What Can You Simulate — Starling's Five Categories at a Glance

The previous two posts covered what AI opinion simulation is (part 1) and how it works (part 2). Now for where you actually use it.

Starling is organized into five categories. Each category has different simulation logic, asks different questions, and produces different outputs. Picking the right category for your situation is the first step to getting accurate results.

This post explains each category through two lenses: "when do you use it" and "what outputs do you get." The following posts cover how it differs from traditional research (part 4), how to design good inputs (part 5), and how to read the result report (part 6).

1. Marketing Reaction

When to use it

Right before a decision aimed at consumers — product launches, ad campaigns, pricing, rebranding. Whenever you want to see the post-launch reaction in advance.

What you get

• Sentiment distribution — positive / negative / neutral ratios
• Purchase intent — T2B (definitely/probably will) vs B2B (definitely/probably won''t)
• Top criticism topics — the 3–5 recurring complaints (auto-extracted)
• Competitor mentions — which alternatives consumers brought up

Primarily used for ad copy A/B/C, product concept screening, and pre-launch price sensitivity.

2. Public Opinion

When to use it

Social issues, election campaigns, public announcements — anywhere camps may split. When you want to see how opinion distributes and forms.

What you get

• Distribution by camp — support / oppose / neutral, with intensity
• Generational and regional breakdown — who leans where
• Key issues — what the camps split over
• Opinion flow — how views shift over time

Used for election simulation, social-issue opinion prediction. Where polls show "where opinion is now," this category shows "where it will flow."

3. Policy Debate

When to use it

Before a government, institution, or company announces a new policy or system. When you want to check the for/against structure and expected pushback.

What you get

• Pro/con distribution — ratio and intensity of each position
• Core opposition arguments — main points and reasoning on the other side
• Persuadable middle — what message lands with which group
• Expected backlash scenarios — how the opinion flow unfolds after announcement

Used for corporate policy announcements (salary bands, return-to-office), public policy simulation, and labor-management negotiation strategy.

4. Crisis Response

When to use it

Crisis announcements, apologies, sensitive disclosures — any situation where one message either calms the storm or amplifies it.

What you get

• Reaction differences across message tones
• Trigger keywords to remove
• Branch points between "calm down" and "escalate" scenarios
• Timing effects — announce now vs. later, simulated

Used for pre-validating apology statements, recall announcements, executive replacement disclosures. Crisis response is one of the highest-ROI categories — a single misstep can cost millions in recovery.

5. General

When to use it

Everything that doesn''t fit the four above. Ranges from content pre-validation to personal everyday decisions.

What you get (varies by situation)

• Interest / engagement / share intent (content-style questions)
• Reaction differences by group (relationship questions — family, friends, coworkers)
• Predicted conflict points and intensity
• Messaging guide — how to phrase it to land softly

This is the broadest category. Examples:

• Pre-validating movie synopses, YouTube thumbnails/titles, article headlines
• "How will my family react if I announce I''m not getting married?"
• "How will my friends see me if I tell them I''m 33 and still living with my parents?"
• "What will people around me say if I leave my big-company job for a startup?"
• "What happens to the relationship if I skip a friend''s wedding?"

Thanks to this category, AI opinion simulation is no longer just a corporate tool. Individuals can now preview how a major life decision will be received.

Getting Started — Your First Simulation in 30 Minutes

It looks complicated, but starting is simple.

1. Pick a category — one of the five above
2. Enter the topic — one paragraph on what you want to see reactions to
3. Set the target — whose reactions you want to see (population composition)
4. Run the simulation — 25 agents / 15 rounds as the default
5. Read the report — arrives in 30 minutes to an hour

For a first run, start with a small decision.

If you want to try it yourself, sign up for the free tier — credits are granted immediately on signup.