THE RESEARCH
Better questions.
Better content decisions.
Who is asking? What do they need? Which sources can AI find? We study GEO personas and model worlds to compare possible interventions. The aim: choose the changes with the strongest expected return.
WHAT WE OBSERVE
Pay attention to the answer.
CURRENT PACKAGE
We record AI answers to relevant questions and look for recurring topics, missing context, visible sources, and changes over time. Growth includes 160 observations each month. Those observations help choose what to write next.
THE LIMIT
AI answers vary by question, system, date, and available sources. An observation does not guarantee that a page will be cited or send traffic. A change in an answer does not prove what caused it.
A SMALL PROBLEM WITH THE PERFECT DASHBOARD
A spherical buyer.
In a vacuum.
One generic prompt. One tidy score. A buyer with no budget, no doubts, and no follow-up questions. Very convenient for a chart. Less useful for a sale.
If a GEO dashboard treats every buyer the same, ask whose decision it actually represents. Our research starts with the messy parts: context, constraints, and the next question.
GEO PERSONAS
Same market.
Different decisions.
MODEL THE BUYING SITUATION
A GEO persona describes a goal, a budget, prior knowledge, and a reason to hesitate. It shapes the opening question and the follow-ups. These are research scenarios, not profiles of real people.
FOLLOW THE CONVERSATION
A founder asks about cost. A marketing lead asks about approvals. An agency asks about managing several brands. Each may start with “Which content tool should I use?” The next question changes what a useful answer looks like.
MODEL WORLDS
What if the information
were different?
THE BASELINE WORLD
Describe the pages, facts, and external sources available today. Keep the buyer scenario fixed. Record the answer and the conditions behind it.
ALTERNATIVE WORLDS
Build a version with a clear pricing page. Another with a detailed comparison. Another with a useful external source. Change one element first. Then explore combinations and follow-up questions.
A modeled source can help an answer only if the scenario makes it available. Live search must still find it. That is why retrieval and publication matter as much as wording.
SEARCH FOR BETTER INTERVENTIONS
Which change
earns its cost?
COMPARE THE OPTIONS
An intervention is a change you can make: a new page, a clearer fact, an updated guide, or an external contribution. Compare each option across buyer scenarios. Look for ideas that remain useful when the question changes.
PRIORITIZE EXPECTED ROI
Weigh potential business value against production cost, maintenance, time, and uncertainty. Seek the highest expected return among the options compared. A cheap FAQ may be a better first move than a large content campaign.
Without conversion and revenue evidence, this is a priority estimate, not measured financial ROI. Select a hypothesis, publish the approved change, and check what happens.
THE METHOD
A question.
A change.
A live check.
Define the GEO persona, constraints, and follow-up questions.
Record the answer, visible sources, system, and date.
Create alternative information worlds with specific content interventions.
Compare expected value, cost, and uncertainty across scenarios. Choose an intervention to check live.
After factual approval and publication, repeat comparable observations and record uncertainty.
A before-and-after comparison can inform the next decision. It does not establish causation on its own.
SIMULATION AS A RESEARCH TOOL
Compare possibilities.
Then test reality.
Our experimental framework compares possible content changes across simulated buyer questions. It can help decide which ideas to check in live observations. The comparison uses configured assumptions; it is not a prediction validated in live AI search.
Enterprise can commission a scenario workbench of up to 576 traces across an agreed persona, engine, and information-world matrix. Launch and Growth include content work and AI answer observations. Scenario results are hypotheses, not guaranteed citations, rankings, traffic, or sales.
RESEARCH BEHIND THE QUESTIONS
Ideas we build on.
- Retrieval-Augmented Generation
A study of a model that uses retrieved text while generating answers. It informs our interest in which sources appear.
- GEO: Generative Engine Optimization
Tests how content changes affect measured visibility in a controlled research setting.
- Generative Agents
Explores how simulated agents use context and memory, informing our questions about buyer context.
- Causal Inference: What If
Guides how to define a baseline, an intervention, and a comparison.
These works inform our research questions; they do not establish results for Media Automat customers.
PUT QUESTIONS TO WORK
From observation
to external action.
Setup & Launch is €1,499 per brand: setup, up to 100 articles total in month one, and an initial external source audit. The first month’s Growth articles count toward the 100. Growth is €499/month from month two, including 30 articles, up to 90 social text adaptations, 160 observations, and prioritized external action briefs.
Client fact check and approval required. One brand, one language, one website. Prices exclude VAT. Images, video, paid placements, third-party fees, and custom integrations are extra.