HHelp4GEO

Measurement methodology

How Help4GEO measures visibility across China AI platforms.

Help4GEO compares repeatable Chinese-language question samples across Doubao, Tencent Yuanbao, Qwen, and DeepSeek. Every result must retain its platform, question type, sample period, model context, and classification rule so a metric can be interpreted without implying a permanent platform preference.

Method version 1.0 · Last reviewed August 22, 2026 · Corrections: zhangluease@gmail.com

Metric definitions

Metrics are calculated within a documented sample. They are not global scores for an entire platform.

Brand mention rate
The share of sampled answers that explicitly name the tracked brand under the defined question set and platform context.
Recommendation rate
The share of sampled answers that present the brand as a recommendation, option, or positive fit under a documented classification rule.
Target-source citation rate
The share of sampled answers that cite or link to a customer-defined target source, measured separately from general source-domain coverage.
Competitor context
The competing brands, products, or alternatives that appear in the same sampled answer and the context in which they are presented.

Sample design

A comparable monitoring run should freeze the brand and product context, question set, question language, execution window, answer environment, and classification rules before interpreting change.

  • Question variants are retained because wording can materially change an answer.
  • Model or product version information is recorded when the platform exposes it.
  • Login state, region, time, and accessible citations are treated as context, not hidden assumptions.
  • Missing or blocked answers remain visible as sample limitations instead of being silently discarded.

Cross-platform comparison

The same question can be compared across platforms, but the answer environments are not identical experiments. Help4GEO keeps platform results separate and uses a shared metric definition only where the underlying evidence is comparable.

A difference between Doubao, Yuanbao, Qwen, and DeepSeek is an observation for the sampled period. It is not proof of a fixed algorithm preference or a guaranteed future outcome.

Evidence boundaries

  • A brand mention is not automatically a recommendation.
  • A source domain appearing in an answer is not automatically a causal ranking factor.
  • A target-source citation gap identifies missing observed coverage, not a guarantee that publishing a page will change the answer.
  • Methodology previews and sample reports are not customer results unless the customer, period, baseline, intervention, and outcome are explicitly documented.

From evidence to action

Help4GEO groups source and content gaps by buyer intent, then prepares a brief or workflow for human review. Monitoring, content generation, and publishing remain separate actions so evidence is not confused with execution.

See the complete product workflow or review methodology previews.