# KnowFlow GEO Knowledge Base

> KnowFlow maintains practical guidance for monitoring and improving how brands
> are understood, cited, and recommended by generative AI products. The main
> language is Simplified Chinese. Last updated: 2026-08-31.

Canonical website: https://knowflow.work/
GEO monitoring product: https://geo.knowflow.work/

## What GEO optimizes

Generative Engine Optimization focuses on four observable outcomes:

1. Discovery: whether a brand enters the model's candidate set.
2. Understanding: whether product facts and positioning are accurate.
3. Trust: whether the answer cites the brand or reliable third-party evidence.
4. Selection: whether the brand is recommended, ranked, or preferred.

Recommended metrics include mention rate, positive mention rate,
recommendation rate, first-choice rate, citation rate, factual accuracy, and
share of voice against a fixed competitor set. Every metric should retain its
sample size, model, region, language, connectivity mode, and test time.

## First audit workflow

1. Freeze verifiable brand facts and their public URLs.
2. Build 30-50 questions across discovery, scenario, comparison, brand facts,
   buying decisions, and risk concerns.
3. Choose 3-5 models actually used by the target customers.
4. Record model version, region, language, login state, and web access mode.
5. Run each important question multiple times and keep the raw answers.
6. Score mentions, recommendations, position, citations, sentiment, and errors.
7. Prioritize high-intent gaps, make attributable changes, and retest using the
   same question set.

## Model coverage

- Global conversation and research: ChatGPT and Claude.
- Global search and citation-oriented answers: Gemini and Perplexity.
- Chinese content and consumer scenarios: Doubao and Kimi.
- Chinese office and ecosystem users: Qwen and Tencent Yuanbao.

Model outputs must not be merged into one opaque score. Diagnose results by
model, market, question cluster, and competitor.

## Content and technical guidance

- Give each page one primary question and place the answer early.
- Make important paragraphs self-contained with an explicit subject, claim,
  conditions, date, unit, and evidence.
- Publish accurate product facts, implementation guides, comparison criteria,
  case studies, limitations, and real FAQ answers.
- Keep Organization, Product, SoftwareApplication, Article, and FAQ structured
  data consistent with visible page content.
- Verify HTTP 200 responses, canonical URLs, crawl access, server-visible main
  content, mobile parity, and updated pricing or policy dates.
- Build third-party evidence through verifiable customer cases, transparent
  research, industry relationships, and accurate ecosystem listings. Do not
  fabricate reviews, customer logos, or bulk low-quality articles.

## Market guidance

China and global programs should use separate question sets and fact sources.
Record account and connectivity conditions for Chinese models. For global
markets, localize pricing, regulation, examples, competitors, language, and
third-party evidence instead of translating one generic page.

## Human-readable guides

- https://knowflow.work/#/article/overview
- https://knowflow.work/#/article/quickstart
- https://knowflow.work/#/article/measurement
- https://knowflow.work/#/article/prompt-set
- https://knowflow.work/#/article/model-coverage
- https://knowflow.work/#/article/brand-visibility
- https://knowflow.work/#/article/content-rewrite
- https://knowflow.work/#/article/technical-geo
- https://knowflow.work/#/article/citations
- https://knowflow.work/#/article/china
- https://knowflow.work/#/article/global