Methodology
We measure the answer, not the search.
Traditional rank trackers watch Google. DineVisible watches what AI assistants actually say when a diner asks for a recommendation — because that's where the decision now happens.
Build the query matrix
For each neighborhood we generate 80–150 natural-language queries across four dimensions diners actually use — cuisine (“best Italian near me”), neighborhood (“where to eat in my area”), occasion (“date night dinner”), and preference (“outdoor seating, good for groups”). For claimed restaurants we also phrase queries in the voice of specific diner personas — a tourist asks differently than a local.
Ask every engine
Each query runs against ChatGPT, Claude, Gemini, Perplexity in parallel. We capture the full response and any citations the engine returns — the live web sources behind a recommendation.
Parse mentions + classify sources
We detect which restaurants each answer names (handling short forms like “Padella” for “Padella, Borough Market”), note the position, and classify every cited URL — direct site, booking aggregator, editorial guide, or social.
Score on five dimensions
Each restaurant gets a 0–100 score blended from Mention Frequency (30%), Position (25%), Cross-Engine consistency (15%), Context Richness (15%), and Source Diversity (15%) — published as the public leaderboard, refreshed weekly.
Why four engines, not one?
The engines disagree constantly — one names you first, another skips you, a third cites your rival. A single-engine view is noise. Scoring across all four is what makes a ranking trustworthy, and the disagreement itself is a signal we surface.