Measure
Define the audience, intent, prompt set, model, date and observation schema. Establish a baseline before drawing conclusions.
AI Search / Methodology
Visibility is more than
being mentioned.
Search is changing how people discover, compare and choose products. I study what AI systems retrieve, what they cite, what they say and where the evidence breaks down.
The working cycle
Measure → Diagnose → Optimize → Compare → Report.
Diagnose
Inspect missing context, ambiguous entities, unsupported claims and differences between source evidence and generated answers.
Optimize
Propose clearer category language, product boundaries, structured comparisons and accessible source material.
Compare
Repeat controlled observations. Look for consistent changes without assuming a single prompt proves causality.
Report
Show what was observed, what was inferred, what changed and what remains unverified.
Measurement matters
Four questions, not one score.
Traditional search rankings cannot be copied directly into generative search. A response may mention a brand without citing it, cite a page without recommending it, or recommend a product using incomplete information.
Discovery
Does the entity appear in an answer to a relevant, predefined question?
Citation
Which sources are explicitly referenced, and do they support the claims made?
Recommendation
Is the product suggested for a particular use case, and with what qualifications?
Accuracy & consistency
Are descriptions supported by current evidence, and do they hold across repeated observations?
A practical distinction
A mention is not a recommendation.
Hypothetical answer A
"Notion is one of several tools teams use to organize work."
Brand presenceThe name appears. There is no evidence here that the system prefers it for the user's needs.
Hypothetical answer B
"For a team prioritizing connected notes and documentation, consider Notion; verify the current integration and security requirements before deciding."
Qualified recommendationThe answer connects a product to a use case, but still needs source and accuracy checks.
Illustrative examples written for this page, not observed model outputs or measured results.
Applied work
Research that can be inspected.
01 / Single-brand study
Notion: from diagnosis to proposed interventions
A public-source baseline, an intervention register and original examples. Proposed changes are clearly distinguished from deployed outcomes.
Read the case study ↗02 / Cross-platform comparisonNotion, Asana, ClickUp and monday.com
A 35-prompt protocol and competitor evidence matrix designed for future repeatable observations. No invented AI results.
Read the study ↗What I will not claim.
A framework is not a measured outcome. A public product page is not proof of what an AI model will answer. And an observed difference is not automatically the result of an optimization. Keeping those distinctions clear is part of the work.
Explore all projects →