01
Citation selection
How answer engines attach sources, and how that differs from simply naming a brand in the prose.
AI visibility research
What shows up when someone asks an AI.
Northline Research studies how brands are named, cited, and recommended in answer engines. We publish methods first, then findings.
Focus
What we research
02
Share of voice
How to count mention, citation, and recommendation as separate events inside a defined prompt panel.
03
Prompt design
How a question set becomes a sample frame, and how branded prompts distort an audit if they are mixed in.
04
Platform change
How to observe Google AI Overviews and chat engines without treating every interface shift as a new law.
Method
Research approach
Frame
Write the question before the prompt
Every study starts with a user job: learn, compare, select, verify, or operate. Prompts come second. If we cannot state the job, we do not collect yet.
Sample
Build a panel you can run again
We version prompt panels, separate branded from unbranded items, and keep the list small enough to repeat. Novelty is not a design goal.
Code
Score events, not vibes
Mentions, citations, and recommendations are coded on separate fields with a written codebook. Disagreement between readers is recorded.
Limit
Report what the method cannot say
We do not convert answer text into traffic or revenue. We do not blend engines into one pie. If a claim needs a number we did not collect, it stays out.
Blog
Latest posts
How answer engines choose citations
What can be observed about source selection in AI answers, and what remains hidden inside unpublished ranking systems.
Measuring brand share of voice in AI answers
A measurement frame for mention, citation, and recommendation that does not pretend a single prompt is a market.
Designing a prompt panel for visibility audits
How to build a reusable set of questions that can support an AI visibility study without collapsing into brand theater.
Contact
Briefings, studies, and methods questions
If you want a competitive landscape study, a visibility benchmark, a prompt panel, or a vendor evaluation, write to us. There is no form.