See the Outcome Before You Decide — What Is AI Opinion Simulation?

Whatever you put out into the world, you don''t know how it lands until after it lands.

You only learn the ad copy is hated after you ship the campaign. You only learn what consumers complain about after the product ships. You only learn which side will revolt after the policy is announced. You only learn whether the content will go viral after you publish it.

The same applies to individuals. What will my family say if I announce I''m not getting married? What will my friends think if I open a café? How will people react if I tell them I''m 33 and still living with my parents? In every decision, the thing we most want to know is "the outcome" — but the outcome only ever shows up after the fact.

What if, right before you decide, dozens of virtual personas could discuss and shift their opinions like real people, showing you the reaction in advance? This post introduces that system: AI opinion simulation.

1. How Public Opinion Actually Forms

Public opinion isn''t one person''s view. It forms through people watching, hearing, and influencing each other.

• Someone voices an opinion first
• People who agree pile on
• A counter-opinion appears
• An influential voice picking a side bends the trajectory
• Like-minded clusters form echo chambers
• Some opinions go viral; others get buried

Traditional research (surveys, focus groups) cannot measure this process. They ask "what do you think?" once and stop. What we actually want to know is "where will the opinion go?" — but our measurement tools only show "where it is right now."

2. AI Opinion Simulation: A New Path

With recent LLM advances, it is now possible to build AI agents that hold something like a persona.

Each agent has demographics (age, gender, occupation), personality (extraversion, conscientiousness, openness), and background (values, interests). They post, comment, follow, and like in a social-media-like environment, and across multiple rounds, they form and shift opinions.

The output:

• Direction of opinion — positive/negative ratio, distribution by camp
• The path opinion takes to form — who speaks first, who follows
• Influencers — which agents'' messages spread the most
• Echo chambers — which groups isolate, and how
• Key issues — which criticisms or interests recur
• Outcome predictions — purchase intent, support rate, intensity of backlash

Where surveys show "what people will say", simulation shows "how the opinion itself will flow."

3. Two Validation Cases

Does this actually work? The Starling team ran simulations using only pre-launch / pre-event information, then compared against actual outcomes.

Case A — Galaxy S25 (January 2025)

Right after the S25 Unpacked announcement. No pre-order, sales, or review data was given — only specs, price, and competitor context.

• Sentiment: 51% positive / 38% negative — within the industry benchmark range (50–60% positive / 35–45% negative)
• Top criticisms: battery, charging, S Pen, RAM, camera — same as the actual reviewer/community criticism
• Competitor mentions: iPhone, OnePlus, Pixel, Xiaomi — same as the actual comparison pattern

Case B — Thailand 2026 Election (February 2026)

The parliamentary dissolution and election happened after the model''s training data cutoff (November 2024), so the model did not know the answer. No poll numbers were provided — only the border-conflict timeline and party platforms.

• Starling''s predicted 1st: Bhumjaithai — same as actual result
• Starling''s predicted 2nd: People''s Party (Prachachon) — same as actual result
• Polls had the 1st and 2nd reversed; simulation got the order right

The point of these two cases is simple. The direction of opinion can be simulated in advance. And the result is not a coincidence — it is accurate enough to be used in real decisions.

4. Who Can Use This

AI opinion simulation isn''t a tool for one specific job function. It''s for anyone curious about the reaction before deciding.

• Marketers — how an ad, a product, a campaign will land
• PMs — what backlash a new feature might trigger
• Policy people — opinion distribution after a policy announcement
• PR — apology tone, announcement timing
• Content creators — whether the synopsis, thumbnail, or headline will catch on
• Individuals — how those around them will react to life decisions (changing jobs, family choices, relationships)

That last item matters. AI opinion simulation is no longer the exclusive domain of corporate marketing departments. Ordinary people can now preview how their decisions will be received.

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