For journalists and researchers
How to use the Trust Index in a story
The index measures one thing: which eligible businesses AI assistants named when asked the questions buyers ask, during a stated window. Everything below is written so a number from this site can be quoted accurately and checked by a reader.
Readings publish within 10 days of the capture window · A named analyst signs each reading and answers questions about it
What a number on this site means
Examples are from the provisional November 2026 reading of Greater Toronto home builders: 16 scored questions, 3 assistants (ChatGPT, Gemini and Google AI Mode), 3 runs each, 144 scored answers.
| You see | It means | It does not mean |
|---|---|---|
| Share 42.4 | The builder scored 42.4 out of 100 on AI recommendation share: how often it was named across the 144 scored answers, weighted by how early in each answer it was named. 100 would mean named first in every answer | That 42.4% of buyers choose it, or that assistants “rate” it 42.4 out of 100 |
| Spread 5.7 | How much the share moved between the three runs (standard deviation). A larger spread means the assistants were less consistent | A margin of error around a true score |
| #1 of 54 named · 277 eligible | Highest share among the 54 builders that any assistant named; 277 HCRA-licensed GTA builders were eligible | That it is the best builder, or that the others are worse |
| Not named | No assistant named the business in any of the reading's scored answers | Anything about the business's quality, safety or licensing |
| ▲ 3 | Rank rose three places since the prior reading (shown from the second reading of a market) | That the business changed; assistants and their sources change too |
| Agency client | The business has a commercial relationship with a company related to the publisher. It is scored by the same method as every other row (see independence) | That the relationship affected the score |
| Provisional | The reading is published before its human name-matching check is complete. Figures can change; any change is logged on the reading | That the figures are estimates |
How to cite
“According to the Trust Index, an independent measurement of which businesses AI assistants recommend, [business] had the highest AI recommendation share (42.4 out of 100) among 277 eligible Greater Toronto home builders in the provisional November 2026 reading.”
- Name the metric: AI recommendation share or measured visibility. Avoid “ranked by AI”, “AI's most trusted” or “best according to ChatGPT”; the assistants answered questions, the index did the measuring.
- Name the reading and geography. Every figure is tied to a dated window and a city or province.
- Say “provisional” while a reading is provisional.
- Link to the reading page, not the home page. Readings are dated, and any correction is logged on the reading with its date.
- The eligible count and the named count belong together; one without the other misleads.
What the index can give you
- The file set for any final reading, including cell-level data (/data), free for reporting and research with attribution.
- A briefing on any reading, on the record and attributable to the Trust Index.
- Custom cuts of published data (by municipality, segment or assistant) from the same cells, with the method stated.
- Advance notice of a reading under embargo, on request, for a market you cover.
What it will not do
- Characterise any business beyond the published cells. The index does not publish answer text or anything assistants said about conduct, safety or complaints.
- Comment on a business's quality, licensing status or customer outcomes.
- Run a one-off prompt for a story. Every figure comes from a full reading under the published protocol.
Contact
Press and research enquiries: press@littpro.com. Corrections go through the corrections page so they are logged. The index is published by LittPro Inc.