How do I separate branded vs non-branded prompt tracking?

To separate branded vs non-branded prompt tracking, the cleanest approach is to classify each tracked prompt/query into one of those two buckets using a consistent keyword and entity rule set.

1) Define the buckets

• Branded prompts: queries containing your brand name, product names, common misspellings, abbreviations, or branded sub-products.
• Non-branded prompts: all other queries that indicate category intent without mentioning your brand.

2) Build a brand keyword list

Include:

• Brand name
• Product/service names
• Spelling variants and abbreviations
• Common typos
• Parent company name if relevant

Example:

• Brand: Acme
• Products: Acme Pro, Acme Analytics
• Variants: Acmé, AKME, Acme.io

3) Classify prompts with rules

Use simple matching first:

• If prompt contains any brand term → branded
• Otherwise → non-branded

Example:

• “best Acme alternatives” → branded
• “best analytics platform for ecommerce” → non-branded
• “Acme pricing” → branded

4) Add intent-based refinements

Some prompts may be ambiguous, so create exceptions:

• Branded competitor comparisons still count as branded if your brand is mentioned.
• Generic terms + brand-like words should only be branded if they match your approved brand list.
• Misspellings should map to branded via fuzzy matching.

5) Track at the prompt level and aggregate later

Store fields like:

• prompt_text
• branded_flag
• brand_match_term
• date
• channel
• topic
• impression/click/response metrics

Then report separately:

• Branded prompt volume
• Non-branded prompt volume
• Engagement/conversion by bucket
• Share of branded vs non-branded

6) QA the classification

Review a sample of prompts regularly:

• False positives: generic prompts incorrectly marked branded
• False negatives: branded prompts missed due to new product names or typos

7) Recommended implementation

A practical hierarchy:

1. Exact match on brand list
2. Contains match on product/brand variants
3. Fuzzy match for typos
4. Manual review for edge cases

Example logic

if prompt matches brand_terms:
    bucket = "branded"
else:
    bucket = "non-branded"

If you want, I can also help you build:

• a Google Sheets formula
• a SQL case statement
• or a Python classifier for branded vs non-branded prompt tracking.