AI Competitor Research: Prompts That Require Proof

Public evidence can compare an offer. It cannot reveal a competitor's bank account.

I begin with a small set of businesses and a specific buying decision. Use the channel framework to define the question before collecting screenshots.

Step 1: Fix the comparison

Name the market, customer need and businesses. Choose comparable products or services rather than comparing entire companies with different scopes.

Write the observation date. Public prices, stock and messages can change; the matrix needs a point in time.

Step 2: Collect public records

Use a search-enabled research tool, such as the capabilities documented for OpenAI web search or Gemini grounding.

For [businesses] in [market], find public product or service pages relevant to [customer need]. Return the exact URL, observation date, quoted evidence and stated offer. Separate the business's claim from independently observed facts. Leave unavailable information unknown. Do not seek private accounts or documents.

Open each decisive source before adding its finding to the matrix.

Step 3: Build the same fields for every business

Field	Acceptable evidence
Offer and scope	Exact public page and stated terms
Price	Visible amount, currency and conditions
Buying path	Observed steps, without submitting a real order
Proof	Published case, review or certification; label its origin
Unknowns	Information the public evidence does not establish

Convert these verified records into the matrix above. Every factual cell needs its source URL. Use unknown for missing evidence. Do not infer revenue, profit, conversion rate or private ad spend from rankings, reviews, traffic estimates or visible ads.

Step 4: Compare the decision, not the adjectives

Use a reasoning-capable model to identify differences in scope, terms and the buying path. Supply the matrix; do not ask it to reconstruct the evidence from memory.

Identify differences relevant to [buying decision]. Cite the matrix row for each observation. Distinguish observed differences, company claims and hypotheses. Propose one test on our own site for each useful hypothesis.

Step 5: Challenge the comparison

Ask a separate review pass to find mismatched products, expired pages, currencies, missing conditions and repeated sources. Another model's agreement does not validate the claim.

Find comparisons that would change if scope, date or terms were corrected. Do not add facts. List the exact source checks needed before using the conclusion.

Step 6: Choose a test you can measure

Select one change to your own explanation, offer or inquiry path. Record its owner, baseline and review date through business intelligence.

The output is a testable decision supported by public records. Private numbers stay unknown.