Most "build your personal brand with AI" advice skips the only part that matters. It tells you to generate 30 LinkedIn posts, schedule them, and watch the followers roll in. You do it. Nothing happens. The posts are fine. They're grammatical, they have a hook, they end with a question. They also sound exactly like the four thousand other posts published that morning, because they were built the same way: ask the model for content about your topic, accept what comes back.
The problem isn't the writing. You skipped positioning. A personal brand is a position in someone's head — what you're known for, who you're for, what you'll argue that others won't. Content is just the evidence. If the position is fuzzy, more content makes it fuzzier, not sharper. AI is genuinely useful here, but for the layer underneath the posts, not the posts themselves. That's the part worth getting right.
Why AI "thought leadership" usually flops
Picture what happens when you ask Claude, cold, to "write a thought leadership post about marketing for coaches." It has no idea what you believe. So it does the safe thing. It averages. It produces the median take, because the median is what's most represented in everything ever written on the subject. That's how these models work, as best I understand it. They're pulled toward the centre of the distribution.
The centre of the distribution is the one place a brand can't live. Being known requires being slightly off-centre. A position is, almost by definition, a claim that not everyone agrees with. "Consistency matters" is the average. "Most coaches post too often, and it's quietly eroding their authority" is a position. The first is forgettable. The second makes someone stop scrolling, because it pushes against what they assumed.
So when people say AI content sounds generic, they're half right. The output is generic when the input is generic. You gave it a topic, not a point of view, and it handed back the average of that topic. Fix the input and the same model becomes useful. The fix is doing the positioning work first, and using Claude as a thinking partner for it rather than a content vending machine.
Get your raw material out of your head first
You can't prompt your way to a position you haven't found yet. Before you open a chat window, you need raw material, and the best source is you talking. Record yourself for ten minutes answering three questions, then transcribe it. Your phone's voice memos plus any transcription tool will do.
- What does nearly everyone in my field believe that I think is wrong, incomplete, or outdated?
- What do clients keep getting surprised by when they work with me?
- What's the thing I find myself explaining over and over because people arrive with the wrong assumption?
Ramble. Don't edit. You're trying to capture how you actually talk about your work, including the throwaway lines, because those throwaway lines are usually where the real position is hiding. People bury their best opinions in asides.
Now you've got something Claude can work with that isn't the internet average. It's you.
Use Claude to find the position, not write the post
Paste that transcript in and resist the urge to ask for content. Ask for analysis. A prompt that actually works looks more like this:
"Here's a transcript of me talking about my work. Don't write anything yet. Read it as if you were trying to figure out what I genuinely believe that's contrarian or non-obvious. Pull out every claim I'm making, even the half-formed ones in asides. Then tell me which one or two are the most distinctive — the ones I could build a reputation on — and which are just table stakes everyone in my field would agree with. Be blunt about which of my opinions are actually boring."
That last line matters. Left unprompted, Claude will be encouraging and call everything insightful. Tell it to separate the distinctive from the obvious and it becomes a far better mirror. You're using its breadth. It has read enough of your field to know what's a common take and what isn't, so it can locate where you're standing apart.
What you're hunting for is one defensible claim. Not your whole philosophy. One sentence you'd be willing to put your name to, that a reasonable competitor might disagree with, that your ideal client would feel relieved to hear someone finally say. A fitness coach's might be: "Most plateaus aren't a training problem, they're a recovery problem, and the industry sells more training because that's what's profitable." A fractional CFO's might be: "Founders don't have a cash-flow problem, they have a pricing problem they've labelled as cash flow." Specific. Slightly uncomfortable. Yours.
Then pressure-test it. Ask Claude to argue the opposite as persuasively as it can. If the counter-argument is stronger than your position, you've found a weak spot to either drop or sharpen. If your position survives a genuine attempt to dismantle it, it's solid enough to build on. The "generate 30 posts" approach never reaches this step, and it's the whole game.
Then the boring-but-load-bearing definitions
A position needs three boundaries around it, and these are worth nailing down in writing before any content exists.
Who it's for, named tightly. Not "entrepreneurs." "Solo course creators doing their first £50k who've never run paid ads." The tighter the who, the sharper everything downstream reads, because specificity is what makes a stranger think that's me. Have Claude draft three candidate audience definitions at different widths and pick the one that feels almost too narrow. Almost too narrow is usually correct.
What you're against. Every strong brand has a quiet enemy: a common practice, a lazy default, a piece of received wisdom. Yours falls out of the contrarian claim you already found. Make it explicit. "I'm for X, which means I'm against Y" gives every future post a spine.
What you won't talk about. Counterintuitive, but the things you refuse to comment on define you as much as the things you push. A brand about deep work shouldn't be posting hustle-culture motivation. Write the no-list down. It saves you from the slow drift into commentary on everything, which is how distinctive people become background noise.
Keep all of this in one short document. Call it a positioning brief. A page is plenty: the audience, the core claim, the enemy, the no-list, and five or six phrases in your own words pulled from that transcript. This page is the thing you feed Claude every time you sit down to write, so the model is expressing your position instead of reaching for its default average.
Now, and only now, the posts
With the brief in hand, content generation changes character. You're no longer asking Claude to invent opinions. You're asking it to dress opinions you already hold in different formats. "Take claim three from my brief and write it as a short LinkedIn post that opens with the moment a client realised it." "Turn the enemy from my brief into a newsletter intro." "Give me five post angles that all defend the same core claim from different entry points."
The output stops sounding like everyone else, because the substance is no longer the internet's substance. It's the brief's. One useful discipline: every post should trace back to a line in the brief. If you can't point at which claim a post is making, it's probably filler, and filler is what dilutes a brand back toward the average you worked to escape. Run drafts back through Claude with one more instruction — "flag anything here that sounds like generic advice anyone could have written" — and cut what it flags. You're turning the model's pattern-matching against its own worst tendency.
This is also where you'll feel the difference over time. Positioning compounds. Ten posts that all reinforce one sharp claim build a reputation. Fifty posts that wander across topics build nothing, however polished each one is.
Where this sits
None of this requires AI, strictly. People built sharp personal brands long before any of these tools existed. What Claude changes is the friction. The transcript-to-position work that might have taken a strategist and a few hundred pounds now takes an afternoon and a good prompt. And the pressure-testing step — having something argue against you, hard, before you commit — is genuinely hard to do alone and easy to do with a model that won't get defensive on your behalf.
If you want this as a repeatable system rather than a one-off — the positioning brief, the contrarian-claim extraction, the voice-matching, the formats — that's roughly what we packaged into the Personal Brand pillar at Skill Locker. It's our product, so take that as the disclosure it is. The method above stands entirely on its own. Do the positioning first, keep the brief in front of you, and the posts more or less write themselves. They'll sound like you instead of like the average of everyone who's ever written about your topic. The order is the whole point. Position first. Posts second. Most people do it backwards and wonder why the algorithm ignored them.