Native ResearchLive in Native
Meet Norn
Our model for artificial intuition, researched and built in house at Native. Norn does not write. It predicts which idea people will respond to, before anything is published.
01Artificial intuition
A model that does not write
Language models are brilliant at writing, images and video. They are bad at knowing what to talk about. They are trained to predict the most likely next word, so their ideas drift toward the average, and the average is exactly what people scroll past.
Norn works at the other end of the curve. It predicts which idea will stand out for one brand, with one audience, at one moment, and it learns that from what people do: which ideas they approve, rewrite or turn down, and how the posts perform once they are out. We call it artificial intuition.
- Language models: the most likely words, so the most average ideas
- Norn: the ideas most likely to stand out
02The API
Ask Norn before you publish
Today Norn works inside Native, where more than 2,200 businesses approve, edit and turn down its ideas. The API opens it to any product. Send Norn a brand, and it sends back ideas in order, best first. The model you already use writes the posts from there.
The API is in development, and access opens from the waitlist below.
Ideas, not words
Each idea is a plan, not a post: the point to make, the angle, the format and why now, with a prompt ready for the model you already use.
Sources you can check
When an idea makes a claim about the business, it cites where the claim comes from, such as a page on the website or a customer review.
Learns from decisions
Tell Norn which ideas people kept, changed or turned down, and how the posts did. It learns each brand’s taste, the way it does inside Native.
Request
curl -X POST https://api.native.no/v1/ideas \
-H "Authorization: Bearer $NATIVE_ACCESS_TOKEN" \
-H "Content-Type: application/json" \
-d '{
"brand": "brand_7Hk2pQ",
"count": 2,
"focus": "We want more emergency callouts this winter."
}'The first idea, once Norn is done (shortened)
{
"id": "idea_4Fd9xW",
"object": "idea",
"rank": 1,
"title": "The Sunday frozen-pipe rescue",
"hook": "Frozen pipes do not care that it is Sunday. We do.",
"angle": "Deadpan humour about the worst-timed emergency, ending on proof that the team shows up the same day.",
"why_now": "Winter is starting in Columbus, and burst pipes come with the first hard frosts.",
"format": "video",
"sources": [
{
"type": "website",
"title": "Emergency plumbing",
"url": "https://pipedream.example/emergency",
"quote": "Same-day callouts, seven days a week."
},
{
"type": "review",
"title": "Review from a customer in Clintonville",
"url": null,
"quote": "They came on a Sunday and had the water back on in an hour."
},
{
"type": "focus",
"title": "Your focus",
"url": null,
"quote": "We want more emergency callouts this winter."
}
]
}What you could build
- A content tool that suggests what to post, not only how to write it.
- A website builder or CRM that drafts the first week of posts for every new customer.
- An agency workflow that sends each client the strongest ideas to approve.
- An AI agent that asks Norn what is worth writing before it writes.
Join the waitlist
Leave your work email, and tell us what you would build if you like. We will write when your access is ready.
03The name
Named for the weavers of fate
In Norse mythology, three norns sit at the roots of Yggdrasil, the world tree. Urd, Verdandi and Skuld. They weave the fate of every world, of gods and people alike.
We chose the name because prediction is the same craft. Read what has happened with care. Watch what is happening closely. Then the future stops being a guess.
- What has happened
- What is happening
- What comes next
Urd
What has happened
Every post that worked, and every one that did not. Norn learns from performance history across industries, platforms and formats.
Verdandi
What is happening
Trends, conversations and rising formats. Norn reads the present as it unfolds, so an idea lands in its moment.
Skuld
What comes next
The prediction itself. The idea your audience is most likely to respond to, suggested before you ask for it.
04Training
Trained on four signals
Most models are trained to generate. Norn is trained to predict. Four signals ground every suggestion the model makes in observed human behaviour.
Brand archetype
Who the brand is. A jester does not speak like a sage. The archetype decides the voice, the humor and the angle an idea can take.
Jung 1959 · Aaker 1997 · Mark & Pearson 2001Read the researchSocial performance
What actually happened. Engagement across platforms, industries and formats gives every prediction a base of real outcomes, not opinion.
Berger & Milkman 2012 · Khosla et al. 2014Read the researchTrending data
What the world cares about right now. Timing is part of the idea, so Norn weighs what is rising and what is fading.
Choi & Varian 2012 · Asur & Huberman 2010Read the researchAcceptance rate
What people keep. Every suggestion a user accepts, edits or rejects teaches Norn what a good idea looks like. We call this taste, and we published our research on it.
Hu et al. 2008 · Christiano et al. 2017Read the research- Brand archetype
- Social performance
- Trending data
- Acceptance rate
For this brand
New profile, strong voice. The archetype and the moment lead.
- Brand archetype38%
- Social performance16%
- Trending data30%
- Acceptance rate16%
05Method
From context to idea
Norn sits between the world and the models that make things. The model does not make the video. It decides what the video should be.
Context in
Your website
- 01Brand archetype
- 02Social performance
- 03Trending data
- 04Acceptance rate
Norn
predicts
One idea
A generative model
- Text model
- Image model
- Video model
The idea arrives as precise instructions for whichever model renders it best.
Approve / SkipAcceptance rate · Taste
The content Video · Image · Text
Context in
Your website, your archetype, public data and anything else relevant we can gather. Norn starts from who you are and where you are.
Norn predicts
The model weighs the four signals against each other and predicts the idea your audience is most likely to respond to.
An idea out
The idea leaves Norn as precise context for whichever model makes it best, like ChatGPT, Claude or Grok. Text, image or video. The medium changes. The idea holds.
- Context in
- Norn predicts
- One idea
- One idea
- A generative model
- The content
06The system
Norn predicts. Native does the rest.
A prediction is only useful if you can act on it without thinking. Native is the system around the model. It turns Norn’s idea into a finished post and places it in an interface anyone can use.
You see finished suggestions, not settings. Approve the ones you like. Native handles production, scheduling, publishing and the replies.
Illustrative
Suggested by Norn
Frozen pipes do not care that it is Sunday. We do. Same-day fix, no drama.
A worked example
A plumber in Ohio
A plumber logs in to Native from Ohio. The brand archetype is the jester.
Norn does not reach for a generic plumbing post. The model predicts how this plumber should talk to their audience. What jesters in home services get away with. Which jokes carry a message about trust and craft. What people in Ohio respond to this time of year.
What comes back is not content yet. It is a prediction of the best move this specific brand can make today. Native takes it from there.
Candidate ideas · Illustrative
Plumbing · Ohio · The jester · January
- 01
A before-and-after reel of a frozen-pipe rescue, with a deadpan line about Sunday emergencies. Post Thursday at 18:00.
- 02
A meme about the one radiator every house refuses to bleed, ending on a book-us line. Post Saturday morning.
- 03
A 15 second guess-the-leak quiz from a real callout, answer in the comments. Post Tuesday at 09:00.
An idea, not content yet. Native takes it from here.
07The literature
Standing on published ground
Norn is not a bet on intuition. Each part of the model follows a line of peer-reviewed research, from the psychology of archetypes to learning taste from human choices. A selection of what we build on.
Published results from the field, not Norn’s own numbers. Norn is early, and its own benchmarks will be published here as they firm up.
+40%
more clicks when a message matches the reader’s psychology
Matz et al. 2017, PNAS
+50%
more purchases from psychologically matched ads
Matz et al. 2017, PNAS
2.3M
images used to show popularity can be predicted before posting
Khosla et al. 2014, WWW
<1%
of interactions needed as human feedback for a model to learn what good means
Christiano et al. 2017, NeurIPS
Predicting behaviour
- Kosinski, Stillwell & Graepel (2013). Private traits and attributes are predictable from digital records of human behavior. PNAS, 110(15).
Ordinary digital traces are enough to predict who someone is. Behaviour is far more predictable than it feels.
- Matz, Kosinski, Nave & Stillwell (2017). Psychological targeting as an effective approach to digital mass persuasion. PNAS, 114(48).
Messages matched to a person’s psychology performed up to 40 percent better in field experiments reaching 3.5 million people.
The archetype
- Jung (1959). The Archetypes and the Collective Unconscious. Collected Works vol. 9.1, Princeton University Press.
The archetypes are old and shared. People recognize a jester or a sage without being told.
- Aaker (1997). Dimensions of Brand Personality. Journal of Marketing Research, 34(3).
Brand personality is measurable. Consumers read brands along stable, testable dimensions.
- Mark & Pearson (2001). The Hero and the Outlaw. McGraw-Hill.
Strong brands hold one archetype and keep it. This is the playbook Norn learns per brand.
- Malär, Krohmer, Hoyer & Nyffenegger (2011). Emotional Brand Attachment and Brand Personality. Journal of Marketing, 75(4).
Attachment grows when a brand’s personality matches the audience’s actual self. Congruence is the lever.
The performance
- Berger & Milkman (2012). What Makes Online Content Viral? Journal of Marketing Research, 49(2).
High-arousal emotion travels. What gets shared follows patterns, and patterns can be learned.
- Khosla, Das Sarma & Hamid (2014). What Makes an Image Popular? WWW ’14.
Popularity can be predicted from content features before anything is posted.
The moment
- Choi & Varian (2012). Predicting the Present with Google Trends. Economic Record, 88(s1).
Search data reads the present before official numbers do. Timing is measurable.
- Asur & Huberman (2010). Predicting the Future with Social Media. IEEE/WIC/ACM WI-IAT.
The volume of social chatter beat market experts at predicting box office. Attention foreshadows outcomes.
The taste
- Bourdieu (1984). Distinction: A Social Critique of the Judgement of Taste. Harvard University Press.
Taste is not random. It is structured, social and therefore learnable.
- Hu, Koren & Volinsky (2008). Collaborative Filtering for Implicit Feedback Datasets. IEEE ICDM.
What people do, keep and skip reveals preference better than what they say.
- Christiano, Leike, Brown, Martic, Legg & Amodei (2017). Deep Reinforcement Learning from Human Preferences. NeurIPS 30.
Models can learn what good means from simple human choices between two options. Acceptance and rejection are enough.
08Why we build it
The end of doing marketing
Marketing is applied psychology. Our vision is to predict human behaviour, because that is what a great marketer actually does. They know what will land before they make it.
Norn makes that call today, and it is built to make it better than any other model, in time better than human intuition. Trained on real behaviour, it is learning what a great idea is for any plumber, baker or founder anywhere in the world.
The mission is simple. You never do marketing again. Norn predicts. Native produces. You run your business.