# How to Measure Your Brand's AI Visibility with a Spreadsheet > You do not need expensive software to track AI recommendations. A three-tier prompt framework and a simple spreadsheet can score your brand visibility in half a day. Source: https://www.starbox.work/en/insights/geo-guide/2026-10-21-manual-geo-testing-sheet/ (Starbox AI Insights, updated 2026-10-21) You have probably lived through this scenario: your executive team reads an industry update over breakfast and sends an urgent message asking where the company ranks inside major conversational engines and whether AI is actively recommending you to buyers. If you do not have tens of thousands of dollars in software budget for dedicated AI monitoring suites, can an in-house marketer or operator handle this from their desk? Yes, easily. Monitoring software simply sends batch [prompts](/en/glossary/prompt/) to conversational engines and extracts the output text. If you know how to build standardized test prompts and use a clear quantitative scoring model, a standard spreadsheet will give you accurate, actionable data. --- ## Myth: You Need Expensive Software to Monitor AI Recommendations Many marketing teams assume AI monitoring requires custom API pipelines and automated dashboards to reveal real market standing. That assumption misunderstands how conversational engines work. Most commercial monitoring platforms simply simulate static prompts and scrape brand entity mentions from the generated text. Automation makes sense for tracking daily volatility across hundreds of long-tail keywords. But when you need to know whether an AI engine recommends your brand for high-intent commercial decisions, manual testing is often more reliable. Automated scrapers rely on string matching. They struggle to parse the actual tone and sentiment behind an AI response. A model might mention your brand name merely as an example of what to avoid or as a filler option at the bottom of a list. By structuring your test manually, you avoid software fees and gain direct insight into the specific citations and ecosystem sources each engine leans on. --- ## The Mystery Shopper Method: 3 Prompt Tiers That Mirror Real Buyers Conversational engines are not traditional search engines. Typing an isolated category keyword will not show you where you stand. Buyers turn to large language models with specific technical requirements, operational constraints, and pain points. I recommend adapting the retail "mystery shopper" method by dividing test prompts into three distinct tiers: 1. **Broad Category Discovery (Baseline Recall)**: Tests whether your brand exists in the model's base training corpus and index. For example: *"What are the leading industrial dehumidifier manufacturers in 2026? Recommend five primary suppliers and explain why."* 2. **Scenario and Pain-Point Requirements (Vertical Relevance)**: Tests whether the model connects your product to specific operational constraints. For example: *"A pharmaceutical cleanroom needs relative humidity kept below 30%. What dehumidification equipment brands are most reliable? Compare them by technical specifications."* 3. **Direct Competitor Comparison (Substitution Defense)**: Tests how the model summarizes your strengths and weaknesses against established category leaders. For example: *"Compare [Competitor A] and [Our Brand] in terms of after-sales response time and custom engineering capabilities."* Keep every prompt neutral. Avoid leading phrasing like "Why is Brand X the best?" Instead, prompt the model to answer as an objective enterprise procurement consultant. --- ## How to Turn an AI Response into a Quantitative Score AI outputs long-form text. To turn that text into clear percentages for executive reporting, you need a standardized rubric. You can evaluate every engine response across three dimensions: - **Mention Score (Threshold, 30% Weight)**: Did the engine explicitly name your full brand or recognized trading name? If not mentioned, score 0. If mentioned, score 100. - **Rank Score (Placement Priority, 40% Weight)**: Conversational engines present options in order of perceived relevance. If your brand ranks #1, score 100. Rank #2 scores 80. Rank #3 scores 60. Rank #4 or lower scores 40. An unmentioned brand scores 0. - **Sentiment Score (Recommendation Stance, 30% Weight)**: How the engine frames your capabilities. If featured as the top recommendation with clear supporting strengths, score 100. If listed neutrally as a peer option, score 70. If mentioned with clear caveats or spec deficits, score 30. If flagged as an outright negative or cautionary example, score 0. Use this formula for each test prompt: $$\text{Single Prompt Score} = (\text{Mention Score} \times 0.3) + (\text{Rank Score} \times 0.4) + (\text{Sentiment Score} \times 0.3)$$ Take the arithmetic mean of all prompt scores on a given platform to determine your overall brand AI visibility score for that engine. --- ## The Step-by-Step Testing Workflow: From Clean Sessions to Root-Cause Analysis To get clean data, never run tests from your daily personal or business account. Conversational engines adjust their outputs based on your historical chat logs, which produces false positives. Follow this workflow: 1. **Environment Setup and Session Isolation**: Open an incognito browser window or log into a fresh account with zero chat history. Click "New Chat" before running each individual prompt so context from previous queries does not bleed into the next test. 2. **[Multi-Platform Distribution](/en/glossary/multi-platform/)**: In China, prioritize Doubao (ByteDance's AI assistant), DeepSeek, Tencent Yuanbao, Kimi (Moonshot AI), and Baidu ERNIE Bot. If you manage international operations, include ChatGPT and Perplexity for cross-border comparison. Each engine relies on different source ecosystems: Tencent Yuanbao draws heavily from WeChat Official Accounts, ERNIE Bot relies on Baidu's search index and Baijiahao (Baidu's content platform), while Kimi and DeepSeek frequently surface technical debates from Zhihu (China's largest Q&A platform). 3. **Log Footnote Citations**: Do not stop at the visible text. Copy every reference link and [citation source](/en/glossary/citation-source/) listed at the bottom of the response. Track exactly which article, wiki entry, or forum thread informed the model's recommendation. If a competitor was favored over your brand, capture their sources as well. 4. **Root-Cause Analysis**: Highlight prompt rows with unusually low or high scores. Compare your brand's footprint against competitor citations to see which platforms need immediate content distribution. --- ## Practical Deliverable: Manual AI Visibility Tracking and Reporting Template You can copy this structure directly into Excel or Lark Base to run tests with your team: ### 1. Manual Testing Log and Scoring Matrix | Test ID | Prompt Tier | Prompt | Engine | Mention (0/100) | Rank (0-100) | Sentiment (0-100) | Total Score | Cited Source Type (Zhihu/Wiki/WeChat/Other) | Competitor Mentions | |---|---|---|---|---|---|---|---|---|---| | T01 | Category | What are the leading industrial dehumidifier manufacturers in 2026? Recommend 5. | Doubao | 100 | 60 (#3) | 70 (Neutral list) | 75 | Trade publications, official press releases | Competitor A (#1), Competitor B (#2) | | T02 | Scenario | Cleanroom humidity needs to stay below 30%. Recommend reliable equipment. | Doubao | 0 | 0 (Unlisted) | 0 (Unlisted) | 0 | B2B directory spec tables | Competitor A (#1), Competitor C (#2) | | T03 | Comparison | Compare [Competitor A] and [Our Brand] on dehumidifier service and specs. | Doubao | 100 | 80 (Tied) | 70 (Neutral comparison) | 81 | Tech review portals, Zhihu Q&A | Competitor A | | T04 | Category | What are the leading industrial dehumidifier manufacturers in 2026? Recommend 5. | Tencent Yuanbao | 100 | 100 (#1) | 100 (Top pick) | 100 | WeChat Official Account in-depth article | Competitor B (#2) | *(Note: For T01, the score calculation is: $100 \times 0.3 + 60 \times 0.4 + 70 \times 0.3 = 30 + 24 + 21 = 75$.)* ### 2. Executive Reporting Brief Template When sharing results with leadership, avoid sharing pages of unedited AI transcripts. Synthesize the findings into this format: ```text Executive Brief: Brand AI Recommendation Baseline Audit Period: October 2026 Platforms Tested: Doubao, DeepSeek, Tencent Yuanbao, Kimi 1. Executive Summary - Overall Brand Score: 62/100 (Pass tier; solid mid-market visibility). - Channel Divergence: Strongest on Tencent Yuanbao (85/100), moderate on Doubao (60/100), lowest on DeepSeek (42/100). - Scenario Gaps: 75% mention rate on broad category prompts, dropping to 25% on specific industrial use cases. Engines lack structured technical specifications for our product line. 2. Competitive Benchmarking Competitor A captured a top-two position in 80% of test runs. Their visibility is driven by detailed technical articles on Zhihu and downloadable industry white papers. 3. Action Plan To close the citation gap on DeepSeek and Kimi, we will publish technical spec breakdowns and teardowns on Zhihu next month, accompanied by structured case studies covering our core application scenarios.