sales-skills/sales-societies

Artificial Societies Platform Help

Artificial Societies (societies.io, YC W25) is a synthetic-audience simulator. Where the other synthetic-research tools run interviews against independent AI participants ([[sales-syntheticusers]], [[sales-imario]]) or let you chat with one data-grounded persona ([[sales-delve]], Marketing Mary), Societies models a whole audience as a network — 300 to 5,000+ AI personas placed on a social graph who react to your content and influence each other, so the output is how a message spreads and how opinion propagates, not just isolated reactions. Each run takes ~30s–2 min.

Its self-serve product (the LinkedIn-audience simulator, "Reach") is built for pre-publishing a post or message: pick or build a target audience, paste your content, and get an engagement score plus alternate variations to compare — its headline claim is an R²=0.78 fit to real LinkedIn engagement (personas are built from LinkedIn data). The enterprise Radiant tier builds purpose-built societies from first-party CRM/qual/quant data with follow-up qualitative interviews, SSO/SCIM, SOC2/GDPR.

It is not demand. The founders say it plainly — "synthetic audiences should never replace listening to real people." A predicted engagement score is a directional pre-test that can produce false positives (a bad message can still score high). Use it to shortlist and sharpen; earn the go/no-go from real behavior.

Step 1 — Gather context

If references/learnings.md exists, read it first for accumulated platform knowledge.

Ask only what you can't infer:

1. What do you want from Artificial Societies?
   • A) Run a simulation and interpret the result — the engagement score, the network reactions, the variations
   • B) Pre-test a specific LinkedIn/marketing post or message before publishing
   • C) Understand the accuracy claims (R²=0.78, 95% opinion accuracy, 90% coherence) and how much to trust them
   • D) Pick a plan — Free (3 credits) vs Pro ($40/mo) vs Team vs enterprise Radiant
   • E) Automate/export programmatically (API reality)
2. What's the real question — "how do I use this tool?" or "should I build/ship this?" If it's the go/no-go decision or a validator comparison, that's /sales-idea-validation (a synthetic score is not demand) — route in Step 2.
3. Self-serve or enterprise? The $40/mo self-serve product tests content/messaging; Radiant builds data-grounded societies for larger studies. Answer for the tier the user is actually on.

Skip-ahead: if the user wants to compare synthetic-research tools or the validate-before-building method, route to /sales-idea-validation immediately.

Step 2 — Route or answer directly

| If the user's question is about… | Route to | |---|---| | Comparing Societies vs other synthetic-research/persona/validator tools, or the go/no-go decision | /sales-idea-validation {question} | | The pure-play synthetic-interview study tool (multi-participant studies, with a real public API) | /sales-syntheticusers {question} | | A reusable Synthetic Individual across many jobs (memory, Pro-gated API) | /sales-imario {question} | | Building the real smoke-test landing page to measure demand | /sales-funnel {question} | | Actually posting/scheduling the content once it's tested (LinkedIn/social) | /sales-social-media-management {question} | | Turning tested messaging into a real content program | /sales-content {question} |

When routing, give the exact command: "This is a {domain} question — run: /sales-idea-validation {original question}"

Otherwise, answer Societies-specific questions using Step 3.

Step 3 — Artificial Societies platform reference

Read references/platform-guide.md for the full reference — the two products (self-serve LinkedIn/audience simulator vs enterprise Radiant), how a simulation works (persona network on a social graph, opinion propagation, ~30s–2 min runs, content variations), how to define an audience (demographics / professional targeting / first-party data), the accuracy claims and their caveats (R²=0.78 LinkedIn, 95% opinion-distribution, 90% coherence — all vendor self-reported), the pricing tiers (Free 3 credits → Pro $40/mo → Team → enterprise Radiant), and the no-public-API automation reality plus how to read a synthetic result honestly.

Answer using only the relevant section — don't dump the full reference.

Step 4 — Actionable guidance

• Treat the engagement score as a directional pre-test, not demand. The score is a synthetic prediction — personas modeled from public/social data reacting to your content. It's genuinely useful for A/B pre-testing messaging (which of two hooks lands, where an argument loses people), and the R²=0.78 LinkedIn fit is a real signal — but it is not a stranger taking an action. Have the user keep the comparative read (variation A beats B, this claim triggers pushback) and take the go/no-go from real behavior — a live post's actual engagement, a smoke test, a pre-sale. Route the real test to /sales-idea-validation and /sales-funnel.
• Use it for relative comparison, not absolute truth. Its strength is ranking variations against each other; its weakness is false positives on absolute scores (the founders' own example: a non-working service scored 81, a working one 88 — both "high"). Tell the user to trust "B > A" more than "88/100 = good," pick the winning variation, then validate it for real.
• Match the product to the job. The self-serve tool (a tiny Free tier — 3 credits + a 2-week trial — then Pro ~$40/mo unlimited; flag prices as best-effort to confirm on societies.io) tests content and messaging against a LinkedIn-style audience — perfect for a solo founder pre-testing a post. Radiant (enterprise, contact-sales) is for data-grounded studies on first-party CRM/research data with follow-up interviews — overkill for a solo maker with no data to ground it in. Don't send a solopreneur to the enterprise tier.
• Define the audience as tightly as you can. Accuracy depends on the society matching your real audience — use demographic (age/gender/location/income/education) and professional (company/industry/job title) targeting, or first-party data on enterprise. A generic society gives a generic prediction; a precisely-specified one is the whole point. Flag when the user's audience is too vague to simulate well.
• There's no public API — don't plan an integration around it. If asked to automate, batch, or pipe results elsewhere, say plainly there's no documented public API, no webhooks, no Zapier/Make, no MCP (the docs.societies.io subdomain doesn't resolve). The self-serve product is browser-only; enterprise Radiant ingests first-party data but exposes no public developer API. The thing worth automating is the real signal (published-post engagement, landing-page conversions), not the simulated score.

If you discover a gotcha or tip not in references/learnings.md, append it there with today's date.

Gotchas

Best-effort from research (2026-07) — Societies' products, pricing, and accuracy claims move; verify at societies.io.

• A synthetic engagement score is not demand. The founders say synthetic audiences "should never replace listening to real people." Use it to pre-test messaging; earn the go/no-go from real behavior.
• False positives are the failure mode. A bad message can still score high (81 vs 88 in the founders' own example) — trust relative comparisons between variations far more than any absolute number.
• Accuracy claims are vendor self-reported. R²=0.78 (LinkedIn engagement), 95% opinion-distribution accuracy, 90% persona coherence — none independently audited. Treat as marketing until verified.
• Personas skew to public/LinkedIn data. The R²=0.78 fit is for LinkedIn engagement specifically; generalizing to other channels or niche/offline audiences is weaker (the founders flag this).
• Two products, very different access. Self-serve (societies.io) = $40/mo, content/message testing. Radiant = enterprise, contact-sales, first-party-data societies. Don't quote enterprise capabilities to a self-serve user or vice versa.
• No public API, no webhooks, no Zapier/Make/MCP. It's UI-only; there's no supported way to script, batch, or export programmatically. docs.societies.io does not resolve.
• Free tier is tiny. 3 credits + a 2-week trial, then Pro at $40/mo for unlimited simulations — confirm current terms on societies.io before relying on them.

Related skills

• /sales-idea-validation — The tool-agnostic validate-before-building method + the full synthetic-research and validator landscape (use this to decide build-or-not and to compare Societies against iMario/Synthetic Users/Delve; a synthetic score is not demand)
• /sales-syntheticusers — Synthetic Users platform help (the pure-play synthetic interview-study tool — multi-participant AI interviews with a real public REST API; contrast its independent-participant studies against Societies' influence-network simulation)
• /sales-imario — iMario platform help (reusable Synthetic Individuals with persistent memory across six jobs, Pro-gated API; the reusable-persona cousin to Societies' one-shot network runs)
• /sales-delve — Delve AI platform help (analytics-connected data-grounded personas + Digital Twins you chat with; contrast a single chattable persona against Societies' whole-audience network)
• /sales-funnel — Build the smoke-test / fake-door landing page that measures the real demand a synthetic score only predicts
• /sales-social-media-management — Actually publish and schedule the content once Societies has helped you pick the winning variation
• /sales-do — Not sure which skill to use? The router matches any sales objective to the right skill. Install: npx skills add sales-skills/sales --skill sales-do -a claude-code

Examples

Example 1: "Societies scored my LinkedIn post 88 — should I post it?"

User says: "I ran my launch announcement through Artificial Societies and it scored 88. Is that good enough to publish?" Skill does: Explains the score is a synthetic prediction, useful for comparison but prone to false positives on absolute numbers (the founders' own example: a non-working service scored 81, a working one 88 — both "high"). Advises running two or three variations and trusting "B beats A" over the raw 88, then picking the winner. Notes the R²=0.78 LinkedIn fit means the relative read is trustworthy, but real engagement on the live post is the actual signal — and if this is really a build-or-not question, routes to /sales-idea-validation. Result: The user pre-tests variations, ships the winner, and reads the real post's engagement as the truth.

Example 2: Free vs Pro vs enterprise Radiant

User says: "I'm a solo founder. Do I need the enterprise version of Societies, and what does it cost?" Skill does: Maps the tiers — Free (3 credits + 2-week trial) → Pro ($40/mo, unlimited simulations) for self-serve content/message testing → Team → enterprise Radiant (contact-sales, first-party-data societies with follow-up interviews). Explains Radiant is for data-grounded studies a solo maker has no data to ground yet, so Pro is the right tier for pre-testing posts. Flags prices as best-effort to confirm on societies.io. Result: The user picks Pro and skips an enterprise sales call they don't need.

Example 3: Can I automate Societies via API? (developer/automation)

User says: "I want to run posts through Societies from a script and pull the scores into my own dashboard." Skill does: States plainly there's no documented public API, no webhooks, no Zapier/Make, no MCP — the docs.societies.io subdomain doesn't resolve and the self-serve product is browser-only; enterprise Radiant ingests first-party data but exposes no public developer API. Suggests that if a pipeline is required, the thing worth automating is the real signal — published-post engagement (a social API) or landing-page conversions — not the simulated score, and points to /sales-social-media-management and /sales-funnel. Result: The user avoids building on a non-existent API and automates the real signal instead.

Troubleshooting

The score feels high even for a message I know is weak

Symptom: Societies returns an encouraging engagement score for content that flopped or feels off. Cause: Synthetic personas modeled from public data over-predict plausible-sounding content; absolute scores carry false positives (the founders' own 81-vs-88 example). Solution: Stop reading the absolute number as a verdict. Run multiple variations and use Societies for the relative comparison (which variation wins), then validate the winner with real behavior — a live post's engagement or a smoke test via /sales-idea-validation and /sales-funnel.

The prediction doesn't match my real audience

Symptom: Reactions don't look like how your actual audience responds. Cause: The society is too generic or the R²=0.78 fit is LinkedIn-specific and you're testing a different channel/niche. Solution: Tighten the audience definition (demographics + professional targeting, or first-party data on enterprise) so the society matches your real audience; treat non-LinkedIn or niche/offline predictions as weaker, and always confirm against a small real test.

I want an API or a way to export/automate

Symptom: Want to script Societies or pull scores into another system. Cause: Societies has no documented public API, webhooks, or iPaaS connectors; docs.societies.io does not resolve. Solution: There's no supported programmatic path — the self-serve product is used in the browser. Automate the real signal instead (published-post engagement, landing-page analytics), and see /sales-idea-validation and /sales-funnel.