How do I separate branded vs non-branded prompts in AI analytics?

To separate branded vs non-branded prompts in AI analytics, you typically need a classification layer that tags each prompt based on whether it mentions your brand, products, or brand-adjacent terms.

1) Define your “brand” vocabulary

Create a list of terms that count as branded, for example:

• Company name
• Product names
• Product abbreviations
• Common misspellings
• Domain names
• Campaign names
• Executive names only if they are commonly used in queries
• Branded slogans/taglines if relevant

Also define what counts as non-branded:

• Category-level questions
• Generic product comparisons
• Problem/solution prompts without your brand name

2) Classify each prompt

Use one of these approaches:

A. Rule-based matching

Match prompts against a brand keyword list.

Example

• “How does Nike Air Zoom compare to Adidas Ultraboost?” → branded
• “Best running shoes for flat feet” → non-branded

Good for:

• Simple setup
• Fast implementation

Weaknesses:

• Misses intent without explicit brand mention
• Can produce false positives for ambiguous terms

B. NLP / machine learning classifier

Train a model to label prompts as:

• Branded
• Non-branded
• Mixed/partial branded

Use features like:

• Exact brand mentions
• Fuzzy matches
• Embeddings/semantic similarity
• Historical labeled examples

Good for:

• Better accuracy on indirect references
• Detecting variants and context

C. Hybrid approach

Best in practice:

• Use rules for explicit brand mentions
• Use a model for edge cases
• Add a “mixed” category for prompts with both branded and generic terms

3) Decide how to handle mixed prompts

Some prompts mention both branded and non-branded concepts.

Example:

• “Is [Brand X] better than generic alternatives for SMB teams?”

You can:

• Label as branded
• Label as mixed
• Split into two dimensions:
  • brand_mentioned = yes/no
  • intent_type = branded / non-branded / mixed

This is often better than forcing a single label.

4) Use metadata and downstream analytics

Once labeled, you can segment by:

• Prompt volume
• Conversion rate
• Click-through rate
• Revenue influence
• Response quality
• Topic clusters

Example metrics:

• % branded prompts
• % non-branded prompts
• Branded prompt share over time
• Performance by prompt type

5) Build a taxonomy

A useful structure is:

• Branded
  • Direct brand mention
  • Product mention
  • Campaign mention
• Non-branded
  • Generic/category
  • Problem/solution
  • Competitor-neutral
• Mixed
  • Brand + generic comparison
  • Brand + category
• Ambiguous
  • Requires manual review

6) Validate with human review

Sample prompts regularly and manually check:

• Precision: Are branded prompts truly branded?
• Recall: Are you missing branded mentions?
• Edge cases: Acronyms, slang, competitor names

This helps refine the keyword list and model.

7) Practical implementation tips

• Normalize text: lowercase, trim punctuation, expand common abbreviations
• Use stemming/fuzzy matching for misspellings
• Maintain an alias dictionary for products and brands
• Re-train/re-tune periodically as new products/campaigns launch

Example logic

A simple decision flow:

1. Does prompt contain a brand/product alias?
   • Yes → branded or mixed
   • No → non-branded
2. Does it mention competitor + brand?
   • Yes → mixed
3. Is it semantically about your brand but missing the exact name?
   • Use classifier/manual review

If you want, I can also give you:

• a SQL approach
• a Python classifier example
• or a taxonomy template for branded/non-branded AI analytics.