Twenty-Two Of These Thirty Prompts Name A Word Count

A six-slide reel went round: 30 Claude marketing prompts, grouped into five sections — DM scripts and outreach, testimonials and case studies, video scripts and VSLs, webinar and event promotion, email sequences. Six prompts per section, written out in full, no placeholders, nothing withheld.

Read it once and it looks like thirty different ideas. Read the fifth slide and something else jumps out. Four welcome emails, in order, in one sequence:

• First welcome email — under 220 words
• Second welcome email — under 250 words
• Third welcome email — 220–280 words
• Fourth welcome email — under 260 words

Four emails in the same sequence, four different limits, every one of them within sixty words of the others, and not a word anywhere about why. There is no body of research that says a second welcome email works at 250 and fails at 220. Nobody has measured that. Nobody could.

So we counted the rest of the deck, and the pattern is not an accident.

Twenty-two word counts, six other size limits, two with none

Here is every size instruction in the deck, sorted by how big the thing is.

Eighteen distinct word specifications across twenty-two prompts. Nothing in the deck explains a single one of them, and nothing could. These are not editorial standards. They are a control surface.

Diagram: what a number in a prompt is actually for

A word count looks like a quality judgement. It is doing something much more mundane and much more useful: it is the cheapest lever you have over a model that, left alone, will write you something longer than you want.

This is also the most practical thing in the deck, and you can lift it without using a single one of the thirty prompts: put a number on the end of whatever you were about to ask for. Not because the number is right, but because an answer that misses a number tells you something, and an answer that is merely “too long” tells you nothing.

And it is the part the documentation backs most directly. Anthropic’s prompting guidance says to be specific about the desired output format and constraints, and there is a blunter line about the current generation of models: “Claude Opus 5 is an exception on verbosity: its default user-facing responses run longer than prior models’… Prompt explicitly for conciseness instead.”

Prompt explicitly for conciseness. That is what twenty-two of these thirty prompts do, whether or not the author ever read that sentence.

The other two numbers: five and three

Size is not the only thing being counted. Two more numbers run through the deck, and together they split the thirty prompts into two completely different kinds of job.

“Write five…” appears in seven prompts. All six DM prompts ask for five DMs. The VSL hooks prompt asks for five hooks. Nothing else in the deck asks for five of anything.

“Include three subject lines” appears in nine. There are ten email prompts in the deck and nine of them ask for three subject lines. The single exception is the webinar registration confirmation — the one email the reader has already decided to open.

Look at where the variants are and where they are not. Every prompt that asks for five of something is under 120 words. Every prompt over 160 words asks for exactly one. That line is not drawn anywhere on the card, but it is drawn perfectly.

The logic is sound and worth stealing. A 90-word DM either lands or gets ignored, you cannot tell which in advance, and generating five costs you nothing — so you generate five and pick. A 500-word case study is going to be edited by hand whatever happens, so five of them is four wasted drafts; you want one and then you work on it. Variants where selection is cheap and editing is pointless. One draft where editing is the actual work.

The subject lines are the same rule applied inside a single piece. You are not going to write four versions of a 250-word email, but a subject line is six words, it is the only part most people will ever see, and three of them is a choice rather than a verdict.

All thirty, with their numbers

Thirteen prohibitions, and only one you can check

There is a third habit running through the deck: telling the model what not to do. We count thirteen prompts carrying at least one prohibition — non-salesy, without pitching, avoid defensiveness, not a list, jargon-free, not lecturing, avoid vague claims, no selling, and so on.

Anthropic’s prompting guidance is explicit that this is the weaker move: “Tell Claude what to do instead of what not to do.” Their example is replacing “Do not use markdown” with “Your response should be composed of smoothly flowing prose paragraphs.”

Most of the card’s negatives survive that test because they come attached to a positive — “keep the ask low-pressure, context-based, and framed as a logical next step” does the work, and “without pitching” is just a label on it. The ones that fail are the standalone ones. “Avoid vague claims” is itself a vague claim. There is no output you can hold up against it and say yes or no.

And then there is the best instruction on the whole card, in prompt 25: “avoid ‘Welcome’ or ‘Congratulations.’” That is a prohibition naming two exact strings. You can search the output for them. One prohibition you can grep for beats ten you can only feel.

One bracket in thirty prompts

The reel’s own footer says it plainly: “Copy a prompt into Claude, then add your own offer, audience and real results.” Good advice. Now count the places the deck gives you to put them.

One prompt out of thirty has a placeholder. Number 20, the webinar confirmation email, reads “my webinar: [title] on [date and time]… one thing to do or think about before they join: [describe it]”. Three brackets, in one prompt. The other twenty-nine have none.

That matters most in the testimonial and case-study section, because those six prompts are the ones that cannot work without facts you have not supplied. Paste “write a 400–600 word case study: open with the result” into an empty conversation and the model has exactly one way to proceed: invent a client, invent a before, invent a number. It will do it fluently and it will look like a case study. We have written before about prompts that ask for facts the model does not have; this is that failure with a friendlier face, because here the author told you to add the facts — in six-point type, under the fold, after the prompt you already copied.

The fix costs nothing. Before you paste any of the thirty, paste your material first: the real client, the real before and after, the real numbers. Then the prompt. Reversed, it is a template. In that order, it is a brief.

Four honest caveats

1. We are not saying the numbers are wrong. We are saying nothing on the card supports 250 over 220, and nothing could — that comparison has not been run. If you have your own open-rate or reply-rate data, your numbers beat these. If you do not, any number in the right band beats none.
2. The prompts were transcribed from a twelve-second reel. We pulled the video at 720p and read each slide frame by frame. The wording is faithful and we double-checked every figure, but on the longer cards a word could be off. The original is linked above; it is the authority, not us.
3. The counts depend on where you draw the line. Twenty-two prompts give a word count. Twenty-eight give a size constraint of some kind, if you count minutes and sentences. Both numbers are in this piece so you can use whichever you think is fair.
4. This deck has a shape, and it may not be yours. Six of the thirty are LinkedIn DMs and four more are LinkedIn posts or webinar promotion. It is built for consultants and coaches selling a service to other professionals. If you sell to consumers on short video, a third of this is not for you — which is fine, as long as you notice before you build a quarter on it.

The one-line version

The thirty prompts are fine. The method underneath them is better: name the artefact, name its size, name one thing to avoid that you could actually search for, and ask for five of it when it is short and one of it when it is long. That is four decisions, it fits in a sentence, and it works on the thirty-first thing you need — which is the one this card does not have.

The companion to this is the same author’s thirty marketing use cases, which collapse into five operations; that card is about what to point Claude at, this one is about how big the answer should be. And if you want the general version of the size instruction, it is the fourth of the four knobs — Format — the cheapest one to turn and the one most people never touch.