Use Roles and Perspectives Without Mistaking Them for Expertise

A viewpoint can reveal a useful question. Its title does not prove the answer.

“Act as a world-class event consultant and review my plan.”

The answer sounds impressive. It mentions attendee experience, operational excellence, and strategic positioning. It recommends a bigger room and a more ambitious program.

But which facts support those recommendations? Did the assistant know the room's capacity? Did it see the budget? Did anyone ask the people who would attend?

A role request can focus an assistant's attention. It can encourage it to examine a proposal as a customer, teacher, organizer, or critic might. It does not supply professional credentials, firsthand knowledge, or independent evidence.

The useful part is the set of questions the perspective brings to the task.

Give the perspective a job

“Be an expert” is broad. “Review this plan for costs that are missing from the supplied budget” is a task.

You can combine the two:

Review the plan from a budget manager's perspective. Identify unsupported amounts, missing expense categories implied by the plan, and tradeoffs within the spending limit. Use the supplied figures and label assumptions.

Now you can evaluate the response. Did it find a relevant omission? Did it use the right numbers? Did it invent a cost?

Anthropic describes role instructions as a way to focus behavior and tone. That is a reason to make the role specific; it is not a basis for treating generated advice as a credentialed opinion. Anthropic: Give Claude a role.

Give every viewpoint the same proposal

Use this fictional event proposal:

Hold a 90-minute community workshop for 30 attendees in an upstairs room. The room quote is $60. Materials are estimated at $90, and printing at $20. The spending limit is $200. The plan calls for small printed worksheets and a standing activity halfway through. No elevator or alternative access route has been confirmed. No attendee feedback has been collected.

The listed expenses total $170, leaving $30 under the limit. That margin is real within the supplied arithmetic. Whether it is enough depends on what the plan still lacks.

Ask for three perspectives: participation and access, budget, and first-time attendee experience. Tell the assistant to use the same proposal for all three.

Without a shared input, one viewpoint may quietly imagine a different event, making the comparison meaningless.

Ask for criteria, concerns, and evidence

A useful response separates what the proposal says from what a reviewer infers:

These are reasoned concerns based on the supplied plan. They are not an accessibility certification, a vendor quote, or a survey result.

The assistant should not state, “The building has no elevator.” The actual fact is that no elevator or alternative route has been confirmed. An unknown is a reason to investigate, not permission to declare an answer.

Keep different kinds of statements separate

In a review, look for three categories.

Evidence: The proposal says the room is upstairs. The supplied quote is $60. The worksheet uses small print.

Inference: Some attendees may have trouble reaching the room or reading the material. This is a concern to investigate, not a finding about a particular person.

Judgment: The organizer may decide that uncertainty about entry is enough reason to delay room selection.

All three can be useful. Trouble starts when they are presented as the same thing.

Ask the assistant:

For each recommendation, identify the supporting fact, the assumption or inference, and what evidence could change the recommendation.

This makes it easier to turn a review into a practical next step.

Several simulated voices are not independent consensus

You may ask the same assistant to produce a panel of five advisers. They may all recommend the same choice.

That agreement is generated within the same exercise. It does not show that five independent professionals reviewed the evidence. Even separate AI systems can rely on overlapping information or repeat the same mistaken premise.

Use multiple perspectives to widen the questions you consider. For independent confirmation, obtain relevant evidence: a venue's documented answer, a real quote, feedback from attendees, or a review by a qualified person where the decision requires one.

If the assistant says “all experts agree,” ask which actual experts, which sources, and what they evaluated. If it cannot establish those details, remove the claim.

Make disagreement useful

Suppose the budget perspective favors the inexpensive upstairs room while the participation perspective favors investigating another venue.

You do not need the assistant to force a tidy compromise. You need the decision made clearer.

A useful response might say:

The upstairs room fits the listed budget, but the route needed by attendees is unresolved. Obtain venue access information before confirming it. Meanwhile, compare one alternative room so the organizer can evaluate actual cost and participation tradeoffs.

The proposal has advanced. It now has a question, an owner, and a next action.

Ask for the strongest reasonable objection to your preferred option too. If you already like the cheap room, write, “What fact could make this choice fail for our audience?” That is more useful than asking the assistant to praise your thrift.

Build a short decision note

For the workshop, the note could read:

Decision pending: Room selection.

Known: Proposed upstairs room costs $60. Listed event expenses total $170 against a $200 limit.

Unresolved: Usable access route for attendees; whether required expenses are missing; feedback on worksheet readability.

Next actions: Organizer obtains venue information and one alternative quote. A sample worksheet is reviewed by intended attendees. The activity is adapted to allow seated participation.

Decision criterion: Choose a feasible option that supports participation and fits the confirmed expense limit.

That note is less theatrical than an imaginary panel discussion. It is also easier to act on.

Students: use viewpoints to question an argument

A student can ask for the perspective of a skeptical reader, a person affected by a policy, or a reviewer checking evidence. Each can expose a different weakness in an argument.

Try:

Review my argument from these three perspectives. Identify one question each would ask. Distinguish objections about evidence from disagreements about values. Do not invent quotations or claim to represent what real people believe.

Then return to the assigned readings and real sources. If you simulate a historical figure's response, label it as an interpretation and check it against that person's documented views.

UNESCO's student framework emphasizes human judgment and responsible participation in the use of AI. Keeping generated perspectives separate from evidence supports that educational aim. UNESCO: AI competency framework for students.

Your exercise

Review the workshop proposal using the three perspectives. Select one recommendation from each and classify its evidence, inference, and judgment. Identify one real-world check that could confirm or change it.

Completion check: You can explain why a suggestion is useful without relying on the title of the simulated adviser who delivered it.