Tone is the voice fingerprint.

Tone controls how formal, how direct, and how much personality the output carries. It isn’t a slider between boring and flashy — it’s the specific way your company sounds in writing.

Describe tone the way you’d brief a new hire: which words you use, which you avoid, what a good sentence from your brand sounds like. Vague tone descriptions produce vague content.

The ten tones Voice ships with

Professional

Polished and precise. Neutral, business-appropriate, no slang. Default for most B2B content.

Casual & Friendly

Warm, approachable, contractions allowed. Reads like a helpful colleague — good for newsletters and community posts.

Authoritative

Confident, declarative, fact-led. Backs claims with data and reasoning. Best for thought leadership and reports.

Persuasive

Action-oriented, benefit-driven, clear CTAs. Built for landing pages, sales emails, and offer announcements.

Inspirational

Aspirational and uplifting. Vision-led storytelling that paints a bigger picture. Use for mission, manifestos, brand pieces.

Witty

Playful, clever, sometimes cheeky. Light humor and unexpected turns of phrase. Great for X threads and social.

Conversational

Speaks directly to the reader (“you”). Short sentences, natural rhythm, like a podcast transcript. Good for blog intros.

Empathetic

Understanding, validating, human. Acknowledges pain points before offering answers. Ideal for support, HR, healthcare.

Bold

Punchy, opinionated, takes a stance. Short sentences, strong verbs, no hedging. Use to stand out in a crowded feed.

Analytical

Methodical and structured. Numbers, frameworks, “here’s why” reasoning. Best for whitepapers, case studies, deep dives.

Language is native, not translated.

Voice writes in the generation language you pick. It isn’t translating from English afterwards, so idioms and rhythm come out right for the language you publish in.

Length presets exist per format for a reason.

A 900-word blog post and a 900-word LinkedIn post are two different failures. Longer isn’t better on a feed; shorter isn’t better in a briefing. The defaults are sane. Override them when you know your audience.

Press release boilerplate is set-and-forget.

The standing paragraph at the bottom of every release. Set the “About” copy once and Voice appends it to every press release without you thinking about it.

The mix Voice defaults to: 40 / 30 / 20 / 10.

Roughly 40% educational, 30% point-of-view, 20% social proof, 10% promotional. That’s our position, not a peer-reviewed study — it’s the shape we recommend based on what actually distributes on LinkedIn today.

The exact percentages matter less than the shape: most of your output earns attention before it asks for any. The 10% ceiling on promotional content is the one hard rule worth keeping. LinkedIn’s algorithm actively throttles posts it reads as sales-focused, and one in ten is the line that stays under it.

• Educational teaches something useful with no ask attached.
• Point-of-view takes a position your competitors would hedge on.
• Social proof is customers, numbers, receipts.
• Promotional is the launch, the offer, the ask.

Run the mix for a quarter before tuning it. Shifting percentages every two weeks produces no signal.

Train Voice with every draft.

The thumbs up/down on each generated draft is the feedback loop. Thumbs up means “this is us” — tone, phrasing, and structure similar to this become more frequent. Thumbs down means “not us” — the specific choices get avoided next time.

Rate consistently for a few weeks and Voice starts producing drafts that need less editing. Skip the rating and you’re stuck training with one-off prompt changes.

For line-level corrections (not the whole draft), use phrase flagging — the next section.

Train Voice in your language, one phrase at a time.

Voice writes in 32 languages, but every language has its own bad habits — the AI-sounding phrases, the translated idioms that don’t exist in real speech, the words that read fine in English but stiff in Swedish, German, or Portuguese. The thumbs up/down on a whole draft is too blunt to fix that. The phrase flagger is the precision tool: you highlight the words you don’t want to hear again, Voice stops producing them.

How it works

• Highlight any phrase in the Voice editor (4–512 characters). A small orange Flag phrase button appears next to the selection.
• Click it and pick a reason: Sounds translated, Sounds AI-generated, Wrong word for this context, or Wrong tone. Add an optional note to remind yourself why.
• From your next generation onwards, Voice is told to avoid that exact phrase in that exact language. The block is private to you and tied to the language you flagged in — flagging a Swedish phrase doesn’t affect your English drafts.
• Free. Flagging a phrase doesn’t cost a task. You can flag as many as you want.

The community loop

When five different users independently flag the exact same phrase in the same language, Adamiro adds it to a community list and starts blocking it for everyone writing in that language. No admin approval — the threshold is the signal. If the consensus changes (users un-flag the phrase and the count drops below five), the block lifts automatically.

This means every language gets quietly better over time without anyone having to centrally curate it — the people who actually write Norwegian, Polish, or Japanese are the ones teaching the system what doesn’t sound right.

The Flagged phrases section in Settings lists everything you’ve flagged. Click the × on any row to un-flag.

“Rephrase to pass content policy” — what just happened?

When you generate a cover image for a blog draft (the image picker’s Generate tab), the image generator runs every prompt through a content-policy check before drawing. If the prompt mentions a real named person, weapons, sexual or explicit material, copyrighted characters or brand logos, or specific real entities, the check rejects it — even if the rest of the prompt is fine.

When that happens, Adamiro detects the rejection automatically and rewrites the prompt into a variant that passes the policy while keeping the same visual theme. You’ll see a banner with a Rephrase to pass content policy button — clicking it drops the rewrite into the prompt field. Review, tweak if you want, and click Generate again.

• It’s free. The rephrase doesn’t use the image generator — no task is charged. If a task was already consumed on the blocked attempt, it’s refunded.
• It keeps your idea. The rewrite preserves mood, palette, composition. It strips only the parts the policy flagged.
• It’s a suggestion, not a verdict. If the rephrase isn’t what you want, edit the prompt by hand and try again.

Common triggers worth avoiding upfront: named real people (“a photo of Elon Musk”), brand logos, weapons, explicit content, copyrighted characters (Marvel, Disney, Pokémon, etc.).

Image-attempt limits — when Adamiro slows you down

The cover-image picker in the Voice editor is a fast loop, which makes it easy to burn tasks fine-tuning mood when editing the prompt substantially or picking from your library would get you there sooner. Two guardrails protect the task wallet:

• Cost-warning toast at 5+ attempts. Once you’ve generated 5 cover-image variants in the same session, a one-shot toast reminds you that each Regenerate costs 1 task. Not a hard stop — just a louder version of the per-attempt counter.
• Hard cap at 20 attempts per session. Adamiro refuses the 21st gen for the same picker session. You clear the cap by picking from your media library (free) or by closing the picker and opening it fresh on a new prompt. The cap exists so a runaway click-fest can’t drain the wallet.

The picker session lasts 30 minutes — if you close the picker and re-open it later in the same draft, you start a fresh attempt counter.