Position an AI-enabled product around the job it helps a defined customer complete, the outcome that matters, the role AI plays and the boundary within which that outcome is credible. Treat AI as the category only when buyers genuinely organise their search and budget around that category. In many markets, AI is better presented as a delivery mechanism or supporting capability, with clear evidence, human controls and limits.
"AI-powered" may attract attention, but it does not tell a buyer what the product is for, which alternative it replaces, whether it is suitable for the task or what happens when the system is wrong. A durable position gives those answers before asking the buyer to admire the technology.
Decide which role AI plays in the market story
AI can sit in three different places:
- Category: buyers actively seek a recognisable class of AI product, allocate budget to it and compare vendors within it.
- Mechanism: AI is how the product performs part of a familiar job, such as classifying requests or drafting a first response.
- Supporting capability: AI improves a feature inside a product primarily understood through another category.
Do not decide this from the model architecture. Ask how buyers describe the triggering problem, the budget, the comparison set and the intended outcome. If prospects say, "We need to reduce the time spent triaging cases," then "AI case triage" may be a helpful descriptor, while "enterprise AI platform" may hide the job. If a chief data or AI officer is buying shared model governance, the AI category may be central.
The existing To-Do Growth guide to choosing a market category provides the broader category method. The special challenge here is that AI terminology evolves quickly and can cause buyers to infer autonomy or intelligence beyond the actual function.
Lead with a use case that can be bounded
A useful use case identifies:
- the user and setting;
- the input or trigger;
- the task the system performs;
- the output and who uses it;
- the outcome to be improved;
- the cases that remain outside scope.
For example, "draft a first summary of an English-language support conversation for an agent to review" is more positionable than "transform customer experience with generative AI". It establishes a user, material, output and review step. The business can then test accuracy, time and exception handling within that boundary.
The US National Institute of Standards and Technology's AI Risk Management Framework organises AI risk work around govern, map, measure and manage. Its Generative AI Profile applies the framework to generative AI risks. These are risk-management resources, not positioning templates, but they support a vital discipline: describe the context and impact of the system before making broad claims about it.
Build a use-case, risk and proof map
Complete one row for each use case you intend to place in the market story. If one message requires several materially different rows, it may be too broad.
| Field | Question | Evidence or decision required |
|---|---|---|
| Customer job | What is the person trying to complete? | Observed workflow, recent examples and current alternative |
| AI role | What does the system generate, predict, rank or detect? | Product documentation and a plain-language explanation |
| Intended outcome | What should improve for whom? | Defined measure, baseline source and relevant time period |
| Material failure | What could go wrong or be misunderstood? | Risk assessment using the actual context and affected people |
| Human control | Who reviews, overrides, appeals or stops the output? | Tested workflow, authority and escalation path |
| Data boundary | What data enters, persists or trains the system? | Current data-flow and contractual/privacy documentation |
| Performance boundary | For which languages, populations and conditions is evidence available? | Evaluation design, dataset scope and known exclusions |
| Market proof | What can be stated publicly now? | Claim-to-evidence record with date, version and qualification |
The map prevents a familiar failure: using one successful demonstration to support a product-wide promise. It also helps marketing, product, legal, security and customer-success teams work from the same defined use case.
Convert model capability into customer-relevant evidence
A benchmark can be useful without proving the marketed outcome. A model's performance on a test set may not establish how the complete product performs with live inputs, human review, latency, integration failures or changed data.
Create a claim ladder:
- 1 Function claim: what the system does, such as producing a draft or ranking items.
- 2 Performance claim: how well it does within a defined evaluation.
- 3 Workflow claim: how the complete human-and-system process changes.
- 4 Business claim: what happens to cost, revenue, risk or service outcomes.
Evidence usually weakens as a team climbs the ladder without new research. Do not turn a measured function claim into an unmeasured business outcome. The UK Advertising Standards Authority's current discussion of AI claims and the impression of intelligence reinforces that marketers must consider the meaning people take from
the claim, not only the technical words chosen. Objective and comparative statements need suitable substantiation for the relevant task and context.
Keep the conditions close to the claim. State the product version, task, relevant population or language, evaluation method and important exclusions. Avoid a footnote that quietly removes the meaning created by a large headline.
Make explanation proportional to impact
Not every AI-assisted feature requires a technical lesson. Buyers do need enough information to understand the product's role, make an informed choice and use it appropriately.
The UK Information Commissioner's Office sets out principles for explaining AI-assisted decisions, including being transparent, appropriately accountable and sensitive to context and impact. Its practical summary of explanation tasks recommends preparing explanations through the design and deployment process rather than adding them at the end.
For positioning, that means the level of explanation should rise with the consequence. A suggestion that a user can easily ignore needs a different market story from an output affecting credit, employment, health, education or access to an essential service. Name the human role accurately. "Human in the loop" is not enough if the reviewer lacks time, information or authority to challenge the result.
Avoid anthropomorphic and autonomy inflation
Words such as "thinks", "understands", "knows" or "decides" can create an inaccurate mental model. Sometimes they are obvious shorthand; sometimes they imply capabilities the product does not have. Use the most concrete verb available: generates, classifies, predicts, retrieves, recommends, flags or summarises.
The UK Advertising Standards Authority has examined how AI terms can create misleading impressions, including when advertising encourages people to treat systems as if they possess human qualities. Its general guidance on misleading advertising distinguishes obvious subjective exaggeration from objective claims that require evidence.
Audit the whole impression, not just technically defensible sentences. Product names, images, demonstrations, interface copy and claims about autonomy combine to shape what a reasonable buyer expects. If a user must check every output, do not imply a fully delegated outcome.
Illustrative example: AI as mechanism, not category
Illustrative example - this product and evidence are hypothetical.
Imagine a procurement tool that extracts renewal dates and notice periods from uploaded contracts, then sends each item to a contract manager for confirmation. Prospects are not searching for an "AI procurement platform". They are trying to prevent missed renewal windows and reduce manual review.
The useful position leads with renewal control and describes AI as the extraction mechanism. The proof plan evaluates defined document types, English-language contracts, extraction fields and the reviewer workflow. The message states that the output is a draft for confirmation and excludes legal interpretation. A later expansion into negotiation or risk analysis would require separate use cases, evidence and positioning.
The example is intentionally modest. Credibility often increases when the team says exactly what the system does and where professional judgement remains.
Test understanding, trust and action separately
A favourable reaction to the phrase "AI-powered" is not enough. Run three distinct tests:
- Understanding: Can the participant explain the job, AI role, input, output and limit accurately?
- Trust calibration: Does the participant rely on the product to an appropriate degree, neither dismissing it automatically nor assuming infallibility?
- Action: Does the intended buyer take a relevant next step, such as reviewing evidence, starting a scoped trial or involving the correct stakeholder?
Test with people affected by the output as well as the budget holder where the roles differ. Record misunderstandings. If users repeatedly assume the product trains on their data, works without review or covers a language you have not evaluated, the position needs a more visible boundary.
Account for changing rules and jurisdictions
AI rules are developing and differ by market, role and use. The European Commission's AI regulatory framework overview explains the risk-based structure and implementation timetable of the EU AI Act. A company may have different obligations depending on whether it provides, deploys, imports or distributes a system and on the use case.
Do not convert a generic compliance statement into a market advantage without confirming its meaning. "EU AI Act compliant" can be too broad if obligations are phased, the role is unclear or the specific system has not been assessed. Obtain current, jurisdiction-specific legal advice for consequential claims. Positioning should use verified facts such as documented controls, transparency and evaluation scope, not an unsupported badge.
Limitations and trade-offs
AI capability, terminology, product versions and regulation can change quickly. A positioning test captures interpretation at a particular time and does not validate technical performance. Evaluation results may not transfer across languages, populations, environments or changed models. Strong boundaries can reduce short-term drama while increasing credibility; a broad AI category may attract attention while also attracting the wrong comparison set and greater proof burden. Product, data-protection, safety and legal specialists should review claims proportionate to the use case's impact.
Further reading
- Decide whether AI is truly the frame with How to Choose a Market Category for Your Startup.
- Build a defensible evidence set with How to Choose Proof Points for a Value Proposition.
- Test buyer interpretation using How to Validate Your Positioning Strategy.
Sources
- NIST: Artificial Intelligence Risk Management Framework
- NIST: Generative Artificial Intelligence Profile
- UK ICO: Principles to follow when explaining AI
- UK ICO: Summary of the tasks to undertake
- ASA: Regulating the illusion of intelligence in ads
- ASA: Misleading advertising
- European Commission: Regulatory framework for AI
Your useful next step
Choose the one use case most likely to begin a real purchase and complete every row of the use-case, risk and proof map. Draft a position that names the customer job first and the AI role second. Ask five representative people to explain it back to you, log every inaccurate inference, and revise the boundary before adding stronger outcome language.
Publication note
Publication note: This article is a positioning framework, not technical, legal, data-protection, safety or advertising advice. AI obligations and evidence requirements depend on the system, role, impact and jurisdiction. Have qualified reviewers approve consequential claims and recheck cited guidance, product versions and evaluation scope before publication.
