AI × Marketing. Measured.
UNMODELED

AI × Marketing · Real experiments · Real data

Everyone has an
opinion about AI.
We have the data.

Unmodeled tests what AI actually does to marketing performance.

Real experiments. Real data. No AI hype.

AI × Marketing. Measured.

Free. No spam. Just experiments. One-click unsubscribe.

The problem

Your feed is full of AI.
Evidence isn't.

Every day marketers are told that a new model, agent or workflow will change everything. Most of those claims are never measured.

Unmodeled asks a simpler question

Did it actually work?

The Unmodeled method

We test it.
You get the data.

  • No screenshots of impressive prompts.
  • No cherry-picked demos.
  • No vendor talking points.
  1. 01

    Hypothesis

    What marketers claim AI can do.

  2. 02

    Experiment

    We test it in a real marketing environment.

  3. 03

    Data

    Traffic. Conversion. CAC. Revenue. Rankings. Time saved. Whatever actually matters.

  4. 04

    Verdict

    Use it.
    Ignore it.
    Or keep testing.

Latest research

The kind of questions
we investigate.

OpinionSubscribers

Field note · Measurement

Why I paid $1,000 for one word

2 min read
OpinionSubscribers

Field note · AI analytics

What made Google's AI understand users better wasn't more data

6 min readDataset: Offline evaluation against topics clicked by hundreds of users; user survey with thousands of US users
OpinionSubscribers

Field note · AI advertising

The agency fee is about to become ad spend

5 min read
OpinionSubscribers

Field note · Measurement

Google's recommender showed new content 41% less. It performed better

5 min readDataset: Live A/B test on a commercial video platform serving billions of users, equal non-overlapping traffic per arm
OpinionScheduled

Field note · AI analytics

When everyone has the same models, the advantage is the data they don't have

Out Thu 22 Oct
ExperimentIn the lab

AI advertising

Claude vs GPT vs human: Who writes ads that actually convert?

Dataset: being collected

Browse all experiments →Everything, in order →

What you won't get

Less noise.
More signal.

Most AI newsletters are written to be shared.
This one is written to be used.

  • "47 AI tools you need to try this week"
  • Recycled product launches
  • Prompt packs
  • AI doom
  • Vendor talking points
  • Hot takes without evidence
  • → What worked.
  • → What didn't.
  • → What the data says.
  • → What you should do next.

Why Unmodeled

Written by Rafa, founder of Sealmetrics.

I spend my days working with marketing data and measurement. Unmodeled is where I test the claims being made about AI and marketing — and publish what the data actually says.

Some experiments use anonymized or aggregated measurement data. Unmodeled is editorially independent from Sealmetrics.

About the methodology →

Every piece is labelled

Experiment
A controlled test with a stated hypothesis, method, sample and verdict.
Observation
A pattern seen in real data, without a control group. Useful, not proof.
Opinion
What I think, and why. Always marked as such.

Stop collecting AI opinions.
Start collecting evidence.

Get the next experiment in your inbox.

Free. No spam. Just experiments. One-click unsubscribe.