Your content sounds like everyone else's. That is a research problem, not a writing problem

Why does a tone guide not fix this?

Because tone is the last few percent. If two articles contain the same facts, the same structure, and the same examples, changing the adjectives does not make them different. It makes one of them slightly better written.

Watch how the sameness is actually produced. A writer is briefed on a topic, searches it, opens the top handful of results, and builds an outline from what those pages have in common. That outline is by construction the average of the existing coverage. Every competitor's writer did the same thing last month.

So the piece is finished before anyone starts writing. Voice work at that point is decorating a structure that was copied from the consensus, and readers detect it immediately even when they cannot name what is wrong.

This is also why adding a tool to the process changes nothing. Generating the same average faster produces the same average.

Where should the material actually come from?

From inside the company. Sales calls, support tickets, the questions clients ask in month two, the projects that went wrong, the numbers only you have. None of that is in the search results, which is exactly why it makes the piece different.

The right column is harder to produce and that difficulty is the moat. A competitor can commission the left column in an afternoon. They cannot commission your last twelve engagements.

How do you get it out of people's heads?

Interview them, badly and specifically. Twenty minutes with whoever does the work, recorded, asking for examples rather than principles, produces more usable material than a week of desk research.

The point of asking for examples is that principles are already public and examples are not. Anyone can say measure what matters. Only your colleague can say we spent four months reporting on a metric that turned out to be counting the same lead twice.

Record it and keep the transcripts. A single good interview usually contains material for three pieces, and the specifics fade fast if nobody writes them down while they are fresh.

• Ask for the last time this went wrong, and what the warning sign was
• Ask what clients get wrong about this before they hire anyone
• Ask what they would do differently with half the budget
• Ask which common recommendation they disagree with, and push until you get one
• Ask for a number, then ask where the number came from
• Follow anything they hedge on, because the hedge is where the real answer lives

What does a differently sourced piece look like?

It disagrees with something in the first paragraph. Not for the sake of it, but because material from inside a company almost always contradicts some part of the public consensus, and that contradiction is the reason to publish.

That opening is doing real work. It tells a reader within ten seconds that this is not the fourth version of the article they already skimmed, which is the only thing standing between your page and the back button.

It also gives the piece a spine. If the claim is that the usual advice about this is wrong in a specific way, everything after it either supports that or gets cut. Articles assembled from consensus have no spine, which is why they drift into listing considerations.

And be prepared to be wrong in public occasionally. A position that cannot be wrong is not a position, and the pieces that get shared are the ones where the author took a risk a committee would have removed.

Does this change how you use AI in the process?

It clarifies it. A model is very good at the parts that were never the differentiator and structurally incapable of the part that is, because it is working from the same public material that made everything sound identical.

Use it for the scaffolding. Turning a transcript into an outline, tightening a draft, producing distribution variants, catching the sentence you wrote three ways. That is a genuine productivity gain and it costs you nothing distinctive.

Do not use it for the claim or the evidence, because it does not have your evidence. A piece generated about a subject the company has not done reads as plausible and slightly wrong to anyone who does the work, which is precisely the audience you were writing for.

The useful mental test: if the model could have written this paragraph without ever speaking to your team, ask why it is in the article at all.

Does this survive at volume?

No, and that is the finding rather than an objection. A process that depends on original material cannot produce eight pieces a month, which is a good reason to stop trying to produce eight pieces a month.

Be honest about the arithmetic. If every piece needs an interview and something only you know, the realistic output for a small team is one or two a month. That sounds like a downgrade until you compare it against what the eight were achieving.

The volume can move to distribution instead. One substantial piece supports a month of posts, an email, a talk, and several sales conversations, which is where the return actually is once general answers are free everywhere.

There is a middle setting that works. One original piece a month, plus maintenance on the pages that already earn attention. The maintenance is the part everyone skips and it is usually the better afternoon.

How do you know if it worked?

Two tests, both cheap. Could a competitor have published this piece unchanged, and did anyone inside your company want to argue with it.

If a competitor could publish it with their logo swapped in and nothing would be false, it is not yours. That is a harsher test than it sounds and most content programs fail it on every piece.

The second one is a better predictor of whether it will travel. A draft that provokes a real argument in a review meeting has a claim in it. A draft everyone approves without comment has been sanded down to the consensus, which is where it started.

Then ask the sales team, a month later, which piece came up in a conversation. That is the only measurement here that survives contact with reality.