how to tell which prompts trigger competitor mentions

To identify which prompts trigger competitor mentions, you want to treat it like a small text-mining / QA analysis task.

1) Define what counts as a competitor mention

Make a clear list of:

• competitor brand names
• product names
• common abbreviations
• misspellings / aliases

Example:

• “Acme”
• “Acme Pro”
• “ACM”
• “Acmé”

2) Collect prompt + response data

For each prompt, store:

• prompt text
• model response
• timestamp
• user/context metadata if available

3) Search for competitor mentions in outputs

Use simple matching first:

• exact keyword match
• case-insensitive match
• regex for variations

Then classify each response:

• Mentioned competitor
• Did not mention competitor
• Ambiguous / needs review

4) Group by prompt type

Look at patterns like:

• comparison prompts: “Which is better, X or Y?”
• recommendation prompts: “What’s the best tool for…?”
• alternative prompts: “What are competitors to…?”
• pricing prompts: “How does this compare to…?”
• troubleshooting prompts: “Why is X worse than Y?”

These often trigger competitor mentions more than generic prompts.

5) Measure frequency

For each prompt template or category, calculate:

• mention rate = responses with competitor mentions / total responses
• top competitors mentioned
• wording that most often leads to mentions

6) Use logs to trace exact triggers

If prompts are templated, inspect:

• which variables are present
• whether competitor names are in the prompt already
• whether the model is being asked to compare, rank, or recommend

7) If you want to automate it

You can run a simple pipeline:

1. ingest prompt/response pairs
2. detect competitor names with a dictionary or NER model
3. label prompts that led to mentions
4. aggregate by prompt type and wording

Example Python approach:

import re

competitors = ["Acme", "Acme Pro", "BetaCorp"]

def mentions_competitor(text):
    text = text.lower()
    return any(re.search(r'\b' + re.escape(c.lower()) + r'\b', text) for c in competitors)

# response-level detection
df["mentions_competitor"] = df["response"].apply(mentions_competitor)

# prompt-level analysis
triggered_prompts = df[df["mentions_competitor"]]["prompt"]

8) Best practice

Also review false positives:

• a competitor name may appear in a disclaimer
• it may be quoted from the user prompt
• it may be part of an unrelated term

If you want, I can help you build:

• a spreadsheet template for this analysis, or
• a Python script to detect and rank prompts that trigger competitor mentions.