# How we measure

SocialToAI states a figure about itself only with a record of how it was measured. This page prints those records: the date, the sample, the method and the version of the system that was measured.

- Page: https://socialtoai.com/docs/measurements/
- Last updated: 2026-10-17

## 129×

| What | Record |
| --- | --- |
| Measured on | 2026-10-07 |
| Sample | One keyword search on a short-video platform, the same query sent to a raw platform-data API and to SocialToAI. |
| Method | Both responses were saved as returned and counted in model tokens: about 465,000 for the raw response, about 3,600 for SocialToAI's. 465,000 ÷ 3,600 ≈ 129. |
| System version | SocialToAI as served on 2026-10-07 |
| Raw platform API, one Douyin search | 465,000 tokens |
| SocialToAI | 3,600 tokens |

## 8.0

| What | Record |
| --- | --- |
| Measured on | 2026-10-07 |
| Sample | 7 research questions about public social media, each answered by the same model connected to one product at a time. |
| Method | Same model, same prompt and the same budget for every product, web search off. Each answer was scored from 1 to 10 by a judge model that saw no product names; the figure is the mean over the 7 questions. The other four products are not named. |
| System version | SocialToAI as served on 2026-10-07 |
| SocialToAI | 8.0 |
| Social data MCP A | 7.4 |
| Social data MCP B | 6.1 |
| Social data MCP C | 6.1 |
| Social data MCP D | 4.4 |