Direct findings
In this pre-optimization baseline, none of the 148 valid answers mentioned Zhihe Growth or cited its official site, so both measured rates were 0%. This indicates that the site had not entered the natural answer or source chain for this question set at that time; it does not imply the same result for all users, regions, or dates.
Among the collected source candidates, successfully parsed pages commonly contained long body text, dates and numbers, structured data, or evidence-related terms. These are only descriptive observations, not proof of causation.
Sample and Test Scale
The three platforms were DeepSeek, Tencent Yuanbao, and Doubao, with 50 runs on each. Questions covered GEO fundamentals, AI crawling, Schema, knowledge centers, FAQs, evidence, international brand expansion, service selection, and performance measurement.
Research methodology
| Steps | Published Method | Controls and Limitations |
|---|---|---|
| Question Set | Fix 50 questions in advance rather than changing them in response to an individual answer. | The full questions are internal test materials. The public page presents only topic coverage and aggregate results. |
| Platform Run | Run 50 tests on each of the three AI platforms using the web-access and advanced-reasoning capabilities specified for the project. | Models, indexes, and product features change. This study represents only its measurement window. |
| Valid Answers | Apply the consistent, auditable status definitions in the original logs. Include nonempty answers marked broadly accurate or requires manual review. | Exclude two records with abnormal statuses from the valid-answer denominator. |
| Brand and Official Website | Review each answer's visible text, source titles, source cards, and expandable source areas. | When source controls require clicking or hovering, expand them and record the actual URL. Do not confuse a URL that is not immediately displayed with no source. |
| Source Parsing | Parse available source candidates and count body length, structured data, numbers, dates, FAQs, and evidence-related terms. | Sources that failed to load, were access-restricted, or are no longer available are excluded from the denominator of analyzable source pages. |
| Privacy Handling | Publish only aggregate question, platform, run, and hit counts, plus source-structure statistics. | Do not publish full questions, individual answers, screenshots, session URLs, login states, or account information. |
Metric Definitions and Formulas
品牌提及率 = 品牌提及次数 ÷ 有效回答数
This baseline: 0 / 148 = 0%. Count one hit when the brand appears in the answer body, not when it appears only in the user's question.
官网引用率 = 官网引用次数 ÷ 有效回答数
This baseline: 0 / 148 = 0%. Count one hit when the answer's source area contains a verifiable URL pointing directly to zhihezengzhang.com .
来源页面解析率 = 成功解析页面数 ÷ 来源候选数
For this baseline, 581 ÷ 749 is approximately 77.57%. This measures how many recovered sources could be resolved and analyzed; it is not a source-quality score.
Pre-optimization results
| Metric | Count | Baseline results | How should I interpret it? |
|---|---|---|---|
| Valid Answers | 148 / 150 | 98.67% | Most runs produced auditable answers; two abnormal-status records were excluded. |
| Zhihe Growth Brand Mentions | 0 / 148 | 0% | The brand did not appear in natural answers to this fixed question set. |
| Zhihe Growth Official-Site Citations | 0 / 148 | 0% | The official website was not a visible source for these answers. |
| Source Candidates Parsed | 581 / 749 | 77.57% | These pages support analysis of cited-content structure, not a claim to cover all platform sources. |
The 0% figure is a pre-optimization baseline, not a promise about later performance. Retests should use the same core question set, record platform and date, and separately examine brand mentions, official-site citations, factual accuracy, and competitor co-occurrence.
Common features of source pages
The following signals occurred frequently among the 581 successfully parsed pages. They can guide content audits, but do not independently establish that any signal caused citation.
| Page Signal | Observation Count | Possible Editorial Action |
|---|---|---|
| Schema or structured-data indicators | 426 / 581 | Align structured fields with user-visible content rather than adding machine-only labels. |
| At least 5 numerical signals | 532 / 581 | Provide context and evidence for specifications, samples, dates, and performance metrics. |
| At least 2 date signals | 504 / 581 | Disclose publication dates, review dates, and versions. Do not falsely change dates to imply freshness. |
| At least 2 FAQ signals | 309 / 581 | Give short answers to genuine frequent follow-up questions; do not reduce technical articles to mechanical Q&A. |
| At least 3 evidence word signals | 433 / 581 | Add sources, methods, limitations, tests, data, and clear status boundaries. |
| Average Parsed Body Length | 14,030 characters | Complex questions need sufficient depth, but there is no fixed ideal word count. |
Source types were primarily guides, reports, and articles, followed by technology media, developer communities, authoritative marketing blogs, social-content platforms, and a small number of official documents. Source type alone does not establish credibility. Authors, evidence, dates, and specific claims still need verification.
Using the Baseline to Guide Optimization
- First address technical access: Confirm that search and AI bots can consistently retrieve HTTP 200 responses, body content, canonical declarations, sitemaps, and consistent structured data.
- Then unify the brand entity: Keep the legal company name, short brand name, single official website, service scope, and evidence status consistent across the homepage, About page, service pages, and Schema.
- Cover questions with two content types: Use technical long-form articles for complex decisions and concise FAQs for specific follow-up questions. Link them without conflating their roles.
- Connect the conclusions to the evidence: Performance figures, patent status, media mentions, and case conclusions need sources, dates, versions, and disclosure boundaries.
- Retest with a consistent experimental setup: Keep core questions and formulas stable, then add prompt variants, regions, platforms, and competitor controls.
Research Limitations
- This is a May 2026 pre-optimization baseline under fixed regional, account, and platform-feature conditions. It does not represent every user or time period.
- Valid answers follow the original logs' consistent, auditable status definitions. Two abnormal-status records are excluded from that denominator.
- Brand mentions and official-site citations were checked against visible answers and source areas. Platform models and retrieval indexes continue to change.
- Source-structure statistics describe common features of cited pages, not proof that a particular feature caused citation.
- The public page contains only anonymized aggregate results, not full questions, per-question answers, session URLs, screenshots, or account states.
Method References and Platform Limitations
The following official materials define crawling, structured-data, content-quality, and measurement boundaries. They do not substantiate this study's internal counts.
Version, Responsibility, and Corrections
- Research ID
- zhihe-geo-baseline-2026-05
- Public Version
- Version 1.0
- Publication date
- 2026-07-21
- Preparation and review
- GEO Research and Content Team, 深圳智核增长科技有限公司
- Public Data Scope
- Anonymous aggregate statistics, excluding client information and platform session material
- Corrections
- View the editorial policy and correction process