Practical guide

How to create a positioning brief with AI

A positioning brief is a decision record about who a product is for, the problem it is chosen to solve, the alternatives it must beat, and the evidence that makes its promise credible. AI can accelerate synthesis and pressure-test language, but it should not invent audience insight or turn an internal opinion into market evidence. The FRAME method below produces a reviewable brief and an.

By Two Prune

A positioning brief is a decision record about who a product is for, the problem it is chosen to solve, the alternatives it must beat, and the evidence that makes its promise credible. AI can accelerate synthesis and pressure-test language, but it should not invent audience insight or turn an internal opinion into market evidence. The FRAME method below produces a reviewable brief and an evidence trail: Fix the decision boundary, Research the audience and alternatives, Articulate a position, Map messages to proof, and Evaluate with people and evidence.

F — Fix the decision boundary

Begin with the decision the brief must support. Name the product or offer version, market, audience, buying situation, geography, time horizon, and owner. State what will change if the brief is accepted: a launch message, product-page rewrite, sales narrative, campaign test, or a choice not to enter the segment.

Write exclusions as carefully as inclusions. A positioning brief for a controlled procurement-review assistant is not a position for every legal or purchasing workflow. If several products or audiences require materially different evidence, create separate decision records rather than hiding the differences in flexible prose.

The UK Government Communication Service's OASIS planning guide links objectives, audience insight, strategy, implementation, and scoring or evaluation. A positioning brief is narrower than a full campaign plan, but the sequence is useful: the proposition should follow the objective and audience evidence, not precede them. GCS OASIS guide

Sources: [1]

R — Research the audience, problem, and alternatives

Build an evidence table before asking AI for a positioning statement. Each row should contain an observation, source, date, audience or situation, confidence, contradiction, and permitted use. Useful inputs include recent interviews, observed workflow breakdowns, support themes, win-and-loss notes, product telemetry with clear definitions, competitor materials, and current policy or market constraints.

Separate four things that are often blended:

The GCS guidance on insight describes useful insight as more than raw data: it should explain attitudes or behaviours and help determine action. Apply that standard to every audience claim. A transcript quote can illustrate a concern, but one quote does not establish its prevalence. GCS insight guidance

Ask AI to cluster observations only after removing or protecting sensitive information and preserving links to the original records. Require the output to list counterexamples and unassigned evidence. A neat theme without traceable rows is a hypothesis, not a finding.

  • what people say they want;
  • what they currently do;
  • what creates a material cost or risk;
  • what makes them change or stay with an alternative.

Sources: [2]

A — Articulate the position and its contrast

Draft one internal position using this structure:

> For a defined audience in a defined situation, who need an evidenced problem resolved, the offer is a category or frame that provides a bounded outcome or capability. Unlike the real alternative, it does so through a defensible mechanism, supported by specific proof.

This is a reasoning scaffold, not mandatory public copy. Test every field. Is the audience observable? Is the problem consequential in the stated situation? Is the alternative what people actually use, including doing nothing? Does the category help the audience understand the offer? Is the claimed difference a real mechanism rather than an adjective? Does the proof support the exact wording?

Use AI as an adversarial reviewer. Give it the evidence table and instruct it to identify the strongest alternative interpretation, unsupported leap, missing comparator, and audience for whom the position would be wrong. Do not ask it merely to make the statement more persuasive.

M — Map messages to proof and disqualifiers

Turn the position into a message-proof map. For every message, record the intended reader interpretation, evidence source, owner, qualifier, expiry date, channel, and disqualifier. A disqualifier is a condition under which the message should not be used: an untested configuration, excluded workflow, unavailable integration, different jurisdiction, or audience without the stated problem.

GovS 011, the UK government functional standard for communication, calls for defined objectives, audience insight, propositions and messages, channel strategy, and measurable outcomes. It also separates communication planning from evidence that outcomes were achieved. That distinction prevents a compelling proposition from being reported as a demonstrated result. Government Functional Standard GovS 011

When AI proposes variants, require every variant to retain its evidence identifier and qualifiers. Reject a shorter line if it silently strengthens the promise. Store rejected variants so the same unsupported statement does not reappear in another channel.

Sources: [3]

E — Evaluate before release and after contact

Review the brief in three passes. First, an evidence owner checks that sources support the position. Second, a product or operational owner checks that the mechanism and boundary match the current offer. Third, representative readers explain in their own words what they believe the statement means. Record confusion, disbelief, unintended implications, and groups for whom the statement is irrelevant.

Choose an evidence-appropriate next step: accept for a bounded use, test a contested message, gather missing research, narrow the audience, revise the product boundary, or hold the brief. After release, monitor whether observed questions, objections, qualified demand, and outcomes fit the position. A click or generated lead does not by itself validate the market frame.

Worked artifact: procurement-review assistant

This synthetic example concerns a software assistant that helps mid-sized procurement teams assemble an initial review packet from approved supplier documents.

| Brief field | Recorded decision | Evidence or limit | | --- | --- | --- | | Audience and situation | Procurement operations leads preparing recurring supplier reviews with fragmented approved documents | Six recent workflow interviews and process maps; not evidence for legal teams or one-off strategic sourcing | | Problem | Reviewers lose time locating current files and reconstructing which checks were completed | Observed in four mapped workflows; frequency remains uncertain in the wider market | | Real alternatives | Shared folders plus spreadsheets, a configured procurement suite, or manual coordinator review | Alternatives verified in the interview set; no claim that one dominates the market | | Category frame | Evidence-linked review-packet assistant | Internal comprehension test needed before public use | | Bounded value | Organizes approved source material into a reviewable first-pass packet with source links | Describes the tested workflow; does not claim reduced risk or completed due diligence | | Mechanism | Version-aware retrieval, required source anchors, and human disposition of unresolved checks | Product owner confirms current configuration; excluded integrations are listed | | Disqualifiers | Unapproved repositories, autonomous supplier approval, legal conclusions, or missing source permissions | These conditions force a hold or a narrower workflow |

AI first proposes “the fastest, safest way to approve suppliers.” The team rejects it: neither comparative speed nor safety nor autonomous approval is supported. A second line, “Bring approved supplier evidence into one reviewable packet,” matches the observed mechanism but still requires a channel-specific claims review. The brief is accepted for a limited comprehension test, with a research gap logged for teams already using configured procurement suites.

The artifact retained for review contains the evidence table, two contradictory observations, the internal position, message-proof map, rejected variants, disqualifiers, owner approvals, and the next learning question. That packet is more useful than a polished sentence with no provenance.

Two Prune boundary

Two Prune can assess how a participant frames an ambiguous communication task, uses evidence, works with AI, verifies claims, exercises judgment, and delivers a business artifact in a realistic simulation. It does not conduct market research automatically, establish product-market fit, approve advertising claims, or make employment decisions. Customers remain responsible for the sources, audience research, product facts, release approvals, and real-world commercial decisions behind a positioning brief.

Sources

  1. 1.GCS OASIS guide · gcs.civilservice.gov.uk
  2. 2.GCS insight guidance · gcs.civilservice.gov.uk
  3. 3.Government Functional Standard GovS 011 · UK Government

Explore this topic