Hey, Mark Masters here.

Tuesday morning, new campaign for an existing client. A course launch.

I opened a fresh chat, and before I could write a headline I had to tell Claude what the course was, who it was for, what it promised, what made it different, and what proof we had.

I'd typed almost the same thing four days earlier, for the email sequence on the same launch.

By the afternoon I'd typed a version of it again, for the ad set. Three chats, three times reconstructing the same understanding of the same offer from memory, each one a little different because I was doing it from memory.

That's a tax. I've been paying it for years and calling it "getting started." So I built the thing that ends it.

A while back I wrote up giving every client a brain, a persistent file that holds the client's voice, their audience, the brand rules, so you stop re-explaining the client every session. That brain is real and I still use it, and it has one blind spot.

The blind spot is the offer itself. The brain holds who the client is and how they sound. What you're selling this week, and why anyone should believe it, was never its job.

A single client can run five offers in a year. A flagship course, a low-ticket tripwire, a done-with-you program, a renewal, a webinar.

The client brain holds what's common across all of them, the voice and the audience. What it can't hold is the part that changes with each offer: this promise, this mechanism, this proof, these objections, this price.

That part gets rebuilt in a chat window every time, which is exactly the tax I was paying on Tuesday.

The fix is a second document, scoped one level down. One per offer. I call it the Offer Dossier.

You extract it once, at the start of a campaign, and from then on every prompt for that campaign reads from it instead of from your memory.

The client brain and the dossier stack. Brain loaded for the voice, dossier loaded for the offer. Together they are the brief Claude used to make you retype.

Seven fields. Each one is a thing every downstream prompt needs and usually has to guess at. Here's the template I save at the top of every campaign.

Three of those fields do most of the work, and they're the three people rush.

The one promise is a discipline, not a summary. Most offers arrive as a pile of six benefits.

Your job in the dossier is to name the single promise the other five ladder up to, and to repeat that one everywhere. An offer with two competing promises reads as an offer with none.

The mechanism is the reason-why, the same one Peggy wrote about. The promise is what you get. The mechanism is why you'll believe you'll get it.

If the dossier's promise field is full and its mechanism field is thin, you don't have a copy problem yet. You have a homework problem, and better to see it now than three drafts deep.

The proof field has one rule that matters more than the rest: nothing invented, ever. This is where AI drafts go wrong on their own, reaching for a statistic when the offer doesn't have one.

The dossier fixes that at the source. If the proof is real, it goes in the field exactly as it is. If it isn't there, the field says [to confirm], and every downstream prompt inherits that honesty.

One field people skip entirely is the disliked alternatives, and it's the one that tells you what you're really selling against. The reader is always choosing between your offer and something else. The last thing that let them down, the workaround they've limped along with, or doing nothing for another quarter. Name those, and the copy can speak to the real competitor instead of an imaginary blank slate.

You don't write the dossier by hand. You extract it, from material that already exists, and then you fix what the extraction got wrong.

Gather the raw material first.

The reviews and support tickets matter most, because that's where the real objections and the disliked alternatives live, in the customer's own words instead of the marketing team's.

Then run this. One job: extract, don't sell.

The model hands you a dossier and a short list of its own weakest spots. Those weak spots are your next hour of work, and you'd rather have them named than discover them in a headline that fell apart under a client's question.

Two things the extraction reliably gets wrong, so you watch for them.

It over-trusts the marketing. Fed a glowing sales page, it will copy the page's own inflated proof straight into the proof field, so tell it to privilege the reviews and the tickets, and read that field with a skeptic's eye.

And it reaches for the loudest promise, which is not always the truest one. The promise that actually converts is sometimes the quiet thing a customer mentions in passing, not the line the headline shouts. That pick is the first thing you check by hand.

The extraction gets you most of the way. It does not get you all of the way, and the gap is the whole reason your judgment is still worth paying for.

Read what came back against what you know. The model will have picked a promise. Check that it picked the right one, because the strongest promise is sometimes buried in a review three screens down rather than in the headline the client wrote.

Fill the [to confirm] fields, or send them to the client as the exact questions you need answered. Cut proof that's technically true but soft. Sharpen the mechanism until a skeptic would nod.

When you're done, the dossier is a contract. Everything in it is real, sourced, and yours to stand behind. That's the thing you're about to hand to every prompt for the rest of the campaign.

A few finished fields from that course-launch dossier looked like this.

Notice how much of it is honest gaps. Two [to confirm] flags and one soft proof field. Those gaps are the dossier doing its job, showing me the exact three things to pull from the client before a word of copy depends on them.

From now on, no prompt for this campaign re-explains the offer. Each one loads the dossier and does one job against it, and that's where the tax finally disappears.

Writing a lead looks like this.

No audience description, no product explanation, no list of benefits, no proof. All of it lives in the dossier, so the prompt gets to be short and the output gets to be grounded.

The next prompt, for the email body, references the same dossier. So does the one for the ads, the one three weeks from now when the client asks for a fresh angle.

The objection section is its own prompt, and it pulls the same file.

Same dossier, different job, and neither prompt had to be told what the offer is.

Every prompt you write against the dossier is shorter than the one you'd have written from memory, and grounded in the same facts as the last one.

The whole campaign ends up reading like one writer built it in one sitting, even when it's three formats built over three weeks. Consistency used to take discipline. Now it takes a shared file.

One dossier, one job per prompt, every job pulling from the same sourced truth. That's the whole system.

The dossier is a plain markdown file. Where you keep it depends on how you work.

The mechanics change. The rule doesn't: one file per offer, referenced, never retyped.

Keep it next to the client brain, not inside it.

The brain is the client, persistent across every offer they'll ever run. The dossier is one campaign, and when the campaign ends you archive it.

Muddle the two and you're back to a single giant document that's vague about everything, which is the swamp the brain was built to drain.

The dossier is only as good as the material you feed the extraction. Thin source in, thin dossier out, which is why the model's list of weak fields is a feature and not a nag. It's telling you where reality is thin, and no prompt fixes a thin reality.

Budget the real time up front. Thirty minutes to gather the material, ten for the extraction and the read, another twenty on the phone or over email getting the [to confirm] answers the client alone can give you.

Call it an hour for a real campaign. You spend that hour once and stop spending fifteen minutes at the start of every chat for the next month.

And treat it as living. When a support ticket surfaces a new objection, or a testimonial comes in with a real number, the dossier gets updated, and every future prompt inherits the update for free. It's the one document in the campaign that gets more valuable the longer the campaign runs.

There's a bonus buried in the objections field. Written down and numbered, your objections are also a test backlog. Each one is a headline angle, an email, a subject-line split waiting to run.

When a test wins, you've learned which objection was really holding the reader back, and that finding goes back into the dossier for the next campaign. The brief slowly turns into a record of what this market actually believes.

One thing the dossier won't do is find the big idea for you. It organizes the truth about the offer.

Turning the sharpest fact in it into a campaign that stops the reader is still the creative call, the same one Ogilvy made when he picked the clock out of a stack of true and boring facts.

The dossier makes sure the fact is there and real. What you do with it stays your job.

The brief was always the bottleneck. We just kept rebuilding it from memory, one chat at a time, and calling that work.

Extract the offer once, into a document that never invents a number and never forgets the promise. Then let every prompt after it inherit a brief instead of your best guess at 8am.

More clicks, cash, and clients,
— Mark Masters