How the index is measured

For each vertical we maintain a corpus of real consumer questions — the things Filipinos ask when choosing a bank, a hospital, a university, a home. Each week we put that corpus to the major AI assistants and record the answers verbatim.

Assistants tracked: ChatGPT, Gemini, Perplexity, Copilot, Claude, Google AI Overviews and Google AI Mode, where they answer category questions. Measurement is scoped to the Philippines.

A brand's share of AI voice is the share of measured answers in its category that mention the brand, combined across assistants. It is recomputed from the full response set — never averaged from per-assistant shares. Weekly movement compares the latest measurement against the closest measurement at least six days prior.

Mentions count answers naming the brand; impressions estimate how often those answers are seen, weighted by how frequently the underlying questions are asked.

Share of AI voice ranks brands against each other. The answer breakdown asks a different question: of every answer we measured in a category, how many named this brand — and did it appear alone or in a crowd?

Each measured answer falls into exactly one band per brand: the brand named alone, the brand named alongside rivals, rivals named without it, or no tracked brand named at all. Because the four always sum to the same total, every brand in a vertical is measured against one shared denominator. The bars on the index draw only the bands that occurred in the latest measurement — when every measured answer named at least one tracked brand, the fourth band is empty and is not drawn.

We publish these as counts, not percentages. The same question is put to every assistant we track, so the answers behind a band are correlated rather than independent samples — and on a vertical's curated question set the base is small enough that one question landing differently would swing a rate further than the underlying visibility moved. The bars show the proportions; the numbers stay counts. We also do not grade brands into categories — two brands can be present in a similar number of answers and hold opposite positions, and the bands show that without our adjective.

This breakdown is captured once per collection run rather than daily, so it carries its own measurement date, which can be earlier than the leaderboard's. It is a point-in-time reading and we do not chart it as a trend.

Beside the ranking, each vertical page shows the tone of the sentences that name a brand. Every time a recorded answer mentions a tracked brand, our data partner classifies that sentence as positive, neutral, negative or mixed, and tags what it is about — price, reputation, service, features and so on. We count those classifications and publish the counts. We do not grade brands ourselves, we add no label of our own, and tone never enters the share of AI voice or the ranking.

These are counts, not percentages, for the same reason as the position bands above: the answers are correlated and the curated base is small. The bar shows the mix; the number stays a count. One answer can mention a brand several times, so mentions outnumber answers. A sentence the classifier could not grade is shown as unclassified rather than dropped.

Brand pages may quote the sentences themselves, verbatim, with the assistant and the question it was answering. Only sentences from answers our editors have already reviewed are quoted, and neutral sentences are never quoted. A quote stops serving the moment the assistant’s answer changes, until the new answer is reviewed.

The ranking measures visibility, not quality or accuracy. A mention is not an endorsement, and the tone counts describe how an assistant phrased a sentence, not whether it was right. Assistant responses vary by phrasing, account and time of day; treat single-week movements within a point as noise, not news.

Vertical rosters are curated editorially: the brands a Filipino consumer would plausibly compare, ranked sets of at least four. Inclusion cannot be bought, and a brand cannot pay to be removed. Roster changes are reviewed by Truelogic's research team and noted on this page when they happen.

Corrections: if you believe a measurement is wrong, tell us through our contact form with the vertical and date; confirmed errors are corrected and annotated.

Responses are collected through a commercial AI-search monitoring platform. Aggregation, ranking and editorial judgment are Truelogic's. The index is free to read and cite with attribution: “The PH AI Visibility Index, Truelogic”.

Looking for vertical-level data? Browse the index or request a briefing.