Tone of Voice¶

Phase: 5 — Brand Project: likeness Date: 2026-05-09 Confidence: High — voice is well-anchored by existing repo materials (founder brief, mockup, CLAUDE.md) and by research into creator-side language preferences.

Brand personality traits¶

Four adjectives, each defined so they're operationally useful rather than vibe-only:

1. Direct¶

We say what we mean. We don't bury hard parts in qualifiers. If revocation can't recover downloaded files, we say so the first time the topic comes up — not in a footnote on page 14.

2. Adult¶

We treat the reader as an adult who has read other things and made their own judgments. We don't explain things creators already know. We don't pretend creators need protection from information about their own platform.

3. Specific¶

We use the actual words: license, revocation, watermark, processor, 2257. We don't reach for "innovative AI-powered" when "license-gated inference" is what we mean. Specificity is respect.

4. Plain (not folksy, not bureaucratic)¶

Short sentences, ordinary words, working clauses. Not the poetic startup register ("we believe..."), not the corporate register ("Likeness is a leader in..."), not the social-media register ("✨ excited to share ✨"). Just plain working English.

These four work together. "Direct" without "Specific" becomes blunt. "Adult" without "Plain" becomes stuffy. "Plain" without "Direct" becomes empty.

Voice principles — we are / we are not¶

We are direct. We are not blunt.¶

Direct (good)	Blunt (bad)	Soft (bad)
"Revocation can't recover files that were already downloaded. We can prove they came from your account; we can pursue takedowns; we can't make them disappear."	"If your stuff leaks, that's on you."	"While we strive for comprehensive content control, certain limitations may apply to previously distributed materials."

We are honest about hard parts. We are not apocalyptic about them.¶

Honest (good)	Apocalyptic (bad)	Evasive (bad)
"Adult-friendly payment processors are the single most likely cause of needing a bridge round before seed. We carry redundancy across multiple processors and a reserve to absorb a single bad month."	"Payment processors could end this company any day."	"We have a plan in place to mitigate any payment-related challenges that may arise."

We use the language creators already use. We don't invent jargon.¶

Right (creator-aligned)	Wrong (jargon)	Wrong (cute)
"License"	"Consent token"	"AI mirror"
"Revoke"	"Deprovision"	"Take it back"
"Approval queue"	"Submission DAG"	"Yes pile"
"Your model never leaves the platform"	"Hermetic adapter isolation"	"We never let your AI escape"
"Watermark"	"Provenance manifest"	"Secret signature"

We say "no" plainly when something is out.¶

Plain no (good)	Hedge no (bad)	Disguise no (bad)
"We don't allow third-party reference uploads. Every face on the platform is verified."	"We're cautious about third-party reference uploads at this time."	"Our verification-first architecture ensures all faces are platform-vetted."

We are confident without being grandiose.¶

Confident (good)	Grandiose (bad)	Apologetic (bad)
"We don't have a model export feature. We will not build one."	"We have built the world's most secure, privacy-preserving, future-proof AI likeness platform."	"We have made the difficult decision not to support model export at this time, though we understand this may be inconvenient."

Writing samples by context¶

How the brand sounds across the surfaces it appears on:

Homepage hero copy¶

``` Your likeness. Your rules. Revocable.

Likeness is a consent-first AI likeness platform for verified adult creators. License a private AI version of yourself to your subscribers, on rules you set. Revoke whenever you want. The model never leaves our servers.

[Apply to the concierge cohort] ```

Onboarding screen — what creators see when they start¶

``` Before you train your model, three things to know.

We will never export your model. There is no version of "give me my LoRA" — not for you, not for partners, not for us. The weights stay on this platform. That is the architecture, not the policy.

Every output your fans generate is watermarked, signed with your license ID, and traceable to the buyer. If a generated image shows up off-platform, we can prove where it came from.

You can revoke at any time. Stop new generations, pause the public gallery, ban a fan, or shut the whole thing down. The revocation applies forward immediately. Files already downloaded are still on people's devices — we can't change that, but we can help you take them down.

If any of this is a deal-breaker, this is the right time to say so. ```

Error message — license check rejected¶

``` This prompt isn't allowed by your license rules.

Specifically: you've blocked the "school setting" category in your rules.

You can adjust your license at any time in Settings → Content Categories. Changes apply forward immediately. ```

Customer support email — to a creator with a billing complaint¶

``` Subject: Your March payout question

Hi Jess,

Thanks for the note. Looking at your account, your March payout was held because the processor flagged a chargeback spike on the week of March 15. That's an automated hold, not a decision against you. The hold releases automatically once the chargeback rate normalizes — likely Wednesday.

In the meantime, the funds are visible in your dashboard. You'll get a release notification when they move.

Two things on my mind that I want to flag:

The chargeback spike was on AI generations specifically. We're looking into whether this was a single bad-actor fan or something broader. I'll have an answer by Thursday.
If this happens again, the processor will hold longer. Worth tightening your fan-side rules on per-day generation limits if you're comfortable with that.

Sound okay? I'm here.

— Mara, Creator Ops ```

``` We hired a Compliance & Legal Lead. Their job is making sure the consent and recordkeeping infrastructure actually works — not making it look like it does.

Welcome [Name]. The next month is yours: 2257, processor posture, AI likeness law tracking, agreements. Pick whatever order you want. I'll get you the inputs. ```

Press email — pitching a journalist¶

``` Subject: Likeness — consent-first AI likeness for verified adult creators

[Journalist name],

Likeness is launching a consent-first AI likeness platform for verified adult creators. Models never leave our servers, every output ties to a license, creators can revoke at any time, no third-party reference uploads.

The narrower point: the platforms that fans currently use to generate adult AI content are mostly ungated and mostly unconsented. Our position is the legitimate alternative sitting on the right side of where regulation is going — TAKE IT DOWN Act, NO FAKES Act, California AB 2602.

Worth a 30-minute conversation? Happy to walk through the architecture, the regulatory frame, or the cofounder team.

— [Founder name] ```

Investor follow-up — after a meeting¶

``` Subject: Likeness follow-up — three things you asked about

[Investor],

Thanks for the time today. Three follow-ups on questions you raised:

Payment processor risk — Civitai's processor cutoff in May 2025 is the canonical example. Our mitigation is multi-processor from day 1; CCBill, Segpay, and Verotel conversations are open. I'll have indicative-terms responses from at least two by [date].
Vylit — yes, ex-OnlyFans CEO Ami Gan, $2.7M Seed Sept
They explicitly excluded explicit content. The gap is real, and our positioning is the larger creator monetization opportunity. Happy to walk through the competitive cell analysis.
Why $1.5M, not $1.2M — the Compliance Lead seat is the most expensive single hire and underpins processor relationships, 2257, and the takedown apparatus. $1.2M makes that hire impossible at cofounder-modest cash.

I'll have the cofounder pipeline update for you next Friday.

— [Founder name] ```

Vocabulary guide¶

Preferred terms¶

Use	Why
License	The structured object creators configure. Specific. Adopted.
Revoke / Revocation	Plain word; carries the worker-rights frame.
Approval queue	Concrete; familiar from existing creator workflows.
Watermark / Watermarking	Standard in the industry. Don't reach for fancier.
Provenance	Specific term-of-art for the C2PA and signed-metadata work. Use when accuracy matters.
Verified creator / Verified	The platform's identity floor. Use consistently.
Concierge phase / Concierge cohort	Internal-facing project name; use externally only when explaining MVP scope.
Processor / Adult-friendly processor	Industry standard. Use without explanation among creators.

Avoided terms¶

Avoid	Why
Deepfake (in our materials)	Carries victim/perpetrator framing. Use only when describing the gray-market alternative.
AI twin / AI clone / AI version-of-you (in marketing)	Cute branding that minimizes the worker-control issue. Creators recoil.
Passive income	Wrong frame for what creators actually do; minimizes the work.
Synthetic	Confusing — overloaded with non-consenting fabrication. Use "AI-generated" instead.
Train your model	Sounds technical. Use "build your private model" or "we train a model from your photos and videos."
Unleash / Empower / Revolutionize	Marketing register. Off-brand.
Innovative AI-powered platform	Off-brand. Off-brand. Off-brand.
Cutting-edge	Off-brand.
Disrupt / Disruption	Off-brand.
✨ / 🚀 / 🔥 / any emoji	Off-brand. CLAUDE.md is correct here.
Family / Tribe / Community (when describing creators)	Creators are not our family. They are our customers. Don't pretend otherwise.

Terms used carefully¶

Term	When to use / when not to
Consent	Use when describing the architecture and regulatory posture. Don't use as a tagline ("powered by consent" is off-brand).
Worker / Worker-controlled	Use in policy / labor-aligned contexts (investor or press materials about regulatory direction). Avoid in casual creator-facing copy where it can sound like the founder is co-opting a movement.
Safety	Use when describing trust & safety operations. Don't use as a feel-good adjective ("safe AI").
Open / Transparent	Use when describing audit logs and per-output provenance. Don't use as a vibe ("we're transparent").

How brand voice shows up — register summary¶

Surface	Register	Length
Homepage / hero	Direct, declarative, punchy	Short
Onboarding	Direct, careful, named tradeoffs	Medium
Error / system messages	Plain, specific, action-oriented	Short
Creator support email	Adult-to-adult, named issues, named next steps	Medium
Social media	Plain, infrequent, factual	Short
Press / investor materials	Specific, evidence-led, regulatory-aware	Medium-long
Long-form (founder brief, blog posts)	Direct prose, first-person founder voice, named hard parts	Long

The thread across all of these: an adult talking to other adults about a thing that matters.

What CLAUDE.md got right and what to update¶

The voice in CLAUDE.md is mostly correct. Specifically right: "direct, founder voice; no marketing-speak; no hype; no emoji; honest about hard parts; treats the reader as an adult; avoids both the lurid adult-industry register and the generic AI-app register."

What to consider updating in CLAUDE.md based on this document:

Add the explicit "we are X / we are not Y" examples — they make the voice operational rather than directional.
Add the vocabulary preferred / avoided / careful tables — they prevent drift.
Add the register-by-surface table for designers and writers who join the team.
Pin "Plain talk" to its values entry in mission-vision-values.md so the voice isn't a separate document; it's an output of the values.

Strategic Connections¶

The vocabulary guide reflects the language map in 01-discovery/target-audience.md.
The press / investor register reflects the regulatory and competitive framing in 01-discovery/industry-trends.md and competitor-landscape.md.
The "Adult" trait reflects the founder-market-fit gap analysis in 01-discovery/confidence-dashboard.md — speaking to adult creators as adults is a respect-signal that compensates partially for the founder being outside-in.

Flags¶

Red Flags: - None.

Yellow Flags: - The voice as written is the founder's voice. Once Creator Ops cofounder is hired, the voice should be reviewed against creator-side feedback. Specifically: are creators reading "Plain" as "respectful," or are they reading it as "cold"? Adjust if needed. - Customer support voice ("Mara, Creator Ops") is illustrative. The actual voice should reflect the actual person, with their own register inside the brand frame.

Sources¶

CLAUDE.md — original voice anchor
docs/founder-brief.md — voice reference document
01-discovery/target-audience.md — language map
01-discovery/raw/customers-demand.md — creator concerns and term-resonance signals