The LLM that writes copy
people actually read.

Tired of robotic AI fluff? Standard models produce bloated corporate filler that kills conversion rates. Resonance-1 (70B) is trained on 300M+ real visitor sessions and 30,000+ conversion A/B tests across 2,500 live websites. Clean, punchy marketing copy that turns traffic into revenue.

300M+ Visitor Impressions
2,500 Live Production Sites
30k+ Conversion Pairs
70k+ Full Page Reads
+20.6% Avg Real Conversion Lift
R1

Resonance-1 Live Copywriter Demo

Sub-100ms inference • Post-processed by natural text humanizer • Zero fluff

Model Live & Ready
Try one click:
R1
Ready. Give me your current headline, ad angle, or product description. I will rewrite it to convert without the AI fluff.
Resonance-1 (70B) • Native Copywriter
01 / THE COPY PROBLEM

Why standard AI models fail marketers, and why prompt tricks cannot fix them

Frontier models are optimized for math, code, and synthetic benchmarks. When asked to write marketing copy, their weights default to generic adjective soup. Marketers spend hours editing out filler words that kill conversions.

The reality: Boring AI copy kills conversion rates

When you put generic AI copy on a live landing page, visitors bounce within seconds. Words like "delve into a tapestry of seamless innovation" signal an amateur, robotic brand. We solved this by post-training open 70B weights directly on real human purchase choices from 2,500 active websites. Every word is tuned to capture attention and trigger action.

Standard Frontier LLM (Generic RLHF) Boutique Hotel Hero Copy
Prompt: "Write an engaging, high-converting hero headline and opening for our newly renovated boutique hotel rooms."
Standard AI Output (Losing Copy):
"Indulge in unparalleled modern luxury and refined elegance at our premier destination. Our newly renovated guest accommodations seamlessly blend sophistication with ultimate comfort, boasting state-of-the-art in-room whirlpool amenities designed for discerning travelers..."
× -20.6% High Bounce Rate • Robotic & Verbose • Lost to Human Copy
Resonance-1 (70B) (Trained on A/B Tests) Boutique Hotel Hero Copy
Prompt: "Write an engaging, high-converting hero headline and opening for our newly renovated boutique hotel rooms."
Resonance-1 Output (Winning Copy):
"This hotel has amazing new rooms with a built-in jacuzzi."
✓ +20.6% Verified Direct Bookings on 42k Live Visitors • Clear Human Tone
Generic AI SEO Article Long-Form Article (2,000 Words)
Reader Behavior on 70,000+ Page Snapshots:
Ranks on Google keywords, but 86% of human visitors bounce within 12 seconds. Readers recognize the rigid, synthetic cadence and never reach the purchase button.
14% Read to End • Flat Voice • Zero Brand Attachment
Resonance-1 Long-Form Prose Long-Form Article (2,000 Words)
Reader Behavior on 70,000+ Page Snapshots:
Trained on scroll depth, dwell velocity, and retention. Captivates readers with natural pacing, conversational texture, and genuine voice that drives action.
68% Read to End • 4.8x Dwell Time • +42% Inbound Signups
02 / TRAINING ARCHITECTURE

How we post-train Resonance-1 (70B), from open weights to conversion mastery

We start with open 70B weights for grammar and fluency, then post-train using Outcome Reward Modeling (ORM) and Direct Preference Optimization (DPO) on real human conversion choices.

STEP 01

70B Open Weight Base

Start with open state-of-the-art weights providing baseline linguistic fluency, multi-lingual capability, and 128k context.

Open Foundation Weights
STEP 02

2M+ Live Rewrites & Data

Automated collection across 2,500 Seatext sites. 30,000+ verified conversion pairs and 70,000+ long-form read sessions.

Real Human Behavioral Ground Truth
STEP 03

Conversion Reward Modeling

Train reward models on actual purchase clicks, booking transactions, and scroll dwell time, not synthetic LLM judges.

Outcome & Process Reward Signals
STEP 04

Direct Preference Optimization

Execute distributed DPO training passes to align model weights directly with high-converting human completions.

Resonance-1 (70B) Production Weights
Resonance-1 (70B) Technical Specifications
Base Architecture Llama 3.3 Transformer (70 Billion Parameters, Grouped-Query Attention)
Context Window 128,000 Tokens (Supports full landing page trees, multi-page funnels, and book-length copy)
Post-Training Strategy Direct Preference Optimization (DPO) + Outcome Reward Modeling (ORM) on Conversion Telemetry
Preference Dataset 30,000+ verified A/B conversion pairs • 70,000+ full-page dwell snapshots • 300M+ real reader impressions
Primary Ground Truth Stripe checkouts, form leads, phone clicks, booking confirmations, 100% scroll depth completions
Inference Latency Sub-25ms first-token latency on distributed vLLM tensor-parallel GPU clusters
Integration Interface /v1/chat/completions • 100% Drop-in OpenAI SDK Compatible
03 / DEVELOPER QUICKSTART

One line to switch, 100% OpenAI SDK compatible

Integrate Resonance-1 (70B) directly into your existing marketing tech stack, CMS, copy agents, or SaaS applications without rewriting client code.

Endpoint: https://api.resonancetext.com/v1
# Drop-in replacement with standard OpenAI Python client
from openai import OpenAI

client = OpenAI(
    base_url="https://api.resonancetext.com/v1",
    api_key="res_live_70b_xxxxxxxxxxxx",
)

response = client.chat.completions.create(
    model="resonance-1-70b",
    messages=[
        {"role": "system", "content": "You are an expert copywriter trained on 300M+ real visitor conversion outcomes."},
        {"role": "user", "content": "Write a high-converting headline for our boutique hotel booking page."}
    ],
    temperature=0.6,
)

print(response.choices[0].message.content)
# Output: 'This hotel has amazing new rooms with a built-in jacuzzi.'
04 / APPLICATION SPECTRUM

Where Resonance-1 (70B) outperforms generic AI models

Engineered from the ground up for performance marketing, SEO agencies, and digital brands where text quality directly determines ROI.

Micro-Copy & High-Stakes CTAs

Headlines, subheads, pricing cards, and checkout buttons optimized for raw purchase conversion. Eliminates the verbose, generic adjective soup produced by base models.

Engaging Long-Form Editorial

Articles, product guides, and technical essays that readers actually read to the very bottom. Pacing, rhythm, and cadence tuned on 70,000+ real session scroll completions.

Enterprise Private Fine-Tuning

Train dedicated private Resonance models on your brand's proprietary audience conversion history. Hosted on isolated VPCs with zero data leakage guarantees.

05 / EARLY ACCESS • FOR AGENCIES & MARKETERS

Request early API access to Resonance-1 (70B)

We are rolling out private API access to selected marketing platforms, content teams, growth engineers, and performance agencies. Receive your API key and benchmark dataset access.

  • ✓ Usage-based API access • OpenAI SDK drop-in endpoint
  • ✓ Direct access to conversion preference pairs and telemetry benchmarks
  • ✓ Dedicated technical Slack / Discord support with core engineering team
  • ✓ Priority onboarding for dedicated enterprise fine-tuning runs

06 / FREQUENTLY ASKED QUESTIONS

Everything marketers need to know about Resonance-1

How does Resonance-1 differ from prompting GPT-4o or Claude 3.5 Sonnet?

Prompt engineering can change format or tone, but it cannot alter the fundamental token probability distributions baked into a foundation model's reward model. Standard models default to verbose, robotic, and flattened phrasing. Resonance-1 has been post-trained on real human behavioral outcomes: its weights fundamentally prefer completions that hold attention and trigger action.

Where did the conversion and reading training data come from?

Our underlying infrastructure, Seatext, runs live on approximately 2,500 websites with over 300 million visitor impressions. Over two years, our suite of 30+ marketing agents ran millions of live A/B tests against natural human copy. This produced 30,000+ verified head-to-head conversion pairs (tied to real Stripe purchases, leads, and bookings) and 70,000+ page-level reading snapshots measuring millisecond dwell time and scroll completion.

Can I fine-tune a private Resonance model on my own company's conversion data?

Yes. For enterprise brands and agencies with high visitor traffic, we offer private fine-tuned checkpoints trained on your brand's own historical conversion telemetry, customer reviews, and winning ad copy. These models are hosted on dedicated, isolated VPC endpoints with zero cross-tenant data sharing.

What is the relationship between Resonance and Seatext?

Resonance is our foundational copywriter model venture. Seatext is our operational marketing platform running on 2,500 websites, which serves as the automated continuous telemetry and data collection engine powering Resonance's post-training runs.

Trusted by 2,500+ high-growth brands to automate website performance & power model post-training

Visa
P&G
GoHACCP
Clico
Fjällräven
Digitopia
ESS
Crumbl
Megri Digital
Google Ads Meta Ads TikTok Ads Shopify WordPress Webflow WooCommerce GoHighLevel BigCommerce HubSpot Optimizely