The platform

Everything you need to rehearse reality — in one synthetic society.

From census-grade persona generation to executive-ready reports, every capability below is shipping today — on a curated multi-model catalog with per-organisation governance.

Synthetic Audiences

Any audience on Earth, at population scale.

Describe the audience you need — in plain language, or grounded in your own structured data — and Lanice AI generates a statistically-stratified synthetic population to match. Any demographic depth, any size, any corner of the world, with the click of a button.

  • Describe it in a sentence — "Adults in Sydney aged 21–38 who shop online weekly and earn $50k–$100k" becomes a fully-stratified, test-ready audience
  • Or scale to a whole country — "Generate 500,000 personas for a general population of the United States, reflective of every state" is one instruction, not a research project
  • Ground it in your own data — upload census extracts, syndicated audience profiles or first-party panel summaries, and the population is stratified to match via a documented largest-remainder algorithm
  • 21 structured demographic columns + ~70 attitudinal dimensions per persona — race, ethnicity, geography, income, education, religion, politics, layered with price sensitivity, brand loyalty, media diet, risk tolerance and decision style
  • Big Five personality vectors, directly editable per persona, applied everywhere the persona acts
  • Reusable Demographic Pools — save "the Sydney online shoppers" or "the MI-08 electorate" once, use it in every test
Audience Builder — describe it, upload it, or both
01
Prompt"Adults in Sydney aged 21–38 who shop online weekly and earn $50k–$100k"
1,500 personas stratified · ready in minutes
02
Prompt"Generate 500k personas for a general population of the United States, reflecting every state"
500,000 personas · 50-state stratification plan 91 filterable dimensions
The Synthetic Society

A full social world, not a chat window.

Personas live in a Facebook-class social graph and act autonomously in configurable cycles — an X-style discovery algorithm decides what they see; their personalities decide what they do about it.

  • Posts, threaded comments, six reaction types, friendships, DMs — with deduplication so behaviour stays realistic
  • Groups, events, pages, stories, dating, and a marketplace with real buyer/seller decision tracking
  • Interaction Memory — every interaction stored with an AI-scored sentiment and emotional tone tag
  • Environment shocks — inject "hurricane warning", "rate cut" or "scandal breaks" at controlled magnitude and watch the network respond
  • Decision Traces — every action logged with the model's reasoning, for full auditability
Live feed — 5% of 10,000 personas per cycle
A
Amara O. · Portland, OR · sentiment +0.6"Okay the new campaign spot actually made me tear up a little?? The bit with the dad at the end 😭"
D
Dale K. · Lubbock, TX · sentiment −0.4"Nice ad, but I've been burned by 'limited time pricing' before. What's the catch?"
2,412 posts this runSentiment & tone scored on every artefact
Content Workspace

15 expert workflows on any asset.

Drop in text, an image, a PDF, a video, a YouTube link, a one-click screenshot of any live web page, or a whole reference corpus — and get scored, classified verdicts in minutes.

  • Each workflow scores 6–7 dimensions 0–5 and classifies readiness from not ready to high priority
  • Bring your own scorecard — type the questions your team actually asks and save them as reusable Assessment Templates
  • Pick the model per run from the catalog; the exact model used is stamped for audit
  • Promote to experiment — a polished asset is one click from a full campaign simulation
Campaign Message Resonance Creative Hook & Attention Narrative Frame Audience Reaction Forecast Objection Triage CTA Conversion A/B Test Design Backlash & Sensitivity Claim Substantiation Issue Salience Voice Consistency Competitive Contrast Fundraising Appeal Comment Insight Campaign Learning Extraction
Social Experiments

Real experiments on a synthetic population.

Point-in-time content drops or multi-phase campaign storylines — each phase with its own duration, behaviours, personality settings, environment events, surveys and content tests.

  • Simulate a launch week: Day 1 announcement → Day 3 pile-on → Day 5 opposition counter → Day 7 closing argument
  • Personality dials — emotional tone, controversiality, humour, verbosity and more, targeted at any granularity from one persona to the whole society
  • Clone lineage — fork an experiment, change one parameter, produce a clean comparison run
  • Durable orchestration — leased, heartbeated, cancellable runs that fail honestly rather than silently "completing"
Campaign — "Rebrand Launch Week" · phase 2 of 4
Phase 1 · Announcement✓
Phase 2 · Morning-after62%
Phase 3 · Counter-ad—
Phase 4 · Closing—
Environment event: "competitor counter-ad" · magnitude 0.6 Survey queued end of phase 3
Surveys

Ask the population directly — then interrogate the answers.

Standalone polls, surveys grounded in a finished experiment so respondents answer from lived in-world experience, or surveys primed with background content the persona has "seen".

  • Every response includes the reasoning and a confidence score — the why, not just the what
  • Cross-tabs with two-proportion z-tests flag statistically significant deviations automatically
  • Notable-segments finder scans every demographic × answer combination so the story finds you
  • Wave comparison — hold the questionnaire constant across runs, before and after a message change
  • AI theme extraction and executive summaries, plus full CSV export
Survey — "Purchase intent after spot v3" · 2,000 respondents
Very likely31%
Somewhat likely38%
Unlikely23%
Would never8%
Notable: Gen Z urban deviates −14pts (p<0.01) Reasoning captured per respondent
Debate Engine

War-game the argument itself.

Stage a structured debate over your topic, an uploaded video or article, a multi-document corpus — or today's breaking news, fetched live.

  • Up to five participants, each a demographically-grounded persona with a stance — human analysts can join too
  • Each persona optionally on a different frontier model, for genuine inter-model deliberation
  • Nine formats: formal debate, talk show, courtroom, Socratic dialogue, expert panel and more
  • An AI judge scores every participant 0–100, declares a winner, and extracts lessons for countering the winning narrative
  • Ask the transcript follow-up questions; export as PDF or report cards
Debate — Courtroom format · sourced from uploaded attack ad
S
Suburban independent · challenging"The ad claims premiums doubled. Exhibit A says otherwise — walk me through the arithmetic."
H
Hostile partisan · defensive"Averages hide the outliers. Families like mine saw exactly that increase."
Judge: independent wins 81/100Lesson: pre-empt anecdote-vs-average framing
UX Simulator

AI personas on your real website — with a mouse and keyboard.

Give a persona a URL and a task written in their terms. A real Chromium browser navigates, clicks, types and scrolls; every step is screenshotted and explained twice.

  • Dual-coded reasoning — the technical trigger and the psychological driver, tagged against a controlled vocabulary of ~20 drivers
  • Live view — streamed screenshots with click markers and a running narrative of each decision
  • Session analytics — outcome funnels, driver frequency, page flows, persona comparison, and an AI summary against your research objective
  • Research instrumentation, not automation abuse — step-capped, time-capped, cost-tracked, with an optional stop-before-purchase guard
  • The agentic dividend — the same session audits how an AI shopping agent parses your site, before that traffic arrives
Session — staging storefront · stop-before-purchase ON
Task completed4/6
Lost at shipping reveal2/6
Top driver: price_sensitivity58%
Top driver: social_proof41%
Friction: variant picker stalled 2 personas Every step screenshotted & explained
LLM Benchmarking

A standing observatory for how the models talk about you.

Ask the same questions to the same catalog models, on a daily schedule, and measure how the answers move. Not "which model is smarter" — what did every frontier model say about us this morning, and did that change overnight?

  • Up to 25 questions × 10 models, optionally framed as personas from your own audience
  • Two scoring signals — embedding drift for the charts and alerts; an LLM judge for what actually changed: sentiment, stance, new and vanished claims
  • Version correlation — every answer stores the concrete model version that served it, so a drift spike reads as a release, not a mystery
  • LLM poisoning detection — the flip side of version correlation: a spike in drift with no provider release behind it is exactly what poisoned training or retrieval data looks like. Daily version-stamped baselines make a manipulated answer about your brand, candidate or issue visible the morning it appears — with the changed sentences highlighted — instead of months later
  • Comparable by construction — temperature 0, fixed effort, stable persona sampling; failures leave visible holes, never silent gaps
  • Observational only — the platform records what models say; it never attempts to steer them
Answer Explorer — "What is [brand] known for?"
G
Gemini · Wednesday · drift 0.04"…known for reliable mid-range appliances and a strong warranty program."
G
Gemini · Thursday · drift 0.41 ⚠"…known for reliable mid-range appliances, though a 2026 recall of its kettle line drew criticism."
New claim detected · stance changeProvider version updated overnight
Planning Mode

Describe the goal; get an executable plan.

Type your research goal in plain language. The planner drafts a costed, step-by-step plan across the whole platform — and builds it with one click.

  • Up to 12 typed steps — persona generation, experiments, debates, workspace runs, surveys
  • Per-step price estimates against live model pricing and your organisation's discount, before anything runs
  • Open questions surfaced up front, never buried as silent assumptions
  • One-click build-out — everything a plan creates is a normal platform object afterwards: inspectable, editable, re-runnable
Plan — "Does the rebrand land with young suburban parents?"
1 · Generate 1,500-persona pool$—
2 · Workspace runs × 3 assets$—
3 · Point-in-time experiment$—
4 · Grounded survey + debate$—
Total fits monthly budget with headroom 2 open questions to confirm
The verdict layer

From fifty thousand reactions to one page worth reading.

Messaging Campaign Assessments

Map-reduce synthesis of up to ~50,000 posts into a scored executive readout — overall score 0–100, per-segment receptivity, prioritised recommendations. Queue batches by demographic slice and compare side-by-side, or on a red-to-green geographic score map.

Macro dynamics & network analytics

Network density, sentiment distribution and contagion, influence detection, cluster detection, interactive network graphs, Pearson correlation of personality dials against sentiment output, and adaptive time-series views.

Reports

20+ card types snapshot any view — assessments, sentiment timelines, debate transcripts, top influencers, cost summaries — into stable, point-in-time executive reports with an AI-written summary and clean PDF export.

Data Cuts & raw export

Tag any subset of the society and stream it out as RFC 4180 CSV — posts with sentiment, tones and engagement; personas with full demographics and Big Five columns — up to 100,000 rows per file. Straight into R, pandas or Stata.

The model catalog

Pick the mind behind the persona.

All AI traffic flows through a single gateway to a platform-curated catalog of frontier and open models — selectable per feature, rotatable for direct comparison, and restrictable per organisation by model or by region.

US, EU & CN providers

OpenAI, Google, xAI, Anthropic and Meta alongside Qwen, DeepSeek, Moonshot, Baidu, MiniMax and Mistral — every output stamped with the exact model that produced it.

Per-organisation governance

Restrict a tenant to "US models only", an explicit allowlist, or exclude individual models — enforced server-side everywhere a model is resolved. Built for sovereignty-conscious government tenants.

Rotate & compare

Run the same population, scenario and prompts across different models as the persona "voice" — and let LLM Benchmarking track the differences daily.

Spend governance

Cost transparency that scales down as well as up.

Every AI invocation is cost-tracked and attributed. Each organisation runs on a prepaid budget with a full ledger, self-serve Stripe billing with auto top-up, and a built-in Cost Calculator with presets from "Starter trial" to "Enterprise simulation" — so a two-person team can see exactly what a month of testing costs before committing, and finance gets clean spend reporting.

Prepaid budgets with spend caps Auto-pause on exhaustion Per-line-item ledger Self-serve Stripe billing Cost Calculator presets Per-category & per-experiment cost dashboards

See the whole platform on one call.

We'll run your actual message through a synthetic audience built to your spec — live.

Book a demo