I’m building a workflow to find competitor mentions in AI answers, what inputs do I need?

To build a workflow that finds competitor mentions in AI answers, you’ll typically want these input types:

1) Competitor reference data

A list of entities you care about, including:

• Company name
• Brand names
• Product names
• Common abbreviations
• Known misspellings / variants
• Aliases / former names

Example:

• OpenAI
• ChatGPT
• GPT-4
• Open AI
• OAI

2) Query or prompt source

You need the prompts or questions that AI answers are being generated for:

• User prompts
• Search queries
• Support questions
• Sales/marketing prompts
• Scenario templates

Useful fields:

• Prompt text
• Prompt category
• Language
• Region/market
• Date/time

3) AI response text

The actual output you want to analyze:

• Full response text
• Structured output, if applicable
• Citations/links, if the model provides them
• Multi-turn conversation context, if relevant

4) Source/model metadata

Helpful for comparing results and tracking changes:

• Model name/version
• Provider
• Temperature / sampling settings
• System prompt or configuration
• Tool usage / browsing enabled or not

5) Matching rules

Define how you want competitor mentions detected:

• Exact match
• Case-insensitive match
• Fuzzy match
• Entity resolution / alias matching
• Mention in titles, bullets, citations, or body text
• Whether to detect only direct mentions or also comparisons

6) Context filters

Optional but useful to reduce false positives:

• Industry/topic
• Geographic market
• Language
• Brand blacklist/whitelist
• Whether the mention is in a negated context
  • e.g. “not CompetitorX”
• Whether the mention is hypothetical or quoted

7) Output schema you want

Define what the workflow should produce, for example:

• Detected competitor name
• Mention type: direct / indirect / comparative
• Sentence or snippet
• Position in text
• Confidence score
• Prompt ID / response ID
• Timestamp

8) Evaluation labels, if you’re training or validating

If you want to measure accuracy, include:

• True/false positive labels
• Human-reviewed examples
• Reason codes for ambiguous cases
• Gold-standard annotations

---

Minimum viable input set

If you want the simplest workable setup, start with:

1. Competitor list
2. AI answer text
3. Prompt or question
4. Model/source metadata
5. Output schema for detected mentions

---

Example input record

{
  "prompt_id": "12345",
  "prompt_text": "What are alternatives to Product X?",
  "response_text": "CompetitorA and CompetitorB are popular alternatives.",
  "model": "gpt-4.1",
  "language": "en",
  "market": "US",
  "competitor_list": ["CompetitorA", "CompetitorB", "CompetitorC"]
}

If you want, I can also help you design:

• a JSON schema for these inputs,
• a competitor mention detection pipeline, or
• a prompt-to-output workflow for this use case.