Loading tutorials…
Loading tutorials…
Brand Voice is the single feature that makes Jasper worth paying for versus raw LLMs. Trained well, it sounds like you. Trained badly, you sound like every other Jasper user.
Who this is forMarketing teams who care that their copy has a recognizable voice. Founders who have spent years developing a tone they refuse to dilute. Agencies maintaining distinct voices across clients.
What you'll need
Step 1
Brand Voice trains from samples. Mediocre samples produce mediocre voice. Use only your best.
Pick 5-10 pieces of finished, published copy you would happily put on your portfolio.
Mix formats: 1-2 blog posts, 1-2 long-form emails, 1-2 ads, 1-2 landing pages, 1 social caption.
Avoid: anything ghostwritten by an agency you would not hire again, AI-generated content, or copy that did not actually work commercially.
Length: each sample 200+ words. Shorter samples are statistically too thin to extract voice patterns.
Step 2
In Jasper → Brand Voice → New. Name it after the brand, not the project.
Name: use the company name or product name. "ACME Co." not "Q3 campaign voice."
Description: 2-3 sentences describing the brand in your own words. This shows in dropdowns and helps teammates pick the right voice.
Upload the samples one at a time, or paste them in. Jasper supports files and direct paste.
After each upload, Jasper will extract voice characteristics. Read these — they are your first signal of whether the training is converging.
Step 3
Tone fields control output more than samples do for the first 30 days. Be specific.
Voice characteristics: 3-5 adjectives. "Confident, skeptical, direct" beats "Professional, helpful, engaging."
Style notes: 1-2 sentences about formatting preferences. "Short paragraphs. Avoids subheadings in pieces under 800 words. Uses em dashes for emphasis."
Words to avoid: paste a list. Industry filler is the obvious one. Also include words your brand simply does not use ("game-changer," "world-class," "unlock," "drive results").
Words to use: 5-10 phrases your brand owns. These get reinforced in output.
Step 4
Brand Voice without a reader produces voice with no target. Define who you write for in concrete terms.
Audience name: 2-3 word descriptor. "Bootstrapped SaaS founders" beats "Business decision-makers."
Pain points: 3-5 specific pains. "Burning through cash on agency retainers that produce generic content" beats "Wants better marketing."
Goals: what they want to achieve in their words, not yours.
Vocabulary: what jargon do they use? What do they reject? "Talks about ARR and CAC, hates the word 'leverage.'"
Step 5
Generate 10-20 pieces in the first week. Tag the ones that miss with notes. This is what trains the voice past the initial baseline.
Pick 10-20 prompts representing your real use cases: blog intros, ad headlines, email subject lines, landing page H1s.
Generate each with the Brand Voice applied.
For each output, score: on-brand, off-brand, or close. For off-brand, write 1-2 sentences explaining what is wrong.
Paste those notes into the Brand Voice → Notes field. Jasper uses these as ongoing context.
Re-generate the off-brand ones after each note batch. Track whether the second pass is closer.
Step 6
Brand Voice drifts when your brand evolves. A 15-minute monthly review keeps it tight.
Once a month, generate 5 pieces and grade them blind. Score 1-5 on brand-voice accuracy.
If average drops below 4, add 1-2 new samples reflecting recent brand evolution.
Update the Forbidden Words list with anything you have caught Jasper repeating.
Quarterly: refresh the audience profile. Your audience evolves; the model should reflect that.
Common mistakes
Training on ghostwritten or AI-generated samples
What goes wrong: You train Jasper on copy that already lacked voice. Output becomes a copy of a copy. Genericness compounds. ~40-80 hours of rework eventually required.
How to avoid: Audit samples ruthlessly. If the original was written by an LLM or a freelancer you would not rehire, do not feed it back into the system.
Vague tone descriptors
What goes wrong: Voice characteristics like "Professional, friendly, engaging" describe 90% of business writing. Jasper has no constraint and produces template English.
How to avoid: Use sharp, constraining adjectives. "Skeptical, direct, occasionally dry" produces meaningfully different output than "Professional, helpful, engaging."
Skipping the Words-to-Avoid list
What goes wrong: Output uses your industry's filler words on autopilot. You edit out the same 8 words from every generation. 50 generations/month at 2 minutes of cleanup each = 100 hours/year.
How to avoid: Build the Forbidden Words list from real edits. Every time you cut a word, add it to the list. Compounds in your favor over time.
Treating Brand Voice as one-and-done
What goes wrong: Trained once at setup, never touched again. Brand evolves. Voice does not. 6 months later, output feels dated and you blame the tool.
How to avoid: Calendar reminder: 15 minutes monthly to grade outputs and add notes. 60 minutes quarterly to refresh samples.
Mixing two brands in one Brand Voice
What goes wrong: Agency uses one Brand Voice for two clients to save a seat. Output averages the two brands — sounds like neither.
How to avoid: One Brand Voice per brand. Upgrade the plan if you need more slots. The economics of doing it right are obvious past two clients.
Recap
Done — what's next
How to set up your Jasper AI account the right way
Read the next tutorial
Hand it off
Brand Voice training is high-leverage and patience-heavy — a perfect handoff to a content specialist who has done it before. EverestX content specialists train Brand Voice in 1-2 weeks and maintain it ongoing for $400-1,200/mo.
See specialist rates
Five high-quality samples is the floor. Past 10, returns diminish — you spend more time uploading than you gain in voice accuracy. Quality of each sample matters far more than quantity.
No. Brand Voice should reflect *your* brand, not a brand you wish you were. Competitor copy as inspiration is fine; competitor copy as training data produces output that sounds like them, which is worse than sounding generic.
ChatGPT custom instructions are a single text field applied to every chat. Jasper Brand Voice is a multi-field profile (samples + descriptors + audience + notes) that conditions every generation more granularly. For one-off content, ChatGPT is fine. For consistent brand-voiced output across a team, Jasper Brand Voice is a meaningful step up.
Three usual culprits: (1) samples were too short or off-brand, (2) tone descriptors were too vague, (3) audience profile was missing. Walk through the steps above with fresh eyes. If it is still generic, the issue is usually the samples — replace them with your strongest work.
Jasper AI
Jasper is powerful but unforgiving on first setup — the wrong plan or workspace shape costs you brand-voice quality for months. This is the setup specialists run.
Jasper AI
Jasper's template library is 60% gold, 40% filler. This is how specialists separate them, plus the custom-template patterns that save 10 hours/week.
Jasper AI
DIY Jasper works for a stretch. Then volume, brand voice, and editing time hit a ceiling. This is the framework for when a specialist actually earns their fee.