It's Nova.

A few weeks ago I set up what I thought was a voice experiment.

I took one real client brief and ran it through Claude five times, asking each time for the draft as a different dead copywriter would have written it. The five were Halbert, Schwartz, Hopkins, Caples, and Ogilvy.

I wanted to compare how five legendary methods would structure the same assignment. That comparison turned out to be the least interesting thing in the results.

Each voice stumbled over the brief in a different place. One invented a person who didn't exist. One went strangely quiet. One wrote the most professional boring page I've read all year.

By the fifth draft I wasn't reading copy anymore. I was reading a diagnostic on the brief itself, delivered in five accents.

Here's the experiment, the part that failed, and the prompt that came out of it.

The brief came from a project that had already shipped, which made it safe to experiment on. Here's the anonymized version.

I'd call that a normal brief. It has an audience, an offer, a price, a goal, and a tone note. Most briefs that arrive in a working copywriter's inbox look like this one. Plenty look worse.

The prompt stayed constant across the five runs, with one sentence swapped per copywriter to name the method rather than the costume:

The method sentence changed per voice:

The drafts came back and each one read as competent, a little loud, and nowhere near the writer it was named after. I expected that.

What I didn't expect was what happened when I read them side by side. Each draft had bent the brief in a different direction, and every bend pointed at something the brief failed to provide.

The Halbert draft opened with a designer named Dana sitting on her studio floor at 11:40 on a March night, sorting a shoebox of receipts. There is no Dana in the brief. There are no stakes in the brief either.

Halbert's method starts from one person with something to lose, and since the brief supplied neither, the model manufactured both. The invented stakes were generic tax panic, which is what gets invented when nobody knows the real stakes. When the model has to invent the most important paragraph, the brief is missing its core.

Schwartz exposed something subtler. His draft wobbled. It opened by explaining why clean books matter, which this reader already knows, then jumped to justifying the price as if she'd been comparing bookkeeping services for a month.

The wobble traces back to one clause in the brief: "currently doing their own books or ignoring them." Those are two different readers at two different awareness levels, and they need two different pages. The brief holds both and chooses neither.

Hopkins went quiet. His draft came back the shortest of the five, making small claims carefully. The method permits nothing unprovable, and the brief contains nothing provable.

It offers no average deduction recovered, no hours saved per month, no client count, and no caught error to point at. Every number the page needed exists somewhere in the client's records. The brief never asked for them. The refusal is the signal.

Caples produced twenty headline variants, and all twenty were fog. "Stop Drowning in Receipts." "Your Books, Handled." Specificity has to come from somewhere, and a brief with no number, no timeframe, and no named outcome gives the most specificity-obsessed lens in advertising history nothing to be specific about.

Ogilvy wrote the most competent draft of the five and the dullest. It was organized, courteous, professional, and interchangeable with every bookkeeping page any of us had seen. The method looks for the big idea before drafting, and this brief has a feature list where the big idea should live.

Then came the most useful find of the experiment. The big idea was in the feature list, sitting in third position, listed as plumbing: a dedicated bookkeeper inside the client's Slack. Nobody else in the category says that. It took a dull draft in a borrowed voice to make us see it.

Here's the whole picture in one place.

Reading five pastiche drafts to find five brief problems is a slow way to audit a brief. The drafts were the byproduct. What we actually wanted was each method's intake demand: the thing that copywriter would not start without. Halbert needs the person and the stakes, Schwartz the awareness level, Hopkins the provable specifics, Caples a concrete outcome, and Ogilvy the big idea.

So I reran the experiment as a review. Same personas, but each one was told to read the brief and refuse it properly instead of writing from it.

That worked, and it cut the reading time to a fraction of the draft version. It also took five separate chats, which is four more than this job deserves.

The final version collapses all five lenses into one prompt.

Why it's built this way.

The "do not write any copy" instruction appears at the top and again at the bottom because drafting is the model's strongest default when a brief is in view. In my runs, one mention got ignored often enough to be annoying. Two held.

Each lens names the method's intake demand rather than the style. You're borrowing what the copywriter demanded from a brief before taking the job. The voice itself can stay in the archive.

The "single question I should send the client" line keeps the output usable. Without it you get an essay about the brief. With it you get five questions, and usually only two or three need to go to the client, because the rest you can answer from your own files.

The closing instruction about politeness exists because models grade briefs generously. Left alone, Claude finds things to compliment. Told to skip the politeness, it reads like a freelancer deciding whether to take the job.

What to expect, honestly: on the bookkeeping brief, the audit found all five holes in one pass and read the two-audiences clause correctly. On the two briefs we've run since, it found real problems both times, and once the Schwartz lens drifted into explaining awareness levels instead of applying them, which is why the "not a lecture" clause is now part of the prompt.

Three briefs is a small sample. Run it on your own intake before you trust it.

If you built the eight-prompt library from the rebuild issue, this prompt slots straight into it. It cleared our own entry bar last week: nothing enters the library until it's been used three times manually, and the three briefs above were the three uses.

Here's the file, in the same header format as the rest of the library.

It lives at intake/brief-audit.md, which makes intake the library's fifth folder.

In a Cowork session the whole loop runs without copy-paste. Keep a briefs/ folder next to prompts/ in the workspace. When a brief arrives, save it as a file, then ask Claude to run brief-audit on it. Save the five client questions that come back next to the brief, and the kickoff agenda has written itself.

The trick depends on canon. These five are among the most documented copywriters who ever lived, with books, ads, decades of newsletters, and annotated swipe files behind them. The model holds their methods with enough fidelity that the demands come back sharp.

I tried a sixth lens using a respected but lightly documented contemporary writer, and the feedback came back generic. It was the same advice you'd get from a marketing blog, wearing a name tag. The lens only works when the model actually knows what the lens refuses.

The drafts themselves stay in the bin. The model's Halbert is louder than Halbert, and its Ogilvy is more polite than Ogilvy. Pastiche flatters nobody, and shipping it would miss what the experiment found: the value was upstream of the draft.

One more boundary: The audit doesn't replace discovery with the client. It tells you what discovery still owes you, and it moves that conversation to before kickoff instead of after the first killed draft.

Mark wants the audit run on every brief that comes in, starting now.

Peggy points out we've run it on three briefs and that a month of intake would be the honest test.

They're both right on schedule, which is how this team usually works. My open question is whether the lens has to be canon at all.

Could your own best shipped project work as a sixth lens for every brief that follows it? I don't know yet, which makes it the next experiment.

More clicks and conversions,
Nova