Couldn't I just use Claude myself?
In first conversations, an accomplished owner will often ask me a version of the same question. Couldn't I just use Claude myself, and skip hiring a marketer or an agency at all?
They ask because the tools invite the question. Why pay for outside marketing help when AI this capable is a subscription away?
My answer surprises people.
Yes. You could.
I should know. Claude is the tool I work in all day, and I have used AI since the first weeks after ChatGPT launched. The model you could subscribe to this afternoon is the model I use. I do not get a better one than you would.
So before anything else, here is what owners who tried the path report back.
Owners who tried it report two failures
The first failure is generic output. An owner asks the model for marketing and gets back work that is competent, correct, and interchangeable with a competitor's. Owners describing this reach for words like gimmicky and soulless, and some quit the tools over it.
That distrust of the output is measurable, and it is the norm rather than the exception. In a June 2026 Bluehost study of 350 U.S. small business owners, 87% already used at least one AI tool, yet only 6% said they highly trust AI to write in their brand voice.
The failure has a mechanism. A model writes from what it was trained on plus whatever you hand it in the moment, and the trained layer is the same public internet a competitor's model holds. Hand it nothing of your own and it writes the average of the internet, fluently, at scale.
The second failure is drift. One working session does not remember the last, so owners producing content in batches report work that loses track of itself. The tone wanders. Small facts wobble. The manual workarounds start eating the hours the tool was supposed to save.
Look at what the two failures share: Both are system problems, and both get blamed on the model. However, the model is the wrong suspect.
The same model, fed your specifics and governed by extensive written rules, can produce work with your fingerprints on it. I make that happen daily.
An interview-trained marketing system fixes both
You could subscribe to an AI tool this afternoon. The part you cannot subscribe to is the system around it.
What I build for clients is an interview-trained marketing system. It has three crucial parts.
- Capture. Hours of recorded interviews (or guided monologues for owners who prefer talking alone) pull out your stories, your numbers, and your judgment. This is the valuable material no model can invent. It's what people, search engines, and AI crawlers want to read. And it's impossible for your competitors to authentically copy it. This is the part that cures the generic, and it is why the interviews carry most of the weight.
- Documentation. More than a dozen custom brand guides are written from those recordings. These guides cover your positioning, your ideal clients, your voice down to your sentence habits, your services, your strategy, and more. The guides live inside a Claude project that carries the same rules into every working session. This is the part that cures the drift.
- Refinement. The brand guides never stop improving. When a produced piece misses, I manually correct it. Then Claude updates the guidelines to make the improved approach more likely to occur in future tasks. Every asset produced makes the system better at producing the one after it. Few DIY owners build brand guides at all, and fewer still refine their guides every day they use them. Yet this feedback loop is what drives compounding improvement and efficiency gains.
Run together, the three parts do the work of a marketing department with just one person working.
How to build the system yourself
I have no interest in keeping the recipe vague. Here is how I would start.
Open a Claude project for your brand. Tell Claude your business goals. Then ask it to interview you: have it write out the questions it would need answered to market your business well. Those questions become your capture agenda.
Answer the questions out loud, recorded. Do it in interviews with a colleague or in solo monologues, whichever you will keep doing. Speech matters because it carries your specifics and your voice in the same pass. Typed notes flatten both.
Have Claude turn the corrected transcripts into brand guides, saved as documents in the project. To give you a sense of what these cover, here are some of the guides in my own system:
- a brand foundation: mission, positioning, and promise
- ideal client profiles: the two or three readers every piece is written for
- a voice and tone guide, plus writing preferences down to sentence habits
- a founder philosophy guide: how you reason and who shaped your thinking
- a services guide that states plainly what you sell and for whom
- a growth playbook: the business model and the levers you will pull
Content planning is its own step, with its own prompts. Ask Claude to interview you again, this time to build a content map: the full list of questions your audience asks, with each question becoming one post. Produce your content from that map, in order of what your best prospects ask first.
Review every piece before it publishes. Whenever you see something that could be better, fix it. Then have Claude revise the brand guidelines to better align with your revisions. That double fix is an important competitive advantage, and it's a step most people will skip.
Finally, keep capturing. Thought leadership holds an audience only while you keep contributing lived experiences that interest your audience. So the capture habit continues, even at the pace of one client question answered out loud each week.
When you keep capturing, the system keeps writing ideas that only you could say. Your time is needed to keep supplying meaningful insights and stories. If you stop feeding the system, it will revert to writing the Internet's average again.
Whether building it yourself makes sense for you
For some owners it does. In my experience, four things need to be true.
You write well and enjoy the craft, because you are the reviewer of last resort.
You know some marketing. Claude can supply a strong strategy when you feed it the right context and instructions. Knowing the field is what lets you do that efficiently: you ask the right questions, use the right vocabulary, and judge the answers. That saves immense time and gets better results than a novice would get.
You have put serious hours into Claude itself, or whatever platform you choose. Power users know the advanced features: projects, custom instructions, how to structure context, how to make a correction stick. Those features are where the high-end results live. Most owners type a prompt into a blank chat and judge AI by what comes back. This difference explains why power users get results the typical DIY effort cannot reach, and why so many owners try AI, get generic output, and give up.
And you can publish consistently. A single considered post a month rarely moves results unless a large audience already follows you. The system pays when it runs weekly, which is exactly what makes recurring hours and ongoing effort worthwhile.
The choice comes down to hours and skill
So you have two options: build and run the interview-trained system yourself, or hire someone to build and operate it for you.
Build it yourself and the hours run in two phases. The build takes a first-timer a long while, at exactly the moment their business demands the most time from them. The weekly operation then continues for as long as you publish.
Alternatively, you can hire someone to build the system for you. In this case, your part compresses down to the capture and the final review. Someone else handles the system setup and refinement. About four hours a month of monologues or interviews keeps your material original while the internet fills up with the generic kind.
The crucial comparison is opportunity cost. Count the hours a proper build and weekly operation would take you at your current skill level in Claude. Price them at what an hour spent with your own clients earns. For most owners, the math points to hiring. The hours you save go back into billable work.
Whichever way you go, aim deeper than a prompt. A prompt a few paragraphs long is no match for a system that knows you deeply. The work I ship draws on more than a dozen documents, deep research, and hours of a founder's recorded thinking. That is context comparable to what a longtime marketing director knows about you and your business.
Build that depth yourself or hire someone to build it for you. But do not publish from a shallow prompt and conclude that AI cannot sound like you. It certainly can.
