Prompt
How do I set up prompt-level brand tracking?
Latest observation
Prompt-level brand tracking usually means measuring how often and how well your brand appears in AI answers for prompts that matter to your business. The goal is to track visibility, sentiment, share of voice, and which prompts trigger your brand vs. competitors.
Here’s a practical setup.
1) Define what you want to measure
Pick 3–5 core metrics:
- Brand mention rate: % of prompts where your brand is mentioned
- Share of voice: your mentions vs. competitors across the same prompt set
- Ranking / ordering: whether you appear first, in top 3, or not at all
- Sentiment / framing: positive, neutral, negative, or mixed
- Recommendation rate: how often the model recommends your brand
- Citation rate: whether answers include sources that mention you
- Intent coverage: prompts by funnel stage or use case where you appear
2) Build a prompt set
Create a list of prompts representative of your audience’s actual queries.
Organize them by category:
- Awareness: “What is the best CRM for small teams?”
- Comparison: “HubSpot vs Salesforce for startups”
- Transactional: “Best email marketing tool under $50”
- Problem/solution: “How do I automate onboarding emails?”
- Brand-specific: “Is [your brand] good for enterprise?”
Best practices:
- Use 50–200 prompts to start
- Include competitor prompts
- Include category prompts where users may not mention your brand
- Keep prompts stable for longitudinal tracking
- Add tags: funnel stage, product line, persona, geography, language
3) Choose the systems to test
Track across the models and interfaces that matter to your users:
- ChatGPT
- Google Gemini
- Claude
- Perplexity
- Microsoft Copilot
- Search AI features if relevant
If your audience uses specific versions or regions, note those too.
4) Standardize the test conditions
Prompt-level tracking only works if runs are comparable.
Control:
- Model name/version
- Temperature / randomness settings if available
- Time of day / location if relevant
- Logged-in vs logged-out state
- Whether browsing is on
- System prompt / custom instructions if relevant
- Number of retries per prompt
If the interface is non-deterministic, run each prompt 3–10 times and average the results.
5) Collect and structure outputs
For each prompt, store:
- Prompt text
- Date/time
- Model / interface
- Raw response
- Brands mentioned
- Brand ranking/order
- Competitor mentions
- Sentiment
- Source links/citations
- Notes on caveats or hallucinations
A simple spreadsheet works at first. Better yet, use a database or dashboard.
6) Create a scoring rubric
You need consistent rules for evaluating answers.
Example rubric:
- 0 = brand not mentioned
- 1 = brand mentioned, but not recommended
- 2 = brand recommended among others
- 3 = brand recommended first / strongly endorsed
Sentiment rubric:
- Positive
- Neutral
- Negative
- Mixed
You can score manually at first, then automate with an LLM-assisted classifier plus spot checks.
7) Add competitor benchmarks
Tracking your brand alone is not enough.
For each prompt, compare:
- Which brands appear
- Which brand is first
- Which brand gets the strongest framing
- Whether competitors are consistently recommended over you
- Whether the model’s answer reflects outdated or inaccurate positioning
This helps you identify where your brand is underrepresented.
8) Segment by prompt type
Analyze results by:
- Category
- Persona
- Funnel stage
- Geography/language
- Product line
- Competitor set
This often reveals patterns like:
- Strong awareness but weak comparison performance
- Great performance on niche use cases but not general category prompts
- Better visibility in one model than another
9) Set a refresh cadence
Models and search behavior change often.
A good cadence:
- Weekly for high-priority categories
- Monthly for broader coverage
- Quarterly for strategic review and prompt set updates
Also rerun after:
- Major product launches
- PR events
- Website/content changes
- Model updates
10) Turn insights into action
Use the results to improve your presence in AI answers:
- Create content that matches high-value prompts
- Strengthen category pages and comparison pages
- Improve factual consistency across your site and third-party sources
- Publish clear product positioning and use cases
- Earn mentions in reputable sources that LLMs may rely on
- Fix misinformation or outdated references
Simple starter workflow
If you want a quick version:
- List 50 prompts relevant to your market
- Run them in 3–5 AI systems
- Save responses in a spreadsheet
- Label brand mentions, ranking, and sentiment
- Compare against 3–5 competitors
- Review monthly and update content strategy
Recommended spreadsheet columns
- Prompt ID
- Prompt text
- Category
- Persona
- Competitor set
- Model
- Date
- Raw answer
- Your brand mentioned? (Y/N)
- Brand rank
- Sentiment
- Competitors mentioned
- Citation/source notes
- Reviewer notes
Common pitfalls
- Testing only branded prompts
- Using too few prompts
- Not controlling for model variability
- Ignoring competitor tracking
- Over-relying on automated scoring without human review
- Treating one-time results as stable
If you want, I can also give you:
- a ready-to-use prompt tracking template,
- a dashboard schema, or
- a Python script to automate prompt-level brand tracking.