Practical guide
How to create a competitor brief with AI
A useful competitor brief helps a named decision maker understand where alternatives genuinely differ, which claims are supported, and what remains unknown. It is not a collection of slogans or a prompt that asks an AI assistant to rank companies from memory. Build it from dated primary evidence, comparable units, and explicit decision criteria. AI can accelerate source inventories, extraction,.
A useful competitor brief helps a named decision maker understand where alternatives genuinely differ, which claims are supported, and what remains unknown. It is not a collection of slogans or a prompt that asks an AI assistant to rank companies from memory. Build it from dated primary evidence, comparable units, and explicit decision criteria. AI can accelerate source inventories, extraction, comparison, and contradiction checks. A human author remains responsible for competitor selection, materiality, interpretation, legal review, and every claim that reaches a customer or decision maker.
Define the decision before the competitors
Write the decision, audience, market boundary, product or service category, geography, customer situation, and evidence cutoff. A sales team choosing how to respond to a named procurement need requires a different brief from a product team testing a roadmap assumption.
Set the comparison unit. Compare products that meet the same need, editions that are actually available to the relevant customer, and features under equivalent plan, permission, and configuration conditions. UK rules for business comparative advertising require comparisons to be objective, material, relevant, verifiable, and representative. The current eCFR business-description rule also points public companies toward material market demand and competitive conditions. These sources support disciplined comparison, not a claim that one universal template satisfies every jurisdiction.
Build a TRACE brief
Use five linked records: Target, Rival set, Assertions, Comparison, and Expiry.
### Target
State what the reader must decide and which evidence could change that decision. Record the use: internal planning, sales enablement, procurement, investor analysis, or public marketing. Public comparative claims need specialist review appropriate to the jurisdiction and channel.
### Rival set
Define inclusion and exclusion rules before collecting evidence. Include direct alternatives, the current manual process, internal build, and “do nothing” when they are credible options. Do not select only competitors that make the preferred product look stronger.
For each rival, record the legal entity, exact product, edition, customer segment, geography, date checked, and why it belongs in the set. Separate a company from one of its products and a product from a configured deployment.
### Assertions
Create an assertion ledger instead of free-form notes. Each row contains:
Prefer current first-party documentation, official filings, regulator records, price and contract documents, and direct product observations. Treat review sites, search snippets, sales claims, and AI-generated summaries as leads until verified. A source can be authentic and still fail to support the proposed claim.
### Comparison
Compare on criteria derived from the decision, not a generic feature inventory. Define how each criterion is tested and what counts as present, partial, absent, unknown, or not comparable. Keep facts separate from interpretation.
| Field | Question | | --- | --- | | Customer and task | Does the alternative serve the same user and decision? | | Evidence | What current source supports the entry? | | Conditions | Which plan, permission, integration, region, or configuration applies? | | Limitation | What does the source not establish? | | Consequence | Why could this difference change the target decision? | | Verification | Who checked it, how, and when? |
Do not turn unknown into “no.” A missing public feature page does not establish that a capability is unavailable. Contact the provider or mark the point unresolved.
### Expiry
Assign a review trigger to every material assertion: a fixed date, release, pricing change, contract renewal, new filing, or procurement event. The brief header shows the evidence cutoff and owner. Expired claims remain visible as history but cannot silently flow into a new proposal.
- the precise assertion;
- its decision relevance;
- the source URL or retained document;
- source owner and publication date;
- access date and applicable product version;
- whether the source is first-party, regulatory, customer-provided, or analyst interpretation;
- confidence and unresolved contradiction; and
- the author responsible for verification.
Use AI at bounded steps
Give the assistant the decision definition, authorised source packet, extraction schema, and output constraints. Suitable tasks include normalising product names, extracting dated assertions, identifying mismatched comparison units, finding contradictions, and drafting questions for unresolved points.
Require source-linked output. Reject invented quotations, inferred plan availability, unsupported market-share figures, composite customer stories, and conclusions that exceed the evidence. Test calculations outside the generated prose and retain rejected suggestions when they could reappear later.
Worked artifact: compare incident-summary options
A synthetic operations team needs an incident-summary workflow. Its alternatives are the existing manual process, an add-on in the current service platform, and a separate AI workspace. The decision is whether to run a controlled pilot, not which vendor is “best.”
The first AI draft says the separate workspace is fastest, the platform add-on lacks source links, and the manual process is safest. The TRACE review rejects all three statements: “fastest” has no comparable timing test, a missing public page does not prove source links are absent, and “safest” has no defined risk measure.
The repaired matrix reads:
| Criterion | Manual process | Platform add-on | Separate workspace | | --- | --- | --- | --- | | Source trace | Current template requires links, sample check pending | Product documentation unclear; provider question open | Admin guide documents citations for the tested edition | | Data route | Existing approved system | Existing system, feature permission under review | New processor and retention review required | | Review burden | Baseline sample: 18 minutes median | No comparable observation | No comparable observation | | Pilot readiness | Available now | Hold until edition and permission confirmed | Hold until data and access review complete |
The brief recommends no vendor. It proposes two evidence-gathering actions: obtain written confirmation for the add-on and run the same synthetic cases through each approved route. The decision owner will then choose pilot, repair, defer, or retain the current process.
Acceptance tests
A reviewer should be able to reproduce every material entry, identify the exact alternatives and conditions, distinguish observation from inference, find the evidence cutoff, and see unresolved contradictions. Check that public comparisons are like-for-like and supported by current evidence. Confirm that legal, procurement, security, privacy, and trademark questions have the right specialist owner.
Two Prune can assess how a participant frames a comparison, navigates evidence, verifies claims, and communicates a decision in a realistic simulation. It does not certify competitor claims, choose a supplier, or make employment decisions.
Sources
eCFR, 17 CFR 229.101: Description of business
- UK Government, Marketing and advertising law
- UK legislation, Business Protection from Misleading Marketing Regulations 2008, regulation 4
Sources: [1] [2] [3]
Sources
- 1.eCFR, 17 CFR 229.101: Description of business · Electronic Code of Federal Regulations
- 2.UK Government, Marketing and advertising law · UK Government
- 3.UK legislation, Business Protection from Misleading Marketing Regulations 2008, regulation 4 · www.legislation.gov.uk