A Gemini competitor-analysis prompt should produce a reviewable evidence pack, not a confident paragraph assembled from unverified web fragments. The hard part is not asking for strengths and weaknesses. It is controlling the comparison set, date, source quality, claim scope, missing evidence and the business decision that follows.

This guide introduces COMPARE-7, a source-ledger prompt and human-review workflow. It addresses the narrow search intent “Gemini prompt for competitor analysis” without replacing primary research, legal review or accountable strategy decisions.

Direct answer

Give Gemini a defined decision, competitor set, geography, date, evidence hierarchy, comparison fields and output schema. Require a source for every factual claim, mark missing evidence as unknown, separate fact from inference and finish with a human claim test. Do not submit confidential strategy, customer data or unapproved copyrighted material.

COMPARE-7

Step Control Output
Context Define decision, audience, market, geography and date Bounded brief
Objects Name the companies/products and inclusion rule Comparison set
Materials Supply approved sources or require grounded retrieval Source ledger
Parameters Define fields, units and comparison date Comparable schema
Assertions Link each factual claim to evidence and scope Claim register
Risks Record unknowns, contradictions and stale sources Uncertainty log
Executive review Test implications and choose next evidence action Human decision memo

Copyable Gemini prompt

<role>
You are an evidence-bounded market analyst. Do not invent facts.
</role>

<decision>
We need to decide: [specific decision].
Audience: [decision owner].
Market/geography: [boundary].
Evidence cutoff: [date].
</decision>

<comparison_set>
Compare only: [Company/Product A], [B], [C].
Inclusion rule: [why these belong together].
</comparison_set>

<source_rules>
Use supplied primary sources first: official product pages, pricing pages,
terms, filings, documentation and dated announcements.
If grounded web retrieval is available, cite the direct page for every fact.
Never treat a search snippet, model memory or unattributed summary as evidence.
If evidence is missing, write UNKNOWN.
</source_rules>

<fields>
For each object compare: target customer, stated problem, core capability,
price/metric/date, delivery model, integrations, proof, limitations and risks.
Use the same definition and unit in every row.
</fields>

<output>
1. Source ledger: source, owner, URL, publication/update date, access date.
2. Comparison table with evidence IDs in every factual cell.
3. Claim register: claim, evidence ID, fact/inference, scope, confidence.
4. Contradictions and unknowns.
5. Three decision implications and the next validation action for each.
</output>

<constraints>
Quote minimally. Separate facts from inference. Do not estimate missing prices,
market shares, customer counts or performance. Do not recommend a decision
when material evidence is missing.
</constraints>

Google's current Gemini prompt-design guidance recommends clear, specific instructions, explicit constraints and response formats, adding relevant context, using consistent structures and iterating based on observed results. It also notes that grounding can help with current or obscure facts. Those techniques improve control; they do not prove that a cited page supports a claim.

Build the source ledger first

Use one row per source:

Evidence ID Organization Direct URL Source type Published/updated Accessed Scope note
E-01 Company A official URL Pricing page 2026-08-20 2026-09-07 US self-service tier

The example is fictional. Prefer first-party documents for product capability and price, regulatory records for formal status, and independent methods for customer or performance claims. A competitor's marketing page is evidence of what it claims, not necessarily of what the product delivers.

Make comparison fields commensurable

Price per month, price per user and usage-based charges are not directly comparable. Neither are “AI assistant”, “workflow automation” and “managed service”. Define the unit, customer segment, contract term, included volume, tax and access date before asking Gemini to compare.

When a field cannot be normalized, keep the original fact and mark the limitation instead of forcing a ranking.

Run the claim test

For every material statement, ask:

  1. Is this a fact, inference or recommendation?
  2. Does the cited page contain the claim?
  3. Does it apply to the named product, market, tier and date?
  4. Is the source primary and current enough for the decision?
  5. Does another source contradict it?
  6. What is missing?
  7. Who is accountable for the final use?

Delete or rewrite any claim that fails. Never let Gemini convert “not found” into “does not exist”.

Worked fictional mini-comparison

Assume three suppliers publish monthly prices of $30 per user, $500 for a ten-seat team and $0.08 per processed item. The numbers cannot be ranked until usage is normalized. For a ten-person team processing 12,000 items:

Supplier Published basis Normalized monthly cost Limitation
A $30 × 10 users $300 Excludes usage add-ons
B Team plan $500 Confirm seat and volume limits
C $0.08 × 12,000 items $960 Excludes platform minimums

This is a fictional calculation, not market evidence. It demonstrates why the prompt must preserve units, assumptions and missing terms.

Convert analysis into a decision memo

The final output should not be a winner score unless the decision criteria justify one. Use four sections:

  • supported facts: what the source ledger proves;
  • bounded inference: what those facts may imply for this decision;
  • material unknowns: what could change the conclusion;
  • next action: demo, interview, test, legal review, pricing confirmation or no-go.

For a repeatable market-research operating sequence, see From Decision Question to Insight Report. That workflow places prompts inside sampling, evidence collection, analysis and accountable communication.

Learning pathway

MTF Institute's Professional Certificate in Market Research programme covers research design, sampling, interviews, surveys, analysis and decision-ready insight reporting. Use COMPARE-7 as one controlled desk-research exercise, not as a substitute for direct customer evidence. The programme is professional education, not an academic degree, and does not guarantee a business outcome.

Sources