How BrandGrowth OS uses AI

BrandGrowth OS uses AI in a few deliberate places — and keeps it away from the parts of the product where measured results matter more than fluent prose. This page explains where AI appears, the rules it operates under, and what happens when it is unavailable. It is the public overview of the platform's AI policy; the product documentation inside the app covers day-to-day use.

Where AI appears

• Assistant — a small chat widget on every signed-in page. Ask about your workspace, your reports, or how a feature works. It knows your workspace name and plan, answers from the product's own context, and is rate-limited per user to keep the service fair.
• "This week in numbers" — an optional paragraph in the Monday digest email, written from the measured rows in that email and nothing else. It is always labeled as AI-assisted, and an empty week simply omits it.
• Plain-English explanations — any visibility finding or action-plan item can be explained in plain language: what it means, why it matters, and what to do first.

Where AI is deliberately absent

Scores, rankings, opportunity matching, and every number in your reports come from deterministic rules — the same inputs always produce the same result. AI may phrase results, never produce them. If something was not measured, the product says "not measured" instead of guessing. That honesty rule applies to the AI features too: a model is never allowed to invent a metric.

Free tiers only

The AI layer runs exclusively on free model tiers across several providers. No payment card is attached anywhere in this pipeline, so there is no billing risk to the service or to you.

Self-updating model selection

Model catalogs change — providers close or open free access over time. So the system never trusts a hardcoded model list: on each provider it reads the provider's own live catalog and price metadata, keeps only genuinely free models, and orders them by preference. Catalogs are cached for a few hours. A model that stops being free drops out of the rotation within that window; a newly-free model joins without any redeploy. If every provider is unavailable, the AI features step aside gracefully — the digest omits its AI paragraph and the assistant shows a friendly retry message. No user ever sees an AI error wall.

Privacy and security

• API keys live only in server-side environment files — never in the browser and never in the repository.
• The model sees only what it needs: your workspace name and plan for the assistant, and a finding's own measured fields for an explanation. It does not receive your other data, and it cannot see other workspaces.
• Per-user rate limits protect capacity and keep the free tiers sustainable.
• AI output is sanitized before it is stored or shown: no invented numbers, no links, no formatting surprises.

Keeping it healthy

The platform runs an automated weekly health check across every AI provider and emails the operator a per-provider status table, along with usage accounting that shows how the provider fallback chain is distributing load. A provider that stops offering free models is caught by that check — and by the self-updating catalog — long before it can affect the product.