Sourced claims · Digital-twin A/B tests
We research your market the way a scientist reads papers, write a field of variants grounded in real evidence, and A/B test them in rounds on digital twins of your audience, built from real audience data. You get the winners, and the winners are what you pay for.
The problem
Generic tools write posts that sound fine, read like everyone else's, and now and then state facts nobody can source.
Each test needs budget and days of traffic, so most teams compare two versions, if they test at all.
Without evidence, the choice goes to whoever argues best in review, not to the version your audience would choose.
How it works
Two engines do the work: one finds what's true, the other finds what lands.
A research agent reads your brand material, reviews and comments, competitor posts and published research, and turns them into ranked, cited evidence.
→ Evidence briefWe write a wide field of variants with different hooks, angles and formats. Every factual claim is tied to a source. Claims without one are cut.
→ A field of sourced variantsThousands of digital twins, each built from real audience data, meet the variants in a simulated feed. They scroll past, like, reply, share and argue.
→ Reactions and objectionsVariants compete in rounds. Losers are cut, survivors are rewritten from what the twins said, and each round faces a fresh panel.
→ Winners and a full reportEngine 1 · Evidence
Before a word of copy is written, a research agent works through your material and the published record the way a careful researcher would. It searches widely, reads closely, weighs every source and keeps only what it can cite. The copy is written from that evidence, not from a model's memory.
Steeped for 18 hours
Brand brief · product spec, brewing process
No added sugar
Brand brief · nutrition panel
Less acidic, so it's easier on your stomach
Cold and hot brew had comparable pH (4.85–5.13) in a 2018 study; hot brew had more titratable acid. None of the sources in this brief test the stomach claim.
Rao & Fuller, Scientific Reports, 2018 ↗
Rewritten as Smooth, never bitter
, the words customers use in their reviews.
Twice the caffeine of regular coffee
No source found
Engine 2 · Digital twins
The readers in every test aren't invented. Each one is a digital twin built from real audience data that we curate, and it carries the views, language and habits found in that data. Twins meet your post in a simulated feed, where they scroll past, like, reply, share and argue with each other, so you see how a post travels, not just how it scores.
No sugar. No syrup. Just beans, cold water and 18 hours.
Carried into round 3: “What's different?” came up across the skeptical segment, so the next rewrite names where the beans come from.
The tournament
One A/B test tells you which of two drafts is better. Several rounds of them, with rewrites in between, find the best of many. Each round, the weakest variants are cut, the survivors are rewritten from what the twins said, and a fresh panel of twins judges the next round.
Round 1
8 advance
Round 2
4 advance
Round 3
2 winners
Winners
V07 · after 3 rewritesNo sugar. No syrup. Just Chikmagalur beans, cold water and 18 hours.
V14 · after 3 rewritesSteeped for 18 hours, so your 7 a.m. doesn't have to be.
What you get
Ready-to-post copy for the channels you choose, in your brand's voice.
Every factual claim linked to the passage it came from. Easy to check, easy to defend.
Who each winner lands with, who it loses, and the objections readers raised, in their own words.
How every variant did in every round, and why the losers lost.
Pricing
Research, drafts and test rounds are how we find them. What you buy is the collateral that comes out on top.
Pricing model
Per winning collateral
Pilot pricing on request.
FAQ
It's closer than a test on made-up personas, because every twin is built from real audience data. It still doesn't replace a live test. Twins are best at fast, directional calls: which hook lands, which claim draws pushback, which version a segment scrolls past. They are less reliable for exact numbers, such as how many people will click. We use them to get from many drafts to a few strong ones quickly. Before you put serious budget behind a winner, a small live test is still worth running.
From the material you share, such as brand guidelines, product specs, customer reviews, comments and past posts, and from published sources such as research papers, industry reports and competitor content. Every factual claim in a winning collateral links back to the passage it came from. If a claim can't be sourced, it doesn't ship.
From real audience data that we curate: customer reviews, social comments, customer interviews and the segment data you share. Each twin is grounded in that data rather than invented from a one-line demographic label, so it carries the views, language and habits found in the data. Twins react to each other as well as to your post. Every round uses a fresh panel, we check that twins disagree as much as real people do, and when you share results from past posts, we test the twins against them first.
Anything that lives in a social feed: post copy, captions, hooks, carousel text and ad copy. Tell us your channels when you request access.
Early access
We're opening early access to a small group of brands and agencies. Tell us about your brand and what you're launching next, and we'll be in touch.
Thanks for your interest. We'll be in touch at .
Aditi K. · 27 · runs before work
No syrup is exactly why I switched. Where can I buy this?
Wants to buyMarcus R. · 41 · wary of new brands
Every cold brew says no sugar now. What's actually different about yours?
ObjectionJun L. · 33 · café regular
Same question. Where are the beans from?
Piles on