GPT
Illustrative writing styleRelax in our newly renovated rooms, complete with private jacuzzis and everything you need for a comfortable weekend away.
Illustrative example · conversion rate: not measuredResonance Writing, with a real-world feedback loop
An AI model that writes
the way people love to read.
Good writing makes someone feel something. Or understand something. Or finally click.
We’re building a language model that learns from what happens next.
Built on Seatext’s real-world copy experiments.
Resonance
Illustrative writing directionA room for two.
A jacuzzi all to yourselves.
Give the reader a moment they can picture.
Illustrative example · conversion rate: not measuredThe story doesn’t end
at the last word.
It ends with a person. Someone who keeps reading, closes the tab, books the room, or buys the thing.
For two years, Seatext has tested AI rewrites against original website copy with real visitors. That gives us something a writing sample alone can’t: the words, the context, and what people did next.
That’s the feedback we’re building Resonance around.
2,500websites in the Seatext network
300M+visitor impressions measured
30,000+conversion A/B tests reported
Seatext infrastructure figures reported by the founder. These describe the data source, not Resonance model performance.
03 / A small writing deskBring a line.
Bring a line.
Find a new direction.
A headline you’ve rewritten ten times.
A benefit that still sounds like a feature.
Start there.
Room to explore01 — 03
01 / Clearer
Get to the thing that matters.
02 / Warmer
Make it feel like a conversation.
03 / More direct
Give someone a reason to act.
04 / From words to outcomesBetter feedback.
Better feedback.
Better foundations for writing.
The research starts with a simple question:
which words worked, for whom, and where?
- 01
Test the words
Compare original copy and rewrites on real websites. Keep the audience, page intent, and offer alongside the text.
- 02
Read the response
Measure reading behavior and business outcomes separately. Staying longer and making a purchase answer different questions.
- 03
Learn from the difference
Use qualified winning and losing text pairs to develop reward models and post-train open language models.
A closer look at the research
Our planned approach combines outcome reward modeling and Direct Preference Optimization (DPO), starting with open 70B weights. The goal is writing that works for its audience and context, rather than one fixed house style.
Evaluation needs to separate training data from held-out websites and future experiments, account for traffic and test uncertainty, and measure reading and conversion outcomes independently. We’ll publish model-specific results with their methodology when available.
The prototype above uses third-party inference. Model weights, API availability, and measured model performance will be announced separately.
Let’s write what comes nextYour next landing page
Your next landing page
could teach us something.
We’re looking for growth teams who care about the words as much as the numbers.
Tell us what you’re working on. We’ll follow up to discuss fit for an early pilot and how you evaluate copy today.
Resonance by Sergei Glukhov · built on SeatextA few things you might be wondering.
Is this a writing tool or a language model?
Resonance is a language model project. The writing desk is an early prototype of the experience we’re exploring for growth teams. It currently runs on third-party models.
Does better writing always mean shorter writing?
No. A good headline might need six words. A useful explanation might need six paragraphs. Our aim is to learn what works for the reader and the task, including voice, pacing, clarity, and depth.
Can I use the API or download the weights?
We aren’t announcing public API or weight availability on this page. Join the pilot to discuss your use case and hear about future access.
How is Resonance connected to Seatext?
Sergei Glukhov built Seatext’s website testing and measurement infrastructure. Resonance is a new venture built on that work, focused on learning from how real people respond to writing.
