Short version: set a low formal tone, keep things simple but not dumbed down, be brief, and allow a healthy dose of imagination. That tiny shift—20% formality, 30% complexity, 70% brevity, 60% creativity—can flip how your Rephrase.ai review reads and how your audience responds. You’ll get a practical, slightly skeptical walkthrough here: what to do, what to avoid, advanced experiments, and how to fix things when the AI goes off the rails.
1. What you’ll learn (objectives)
- Exactly what those percentage settings mean in practice and how they translate into prompts and controls. How to configure Rephrase.ai (or similar rephrasing tools) to consistently produce content with the target profile. Testing and evaluation methods so you can measure whether the result matches your brand intent. Advanced tricks for combining these settings with prompt engineering and post-editing to save time and increase impact. Concrete troubleshooting steps for the most common output problems marketers encounter.
2. Prerequisites and preparation
- Access to Rephrase.ai or a comparable text-rewriting AI with configurable style settings (formality, complexity, brevity, creativity/temperature). A baseline piece of content: your original review, a product description, or a draft email you want rewritten. A simple rubric for evaluation: voice match, clarity, length, and conversion intent. Treat this like QA, not aesthetics. Basic editing tools (Google Docs, Word, or your CMS) and a stopwatch or workflow timer. You’ll iterate, and time matters. Patience for dealing with glib outputs—AI is fast but not always wise. Your job is to guide it.
3. Step-by-step instructions
Step 1 — Translate percentages into tool settings
Most tools don’t accept “20% formality” as a literal input. Convert each percent into concrete instructions:
- 20% formality → “Keep tone casual, avoid formal phrasing and jargon, do not use legalese.” 30% complexity → “Use simple sentence structures; occasional multi-clause sentences permitted but minimize dense concepts.” 70% brevity → “Prefer concise sentences; aim for short paragraphs and cut filler. Target 50–70% length of original.” 60% creativity → “Allow creative metaphors and slight inventiveness in phrasing, but stay truthful and on-brand.”
Step 2 — Build the prompt
Combine those instructions into a succinct prompt. Example prompt:
"Rewrite the following review with a casual voice (20% formality), low complexity (30%), high brevity (70% shorter than original), and moderate creativity (60%). Use short sentences, keep it punchy, include one clever metaphor, and retain all factual claims."
Don’t be shy about adding constraints: mention the desired word count or sentence limit, call out words to avoid, and state the target audience (marketers, in this case).
Step 3 — Run a controlled batch
Generate multiple variants (3–5) from the same prompt. Save them for comparison. Variants help you see how consistently the AI follows the profile and which tweaks produce better outputs.
Step 4 — Evaluate against your rubric
Score each variant on a simple scale for voice match, clarity, brevity accuracy, and creativity fit. Example rubric:
Criteria 0–3 (Fail to Perfect) Casualness Is the text relaxed, conversational, and not overly formal? Complexity Are sentences and ideas easy to parse (no dense jargon)? Brevity Is the output significantly shorter without losing meaning? Creativity Does it have tasteful originality without inventing facts?Step 5 — Iterate and refine
If outputs miss the mark, tweak one variable at a time: increase creativity to 70% if the copy feels staid, or reduce it to 50% if the AI invents details. Track changes so you know which tweak fixed which issue.
Step 6 — Post-edit strategically
Don’t try to fix everything in the AI prompt. Post-editing is cheaper and faster for marketers. Focus edits on CTA strength, brand word choices, and factual accuracy. Use the AI output as a scaffold, not the final product—unless it’s already perfect.
4. Common pitfalls to avoid
- Vague prompts: Saying “make it casual” yields inconsistent results. Convert vague instructions into explicit dos and don’ts. Over-constraining: Telling the AI to be 70% shorter and also to keep every nuance will produce either flabby or hollow text. Decide which matter most—brevity or nuance—and prioritize it. Assuming creativity means accuracy: Higher creativity can invent plausible but false details. Never skip fact-checking. Single-pass complacency: Accepting the first decent output wastes potential. Always generate variants and choose the best. Ignoring audience fit: Casual tone for technical buyer personas can erode trust. Adjust percentages by audience segment.
5. Advanced tips and variations
Tip: Map percentages to model parameters
If your tool exposes parameters (temperature, top-p, max tokens), map the percentages: set temperature to 0.6 for 60% creativity, reduce max tokens by roughly 30% to enforce 70% brevity, and limit specialized vocabulary to enforce 30% complexity. These translations aren’t perfect, but they provide a reproducible starting point.
Tip: Persona layering
Create a persona cheat-sheet: "Marketer Mike — witty, business-first, skeptical." Layer that on top of the percentage brief to anchor style. Personas reduce variation between generated pieces.
Tip: Controlled creativity
Ask the model for "one tasteful metaphor" or "a single bold line" rather than leaving creativity unbounded. This preserves brand safety while giving the copy personality.
Tip: Use A/B splits
A/B test two profiles: the 20/30/70/60 profile vs. a baseline (e.g., 40/50/40/30). Measure open rates, click-throughs, or time-on-page to quantify which voice works for your audience. You’ll learn faster than you think.
Tip: Progressive constraints
Start loose, then refine. First pass: ask for casual and creative. Second pass: ask for shorter and simpler. This staged approach often yields better results than squeezing everything into one prompt.
Thought Experiment 1 — The Skeptical CMO
Imagine a CMO who hates fluff and values speed. If you set 70% brevity and 60% creativity, will the CMO forgive a clever line if it risks clarity? Walk through three scenarios: the copy increases conversions by 8% (accept); it’s witty but confuses users (reject); it’s short and clear but bland (acceptable). This helps you decide which metric the organization prioritizes and adjust percentages accordingly.
Thought Experiment 2 — The Technical Audience
Now imagine a technical audience. Lower the creativity to 40% and bump complexity to 50% while keeping brevity at 60%. Run a mental simulation: does the tradeoff maintain credibility? If not, increase complexity further and reduce brevity. Repeat until the mental metrics align with your stakeholders’ expectations.
6. Troubleshooting guide
Problem: Output is too formal
Symptoms: long sentences, passive voice, business-speak. Fixes:
- Add explicit commands to the prompt: "Use contractions (don't, can't), short sentences, second-person address (you), and active voice." Provide examples of the desired casual tone in the prompt. Regenerate and compare variants focused solely on tone.
Problem: Output is too terse and loses meaning
Symptoms: sentences are choppy, important points removed. Fixes:
- Relax brevity from 70% to 50% for pieces that require nuance. Ask the model to "keep these three facts intact" and list them explicitly. Use a two-step output: a short version followed by a 1–2 sentence explanation of the key points.
Problem: AI invents details or claims
Symptoms: unsupported features or fake stats appear. Fixes:
- Reduce creativity parameter (e.g., from 0.6 to 0.4) and re-run. Prompt the model with "Do not add new facts; only rephrase provided content." Always run a factual verification pass. If budget allows, automate fact-checking via a knowledge base lookup.
Problem: Tone varies between outputs
Symptoms: each variant feels like a different writer. Fixes:
- Create a short persona and include it in every prompt. Use the best variant as a style exemplar and ask the AI to match it. Lock down seed settings or parameters if your tool supports reproducibility.
Problem: Output is creative but off-brand
Symptoms: metaphors or jokes that don’t fit brand voice. Fixes:
- Provide a list of "on-brand" and "off-brand" examples to the model. Limit creativity to a single line or sentence and reserve the rest for straightforward messaging. Implement a short editorial pre-flight checklist to catch off-brand moments before publication.
Final checklist before publishing
- Does the text read casual but professional? (20% formality) Is the sentence structure simple and clear? (30% complexity) Is the copy concise, and does every sentence justify its existence? (70% brevity) Does creativity add personality without inventing facts? (60% creativity) Have you tested variants and picked the best performer for your target audience?
Wrap-up: Don’t treat those percentages as dogma—they’re a starting framework. Think of newsbreak them as a quick personality brief you give the AI. Then iterate: prompt, compare, tweak, and measure. This approach saved me time and made reviews readable and useful for marketers who don’t want to wade through corporate napkin-speak. It’s refreshingly practical, and if you’re slightly cynical about AI marketing copy (good), this method will keep you in control while letting the machine do the heavy lifting.
Next step: pick one piece of content, apply the 20/30/70/60 brief, generate five variants, and run a one-week micro-A/B test. If nothing else, you’ll get a clearer sense of what “moderately creative and very concise” actually looks like in your market—and that alone is worth the experiment.