Audience Voice Intelligence API

Your audience doesn't talk like an NPR host.
Why does your AI?

[Brand] scores and rewrites your campaign copy against the revealed-preference language of the exact audience you're trying to move — not the audience your AI was trained on.

●Proprietary behavioral signal — not another LLM wrapper. Built on 15 years of revealed-preference audience data.
●Authenticity that converts — purpose-built for the voters, donors, and members your AI is mistuning for.
●Center-left audience expertise — campaigns, causes, foundations, labor. We don't work with the other side.

Book a 15-minute demo See a sample authenticity score →

The problem

Your AI thinks everyone talks like it does.

Every large language model on the market — Claude, GPT, Gemini, Llama — was trained with human preference labels. Those labels came from annotators. Those annotators had a voice.

Published research and our own measurement confirm the same thing: that voice is overwhelmingly urban, college-educated, and coastal. When your campaign's generative AI was fine-tuned, the people telling it "this is good writing" were a narrow slice of the country.

That's fine when you're writing a blog post. It's a disaster when you're trying to reach a welder in Youngstown, a retired teacher in Albuquerque, a first-time Latino voter in Phoenix, or a union steward in Pittsburgh.

Your campaign's AI-generated copy sounds like the person who labeled the training data — not the person you're trying to persuade. Every dollar of paid media built on top of that copy has an authenticity tax baked in.

Labeler bias is measurable

Preference-label research shows systematic demographic skew in the data that shaped every frontier model.

Generic copy under-performs

Our internal benchmark shows a consistent authenticity gap versus human-authored control copy for non-coastal audiences.

The gap is growing

Your voter doesn't write like your copywriter. Now your AI doesn't either — and it's compounding with every send.

The solution

[Brand] gives your AI a localized ear.

We've spent 15 years building a proprietary behavioral signal for how specific audiences actually talk, share, and persuade each other — built from revealed preference, not from what people say they like in surveys or what a contract labeler clicked at 2am.

That signal powers four things your team can use on day one:

Score.

Submit any candidate copy. Get an authenticity score against the exact audience you're targeting — before you spend a dollar in paid.

e.g. one fundraising email, scored against "sustaining progressive donor."

Compare.

Submit N message variants. Get them ranked against the audience. Kill the bottom quartile before QA.

e.g. 12 subject lines → ranked, with the weakest flagged.

Rewrite.

Submit LLM-generated copy. Get it rewritten in the audience's voice — without changing the policy point or the campaign's positioning.

e.g. a Claude draft, retuned for a Rust Belt union audience.

Build.

Give us a seed corpus of your audience's actual speech. We return a reusable audience profile your whole team can call against.

e.g. donor emails + canvassing notes → one profile ID.

How it's different

Notanother Jasper/Copy.ai clone. Those tools write copy. We measure and correct copy — against a signal they don't have.
Nota chatbot. It's an API. Plug it into your existing creative stack in an afternoon.
Nottrained on Reddit + Wikipedia. It's trained on revealed behavior data we've refined for over a decade.
Notfor everyone. We work with center-left campaigns, cause orgs, foundations, and labor. If that's not you, we're not your tool.

Track record

Built in the room. Refined across $50M+ in spend.

This isn't a cold-start startup. It's a 15-year methodology, productized.

A national foundation used our audience voice methodology to retool the messaging for a multi-year apathy-reduction campaign. The result: measurable lift in target-audience engagement and a retained proprietary playbook the foundation still uses.

Sector: Global development • Use case: Apathy reduction • Year: recent

A Fortune 500 healthcare client used our audience voice signal to rewrite member outreach for three underserved demographic segments. Authenticity score lifted; downstream engagement followed.

Sector: Healthcare • Use case: Member comms • Year: recent

[ Testimonial coming from our first pilot customer ]

years of proprietary methodology refinement

$50M+

in downstream paid media spend informed

sectors served: political, cause, foundation, Fortune 500

Logos available on request under NDA.

Pricing

Pricing that matches your stage

One API. Four tiers. Pick the one that matches your call volume, SLA needs, and team size.

Starter

$299/mo

For solo strategists and small in-house teams testing audience voice fit before committing.

~2,000 API calls / month
/score-voice and /compare-candidates
Pre-built audience profiles
Self-serve docs and examples
Best-effort latency

Start Starter →

See your own copy scored against your own audience.

Give us 15 minutes. We'll:

Score a sample piece of your current copy against one of our pre-built audience profiles.
Rewrite it in the audience's voice.
Show you the delta in readability, sentiment, and target-audience alignment.

If it's a fit, we'll walk through pricing and next steps. If it's not, you walk away with a benchmark score you can use.

First name

Last name

Work email

Please enter a valid work email.

Organization

Role title

Organization type

Monthly comms / ad spend

How did you hear about us?

What problem are you trying to solve?

We never sell your data. Read our privacy policy.

Got it. We'll be in touch within one business day.

A real human — not a sequence — reads your submission within 24 hours. If it's a fit, you'll get a calendar link for a 15-minute walk-through and a preview authenticity score on a sample of your own copy.

See a sample score while you wait →