A marketer who struggled with copywriting two years ago is now generating website copy without issue. Someone who couldn’t put together a coherent brief is producing one before the kickoff meeting even ends. A marketer who was never particularly analytical can upload a raw .csv export and generate an insightful readout for leadership. This is a great thing.
But it also means the baseline has shifted. If you’re using AI for the same things everyone else uses it for — draft this email, summarize this document, give me five subject line options — you’re not gaining an edge. You’re keeping up.
The marketers who are excelling are using AI to do things they weren’t doing before. Not better versions of existing work. New work entirely.
How great marketers use AI
Three things separate marketers who get genuine leverage from AI and those who are just keeping up.
The first is that great marketers stay focused on results, not throughput. Anyone can point to AI usage. It’s much harder to point to AI impact. You could spend a week building a sophisticated-sounding automated workflow and call that progress, all while your CPAs are rising and growth has hit a wall. The discipline of connecting AI usage to actual business metrics is what separates optics from outcomes.
The second is novelty. The real leverage isn’t getting faster at things you were already doing. It’s figuring out what AI makes possible that you simply weren’t doing before.
When we were building the go-to-market strategy for a new program launch at Udacity, we had five options on the table: a virtual event, an influencer program, a contest, and a couple of others. In the past, we’d have narrowed the list using team judgment, historical performance data, and instinct.
We still did that. But we added a step that wasn’t previously practical (at least not at the same speed and price point): we built detailed synthetic personas of our target students, then used Claude to run a simulated panel, asking which approach would most appeal to them, what their concerns would be, what would push them to convert. We used those responses to pressure-test our instincts before committing.
Is synthetic persona research as definitive as real human research? No. But it gave us higher confidence in the direction we chose, and it surfaced objections we hadn’t considered. That’s a capability we didn’t have two years ago.
The third is that great marketers treat setup as the skill. The quality of AI output is almost entirely determined by what goes in. Great marketers invest time in the frame — the persona, the brand voice guidelines, the evaluation criteria, the example of what good looks like — before asking AI to do anything. Average marketers type a short, generic prompt. Great marketers have already built the context the tool draws on.
Where marketers get stuck with AI
Most people can get to the level of using AI for individual, disconnected tasks. Ask it to analyze a report. Write some copy. Do competitive research. That’s table stakes now.
Few marketers learn how to move beyond disparate tasks and connect the pieces into workflows and systems. Executing individual tasks with AI is easy. The next step is orchestrating how competitive research feeds into the brief, how the brief goes to a synthetic panel for pressure-testing, how the panel feedback gets integrated, and how you eventually collect performance data and feed it back into the process. Building a workflow rather than a task list is where most marketers stall.
A second sticking point is what many are calling multiplayer AI. Most people have their own setup: their own Claude, their own prompts, their own habits. That works fine individually. But when you’re on a team, you need connective tissue. If a colleague does market research that would be useful to you, you don’t know it exists. There’s no shared infrastructure, no institutional memory. Solving that is more of an organizational challenge than an AI challenge — but it’s one of the bigger blockers to AI actually scaling across a marketing function.
A third is subtler. After a while, people get comfortable. Maybe too comfortable. Early on, you spend a lot of time reviewing and correcting output because you don’t fully trust it. Then confidence builds, quality improves, and you start releasing things with less scrutiny. That’s when reports full of vague, generalized language start making their way to stakeholders — language that doesn’t reflect your actual business, your actual voice, or your actual situation. The trap isn’t that AI gets worse. It’s that people stop looking closely.
The stages of AI capability building
Stage 1: Ad hoc tasking. Individual prompts for writing, summarization, research, etc. Useful, real time savings, but it’s all one-off tasks to do the same old work a bit faster.
Stage 2: Context mastery. You’ve figured out that output quality tracks input quality. You start giving AI detailed personas, brand voice guidelines, evaluation criteria, examples of what good looks like. Prompting stops being a search query and becomes a craft. The same tool starts producing noticeably better results, and gets even better with the next stage…
Stage 3: Tool integration. You connect AI to your actual data sources and platforms — your analytics stack, ad platforms, CRM, Google Drive, and Slack. Now it’s getting context on the fly with real, live information instead of what you remember to paste or upload.
Stage 4: Higher-order thinking. With real data and strong context in place, you start using AI for more than execution. You use it to stress-test ideas, challenge strategic assumptions, synthesize research, and model decisions before committing.
Stage 5: Workflow design. You move from individual tasks to connected sequences. AI doesn’t do one thing — it does a chain of things, where each output feeds the next input. You start feeling like you’ve built something, not just used something.
Stage 6: Agentic operation. AI runs systems with meaningful autonomy and you review the outputs rather than doing the work. You’re the orchestrator.
Most marketers are somewhere between stages 1 and 3. Stages 5 and 6 require real intentionality and a willingness to get a little technical, but they’re where your marketing expertise gets supercharged and where your real competitive advantage starts to emerge.
What this progression looks like in practice
Let me walk through how these stages played out in my own work, using the weekly business review my team runs every Wednesday.
We review key metrics across the channels we each own. I manage a handful of them, which means a spreadsheet with several tabs that needs updating before every meeting.
In the early stages, I was doing that manually and then uploading the CSV to Claude to ask for insights. It helped. But Claude kept missing context that changed what the data actually meant. There was a stretch where SEO traffic would spike, and Claude would flag it as a meaningful signal. When I dug in, I’d find the traffic was from a geography where we run a non-commercial program that was irrelevant to the consumer business. Claude didn’t know that. I had to explain it again every week.
So I started giving it more context upfront. But that was still manual and still imperfect. To move to the next level, I needed integrations on two fronts. The first was data: connecting Claude directly to our ad and analytics platforms and Google Suite so it could pull and update the spreadsheet without me. The second was context: connecting it to Slack and email so it could scan what had actually happened that week — campaigns launched, anomalies the team flagged, performance updates from agencies — without me having to brief it manually each time. Once both were in place, the output shifted from noise to something I could actually act on.
Now the whole thing runs as an agentic workflow. I tell Claude to run the WBR update. It pulls the data, scans for context, writes the summary, and delivers it. What used to take two hours takes under 30 minutes. More importantly, I show up to the meeting having actually thought about the data rather than just collected it.
Risks of using AI in marketing
The question isn’t really “what should you not use AI for.” Almost every part of marketing is fair game. The better question is where the temptation to rely on AI completely tends to outrun the results.
Synthetic research for creative evaluation. Synthetic panels work well for stress-testing strategy and narrowing conceptual directions. They work poorly for evaluating finished creative. Ad creative is an emotional, human response, and you won’t get a reliable read from a simulated panel on whether an ad will land. Use AI early in the creative process. Use real humans to evaluate the output.
Unreviewed output going to market. The more content AI generates — social posts, email copy, ad creative, product descriptions — the easier it becomes to approve and ship without really reading. Not just scanning for typos, but checking whether the claims are accurate, the voice is on-brand, and the message is what you’d actually want a customer to see. At scale, this is how brands start sounding generic, stop connecting with customers, and invite errors and inaccuracies. QA shouldn’t disappear because AI wrote the first draft.
Losing touch with the business. There’s a subtler risk in automating well: you can get so many steps removed from the data that you lose the thread. If AI is pulling the metrics, writing the summary, and generating the talking points, you could technically run a weekly business review without ever really absorbing what’s happening. That kind of drift is hard to notice until someone asks a question your agent’s output can’t answer for you. Automation is for the collection; the understanding still has to be yours.
Communication that loses its humanity. AI can make you more articulate and organized in how you communicate with colleagues. That’s useful. But there’s a risk that people end up two or three AI interactions removed from each other. The relationships that matter on a team are worth protecting from that kind of abstraction.
It’s a great time to be a marketer
The duality of the current moment is that it’s both a genuinely exciting and unsettling time to be a marketer. The capabilities available now — things that would have required a team, a budget, and weeks of lead time just a few years ago — are increasingly within reach for anyone willing to learn. And yet, AI can do a lot of things many of us thought we were uniquely good at. Which raises the question, what does it actually mean to be good at this work? I don’t have a clean answer to that. Until I do, my advice is to stay curious, keep learning and experimenting, and never let the tools take over all the thinking for you.
