Research / Field notes

Questions worth testing.
Methods worth showing.

I use research to make decisions more defensible: define the question, preserve the baseline, inspect the evidence and be explicit about what remains unknown.

Current investigations

Work in progress, with the evidence visible.

01 / AI Search · Applied analysis

Notion: SEO, AEO and GEO

From a public-source baseline to a prioritized intervention register. The study examines category clarity, product boundaries, comparisons and answer-ready information architecture.

Status: documented proposals, not changes made to Notion or measured ranking gains.

Explore the study

02 / AI Search · Competitive design

Four platforms, one comparison protocol

A frozen set of 35 prompts and an evidence framework for studying how AI systems describe and compare Notion, Asana, ClickUp and monday.com.

Status: protocol and public-source evidence prepared; AI response observations are not yet collected.

Explore the protocol

03 / Tooling · Reproducibility

AI Search Visibility Lab

Python-based tooling for structured observations, validation and repeatable analysis. The emphasis is on making a measurement auditable before making it persuasive.

Status: research framework and demonstrations; not client performance data.

Inspect the repository

From question to evidence

How an investigation takes shape.

01 / Define

Start with a decision, not a dashboard.

For a product comparison, the question might be whether an AI answer describes a capability accurately, not simply whether a brand name appears.

02 / Record

Preserve what was actually observed.

Store the exact question, model, date, answer, cited URLs and relevant source excerpts. A product page alone cannot stand in for an AI response.

03 / Evaluate

Make room for disagreement.

Score presence, citation and recommendation separately. Mark ambiguous cases for review rather than forcing a confident number.

04 / Revisit

Check whether the finding survives repetition.

Repeat with the same protocol, document changes and report uncertainty. One persuasive example is not a reliable trend.

Research discipline

An observation is not a conclusion.

Freeze the question. Record prompts, versions and evaluation criteria before looking at outcomes.

Separate the layers. Distinguish first-party product claims, independently verified facts, model responses and my own interpretation.

Keep uncertainty visible. Missing observations remain missing; a proposed optimization is not a demonstrated improvement.

Make revision possible. Document assumptions and source dates so findings can be challenged or updated.

Explore the method

What does it actually mean to measure AI Search visibility?

Retrieval, citation, recommendation and answer quality are related, but they are not interchangeable measures.

Read the AI Search methodology