Simulations

A simulation is a complete run of your content through a crowd of AI agents. This guide covers everything from creating simulations to interpreting results.

Creating a Simulation

Every simulation starts with seed content, the material your crowd will react to. Go to /sim/new to begin.

Seed Types

CrowdProof supports three types of seed content:

Text

Raw text like a tweet, press release, product description, or announcement. This is the most flexible option. Agents react to exactly what you write.

We are excited to announce that starting next month, our premium plan will include unlimited API access at no additional cost.

URL

A link to a public page. CrowdProof reads up to the first three links before the run starts, so the crowd can react to what those pages say. A page it cannot fetch, and any extra link, stays as URL text. The report says which pages were read. Useful for testing landing pages, articles, or competitor announcements.

https://example.com/blog/product-launch

Scenario

A hypothetical situation described in third person. This is ideal for testing "what if" questions or competitor moves.

Apple announces that the next iPhone will not include a charging port, requiring all charging to be done wirelessly.

Audience Selection

Choose which crowd should react to your content. Each audience preset contains different archetypes with distinct personalities and perspectives.

See Audiences for detailed archetype lists.

Configuration Options

Agent Count

The number of AI agents participating in the simulation. More agents means more diverse reactions but longer processing time.

• 100 agents: Free first-run size. Camps still form; larger crowds reduce faction noise.
• 250 agents: Pro size. More voices than a Free first run; less faction noise.
• 500 agents: Pro cap. More voices than a Free first run.
• 2,000 agents: Team tier, broader faction dynamics
• 5,000 agents: Enterprise tier, maximum diversity

Simulation Rounds

Each round allows agents to read and react to each other. More rounds show how opinions evolve and factions consolidate.

• 5 rounds: Initial reactions only (minimum)
• 10 rounds: Early faction formation visible
• 20 rounds: Pro size. More rounds than a Free first run, which is capped at 10.
• 30+ rounds: Extended debate, useful for controversial topics (Team and Enterprise plans)

Running a Simulation

Click Launch Simulation to enter God Mode where agents begin reacting in real time, or Launch A/B Experiment to open the dashboard so you can watch both variants. A Free first run is 100 people and up to 10 rounds. Larger crowds take longer. Watch it live in God Mode. When the crowd finishes you get a report. A stopped or failed run has no report.

On a running crowd you can pause, inject events, or let it run to completion. A pending, finished, failed, or stopped run cannot pause or inject. See God Mode for interface details.

Interpreting Results

When the crowd finishes you get a synthesis report. A stopped or failed run has no report. Here is how to read each section:

Overall Sentiment

A summary of whether reactions landed positive, negative, mixed, or divided. A crowd that split into opposing camps reads Divided, matching the report. Mixed is a crowd that stayed near the middle. This is the quick answer: how did people react?

Faction Breakdown

A card per faction showing how agents clustered into opinion groups, each with its share of the crowd, average sentiment, and a representative quote. Look for:

• Dominant faction: What is the most common reaction?
• Opposition size: How large is the negative response?
• Curious/Pragmatic: Are people asking questions or waiting for more info?

Key Arguments

The most compelling points raised by agents on each side. These are summarized from actual agent posts. Use these to anticipate real objections and talking points.

Viral Moments

Content that spread rapidly through the crowd, receiving high engagement and reshares. This shows what messages resonate and which phrases stick.

Notable Quotes

Standout reactions from individual agents. These are selected for their representativeness, persuasiveness, or unexpected perspectives.

Sharing Reports

Share Report copies a public link. While the crowd is running, a colleague watches the live spectator feed. When the run finishes, the same link opens the report. A stopped or failed run has no report. You can also:

• When the crowd finishes, download the report as Markdown. A stopped or failed run has no report.
• Export the raw feed as CSV when the crowd has posted. A paused, stopped, or failed run with posts can too. Survey CSV is for collected responses.
• On Enterprise, scoped keys create, retrieve, and export. Live stream and inject stay in God Mode.

Running Follow-up Simulations

After reviewing results, you can run variations:

• A/B testing: On Team and Enterprise, turn on the A/B test toggle in the wizard and enter Variant B. Both versions then react to one shared crowd, follow network, and activity schedule. Two separate launches with the same audience each draw their own crowd, so part of every gap is sampling noise.
• Different audiences: See how Tech Twitter vs General Public reacts
• Refined messaging: Address objections raised in the first simulation

Next Steps

• Audiences - Deep dive into audience presets and OCEAN model
• API Reference - On Enterprise, scoped keys create, retrieve, and export