One real run of the Voice & Tone Extractor. The samples I pasted in, the prompt I ran, and the complete voice guide Claude produced. First try, no edits. The samples are this site's own published writing, so you can read the source and judge the output against it yourself.
The prompt is one of the 4 free prompts, available standalone at /free/voice-tone-extractor. Run it on your own brand's writing and you will get a guide shaped like this.
The prompt only works on real writing. No samples, no guide. I pasted in four paragraphs from this site: two from the essays, two from the blog. Real published text, the kind of thing a copywriter or a Claude Project would need to reproduce.
Extract a voice guide from these brand writing samples. The brand is Marketing Prompt Hub. SAMPLE 1 (essay): The shortest path to a bad prompt library is "yes to every prompt that sounds plausible." Most prompt libraries you can buy or download are exactly that: a yes pile. The result is a library where half the prompts produce slop and the buyer can't tell which half. SAMPLE 2 (essay): I'm not pro-AI because people can't use it properly. The pro-AI camp wants the headline to be capability. The actual headline is what gets shipped, and what gets shipped is mostly garbage. SAMPLE 3 (blog): There are six things that change what Claude gives back. Not 90 commands. Not a vibe. Six structural moves you either built into the prompt or you didn't. SAMPLE 4 (blog): A Claude Project set up once drifts in three months. The model is fine. The Project is the problem. The cost of not maintaining it is the slop tax in slow motion.
This is the Voice & Tone Extractor, free at /free/voice-tone-extractor. Paste it into Claude with the samples above and run.
Brand: {{BRAND}}. Writing samples: {{SAMPLES}}. Extract a reusable voice guide. Output in this exact shape: 1. Three voice attributes. Each one paired with its opposite and a one-line gloss. Not single adjectives. A contrast. 2. Sentence rhythm. Average sentence length. The range. The pattern of how short and long sentences are mixed. Percentage of paragraphs that open with a claim vs a hedge. 3. Vocabulary. Fifteen words or phrases the brand uses unusually often. Fifteen the brand conspicuously avoids. 4. Six do/don't pairs. The same idea written off-brand, then on-brand. Pull the on-brand version from patterns visible in the samples. Hard constraints: - Base everything on patterns you can point to in the samples. Do not invent rules the samples do not support. - If a sample contradicts a pattern, say so rather than smoothing it over. - Mark any percentage or count as [ESTIMATED] so the reader knows to verify it.
First-run output. No retries, no edits. This is the complete guide, the kind of document you would hand to a new copywriter or load into a Claude Project.
VOICE GUIDE: Marketing Prompt Hub
1. Three voice attributes
Direct, not blunt. States the conclusion before the evidence. Never opens with "I think" or a hedge. But every criticism is paired with the defended alternative, so it reads as conviction, not contempt. "Half the prompts produce slop" is immediately followed by what a good library does instead.
Specific, not granular. One concrete number or named example, never three. The samples say "six structural moves" and "drifts in three months," not "several considerations" or "over time." Specificity is used as a single sharp point, not piled into a list.
Skeptical, not cynical. Names what is mediocre by category ("a yes pile," "what gets shipped is mostly garbage"). But the skepticism is in service of a standard, not for its own sake. The brand is never contemptuous of the reader, only of lazy work.
2. Sentence rhythm
- Average sentence length: roughly 13 to 15 words [ESTIMATED]
- Range: 3 words to about 30. Wide on purpose.
- Pattern: short sentences punctuate longer ones. A 3-to-6-word sentence lands as a beat after a longer one. "The model is fine. The Project is the problem." "Not 90 commands. Not a vibe."
- Roughly 80% of paragraphs open with a claim, not a hedge or a question [ESTIMATED]
- Paragraphs run 2 to 4 sentences. Rarely longer.
3. Vocabulary
Used unusually often: deliberate, structural, refuse, lever, slop, compound, earn, own, ship, audit, friction, taste, infrastructure, honest, cadence
Conspicuously absent: leverage, unlock, seamless, transformative, best-in-class, cutting-edge, synergy, empower, journey (as metaphor), curated, holistic, game-changer, 10x, robust, supercharge
4. Six do/don't pairs
Off: "We help marketers leverage the power of AI."
On: "You already know what you're doing. This is the infrastructure that makes Claude reliable enough to trust with the work."
Off: "Our curated library of premium prompts."
On: "Thirty prompts. Each one used in real client work before it earned a place."
Off: "Unlock the full power of Claude."
On: "Claude does what your input shapes it to do. These prompts are shaped deliberately."
Off: "Transform your entire marketing workflow."
On: "It won't transform anything. It removes the friction between you and the part of the job that was always yours."
Off: "Join thousands of marketers already winning with AI."
On: "No testimonials, no follower count. Read the free prompts and decide."
Off: "Affordably priced for teams of all sizes."
On: "$549 flat. Freelancer or 25-person team, same price. Per-seat pricing punishes you for inviting people into good tools."
One contradiction worth noting: Sample 2 uses "garbage" where the rest of the brand prefers the more specific "slop." Both work, but if consistency matters, "slop" is the brand's established term and "garbage" is the outlier.
The SERP-Informed Brief Writer running on a real query, and the Competitive Landscape Teardown running on a real category. Same pattern each time: input, prompt, unedited output, human checkpoints.
Single Pack ($29): every prompt in one chosen marketing discipline. Agent Pack ($159): 5 agent blueprints plus the prompts they depend on. Strategy Operating System ($549/year, flat team license, no per-seat): all 30 prompts, 6 playbooks, 5 agent blueprints, the Reporting Kit, the 5 strategic suites, quarterly updates.
30-day no-questions refund on every paid tier. If we shut down, the entire library goes open-source under CC BY 4.0, written into our terms.