If you run an accounting firm, you've probably been hit with the same wave of AI tools and AI marketing pitches as everyone else. ChatGPT can write your blog posts. Midjourney can generate your images. There's an AI tool for every part of your marketing stack, and most of them promise to do the work faster and cheaper than a person.
The question isn't whether AI is real or whether it matters. It is, and it does. The question is how to use it well.
The short answer: Yes, you should be using AI in your marketing, but only in specific ways. AI is great for certain shapes of work and dangerous for others. Used well, it speeds up the parts of marketing that benefit from raw transformation — transcription, reformatting, framework analysis. Used badly, it produces marketing that looks polished and feels professional but is indistinguishable from every other accounting firm out there. The difference between those two outcomes is not about which AI tool you use. It's about whether the expertise that determines good marketing is in the loop at all.
A note on where this is coming from
Before we go further: an agency telling you to be careful with AI in your marketing sounds suspicious. The cynical read is that we're protecting our retainer by talking you out of doing the work yourself with cheap tools. We've thought about that, so let us state our position plainly.
We are pro-AI. We use it every day in our own work. We teach our clients how to use it. It's a real and important part of modern marketing, and that role is only going to grow. Anyone telling you to avoid AI entirely in 2026 is wrong, and probably about to fall behind.
What we're cautious about is specific uses, not the technology itself. The same way you'd tell a client that AI is fantastic for some accounting work and dangerous for others, we have a developed view on where AI fits in marketing and where it doesn't. Our skepticism is earned, not reflexive. We've watched clients use AI in ways that made their marketing worse, and we've watched ourselves use AI in ways that made our work better. The difference is not random.
The accounting parallel
You already do this work in your own field. When a client asks whether they should be using AI for their accounting, you don't say "yes, replace your bookkeeper with ChatGPT." You also don't say "no, AI is bad." You give a more textured answer.
Categorization of transactions, document summarization, first-pass research, drafting a client email: AI is genuinely useful for these. The cost of getting them subtly wrong is low, the work is structured, and a knowledgeable human is reviewing the output before it goes anywhere consequential.
Final review of a tax return, advisory recommendations to a client, judgment calls about what a particular business should do next: you tell them not to touch it. The cost of being subtly wrong is high. The work requires expertise that AI doesn't have. And the client can't tell the difference between confident-and-correct and confident-and-wrong, because they don't have the expertise either.
That second category is most of what people pay you for. It's where your judgment lives. Marketing has the same shape. Most of what makes marketing actually work for an accounting firm is in the high-stakes, judgment-driven category, and that's exactly where AI fails the most.
The closed-loop trap
Here's the specific failure pattern we see most often when an accounting firm owner tries to do their marketing with AI directly. It has three connected pieces, and once you're in it, you can't see your way out.
The first piece is bad inputs. AI's output quality is governed by the quality of what goes in, and the quality of input is governed by knowing what questions to ask. For an accounting firm trying to differentiate from every other firm in the same market, that means knowing what actually makes one firm different from another in this specific industry. That's its own form of expertise that takes years to develop. Most firm owners haven't done that work because it isn't their job — your job is accounting, not marketing. There's nothing wrong with that. But it means the inputs going into AI are usually generic, and generic inputs produce generic outputs.
The second piece is the inability to evaluate the output. This one is harder to talk about because it sounds condescending, but it's true and it matters. If you're not a writer, you can't reliably tell good writing from mediocre writing. If you're not a designer, you can't reliably tell good design from mediocre design. AI removes the technical barrier to producing copy and design but doesn't develop the trained eye needed to evaluate it. So firm owners end up with output that looks fine — it's grammatical, it's structured, it hits the obvious points — but is missing the things that actually make marketing work: specificity, differentiation, voice, the precise claim that makes a prospect feel like you're talking directly to them.
This is the part that's most invisible. Your prospects don't usually articulate "this copy is bad" or "this website looks amateur." They just feel something off and don't engage. The signal that something is wrong with your marketing is the absence of response, which is hard to read. Meanwhile, the output looks great on your screen.
The third piece closes the loop. When the firm owner wants a second opinion on the work they've produced, the natural move is to ask another AI tool. "Hey ChatGPT, is this blog post good?" The problem is that AI doesn't have taste. It has an averaged view of what's been written before. Asking ChatGPT to evaluate copy is like asking a focus group of every piece of writing on the internet what they think. The feedback regresses toward the mean. It tells you whether things are technically correct and whether they follow common conventions, not whether they're actually good for your specific firm and your specific audience.
This is worth flagging directly because we're seeing it constantly with our own clients. We'll deliver a piece of copy or a design comp, and the client will run it through ChatGPT and come back with the AI's feedback. Please don't. The feedback isn't useful. It's an averaged, generic critique with no understanding of the strategy behind the work, the audience the work is targeting, or what we're deliberately doing to differentiate your firm. Your own gut reaction — "this doesn't feel like us" or "I don't love this word" — is far more useful than anything AI will tell you, even when you can't fully articulate why.
The three pieces compound. Bad inputs produce generic output. The firm owner can't tell the output is generic. When they look for a second opinion, they get the same averaged feedback that produced the work in the first place. It's a closed loop, and every step depends on the same missing skill: industry-specific marketing expertise. AI doesn't supply any of it. AI just makes the loop run faster.
Why input expertise is the actual moat
The reason we spend so much time on industry-specific work isn't because we love accounting firms (though we do). It's because the input is what determines the output, and developing the right inputs for an accounting firm takes serious work.
When we sit down with a new client, the first phase isn't writing or designing. It's a structured process of figuring out what's actually different about this firm. Not what the firm owner says they do (which usually sounds like every other firm), but what they actually do that's distinct: who their best clients are and why, what they refuse to do that competitors do, what their owners believe about service or pricing or scope that drives real decisions, what their best clients say when they tell someone else about the firm.
That input work is where the differentiation lives. Once you have it, every piece of marketing that comes after has a real signal to work from. Without it, AI is just churning the same undifferentiated accounting-firm content that every other firm is also producing.
This is also where our marketing fiduciary positioning shows up. A vendor's incentive is to deliver work that looks like marketing happened. A fiduciary's incentive is to deliver work that actually moves the firm forward. Doing the input work properly is slower and harder than skipping it. We do it anyway, because that's the job.
AI is a real multiplier on top of that input work. Once we know what makes a firm different and have captured the firm owner's voice and perspective, AI can help us draft, reformat, and analyze faster than we could without it. But the multiplier is doing nothing if the input is bad. A 10x multiplier on zero is still zero. The expertise is the thing.
Where AI genuinely helps
The use cases below work because something else is supplying the signal — the firm owner's actual thinking, an existing piece of well-crafted work, or a defined framework. AI is doing transformation, not creation. That's where it shines.
Capture and transcription. Most firm owners think faster than they write. Voice-to-text on your phone, Otter or Fathom for client conversations, transcription of weekly video recordings: these all turn raw thinking into text that can become marketing material. The signal (your actual perspective, your actual voice, your actual experience with clients) is high-quality. AI just makes it available in a form you can work with.
Framework analysis. Feeding existing content through a defined lens — StoryBrand, jobs-to-be-done, the Endless Customers framework — to surface gaps and weaknesses works well. The framework provides the structure for the analysis. AI does the mechanical work of applying it. You still need someone with judgment to act on what comes back, but the surfacing step is genuinely faster.
Reformatting and repurposing. If you have a strong long-form piece (a blog post, a podcast episode, a webinar transcript), AI is excellent at turning it into derivative formats: social posts, email summaries, video scripts. The original work is the source of quality. AI is just adapting it. As long as the source is good, the derivatives can be good too.
Research synthesis. Reading three industry reports and asking AI to summarize the common themes is faster than reading them yourself, and the output is usable. You'll want to verify any specific claim before citing it, but as a starting point for your own thinking, it works.
Tactical drafting on structured tasks. Email subject lines, social caption variants, headlines for an ad you've already written, alt text for images. AI is fine for the kinds of small-scope tasks where the criteria are clear and the cost of a misfire is low.
Where AI fails specifically
The anti-patterns below all share the same shape. AI is being asked to create where it should be transforming, and to substitute for expertise where it should be amplifying it.
"Write me a blog post about [topic]" with no input. This is the cleanest example of the closed-loop problem. The prompt is generic, so the output is generic. The firm owner publishes it because it looks fine. Nothing about it differentiates the firm from any other firm. Repeat fifty times and you have a blog full of content that nobody reads and that doesn't move the needle on anything.
AI-generated headshots and stock-style imagery. A real photo of a real person at the firm signals trust. An AI-generated image signals the opposite, even when readers can't articulate exactly what's wrong. The uncanny-valley problem is getting better, but the deeper problem — that the image isn't actually evidence of anything — doesn't go away with better technology. Trust signals have to be true.
Running agency work through ChatGPT for feedback. Already covered above, but worth restating because it's the issue we deal with most often. The feedback is generic, the AI doesn't know your strategy or audience, and it crowds out your real reaction — which is the feedback we actually want.
"AI-powered" marketing pitches that promise volume at low cost. When a vendor's pitch is that they can produce 50 blog posts a month for $500 because of AI, what they're really pitching is the closed-loop problem at scale. The work will get produced. It won't differentiate you from anyone, and it may make your firm look worse rather than better, because the bar for noticeable content keeps rising as more AI content gets published.
How to think about this for your firm
The honest answer to "should I be using AI in my marketing?" is the same answer you'd give a client about AI in their accounting work. Use it heavily for the right things. Don't use it at all for the things where the cost of being subtly wrong is high.
The right things in marketing are the structured, transformation-shaped tasks: transcription, reformatting, framework analysis, research synthesis, tactical drafting on small-scope work. AI makes you faster at these without making you generic.
The wrong things are the strategy-shaped tasks where industry expertise determines whether the output differentiates you or makes you sound like every other firm. Positioning, messaging, cornerstone content, brand voice, design that builds trust. These require knowing what questions to ask and having the trained eye to evaluate the answers. AI doesn't replace that, and the closed-loop problem means that using AI for these tasks usually makes the output worse, not better.
We use AI every day. We're going to keep using more of it, in more sophisticated ways, as the tools improve. We also know which parts of our work it can't do, and we're not in any hurry to pretend otherwise. That distinction — heavy AI use in the right places, careful avoidance in the wrong places — is what we mean when we say we're pro-AI but cautious and intentional. It's the same disposition we'd hope you'd bring to AI in your accounting practice. The technology is a tool. The expertise is what makes the tool worth anything.