State-aware GEO, end to end.

One project = one brand × one market × (optionally) one product line. In the US, a market can be the whole country or a single state — “California, US”, “Texas, US”, “Florida, US”, “New York, US”. State projects measure the state’s own answer surface and its own local competitors.

Multi-line brands (e.g. a flagship SMB product and an emerging enterprise line) should run one project per line: each line gets its own prompt library, competitor set and score. Composite scores mask exactly the differences you care about — in our Payoneer test the two lines scored 58 vs 57 while sharing only 2 of 12 AI-named competitors.

Run Profile before content agents: it fetches the official site into one canonical fact base. Every asset is grounded on it — agents never invent facts.

Discovery → Monitor → Report in one click (~4–6 min). The first scan becomes the Day-0 baseline.

Work the gaps with Optimize / Site / Distribute / Encyclopedia, then re-scan to prove the lift. The trend panel tracks per-scan deltas and month-over-month.

The “Sets” button in the workspace header (project admins). Edits live in project config — the Discovery asset and scan history are never touched.

Tag each entity competitor / partner / directory / self. Only “competitor” enters Share of Voice, the benchmark and gaps; tags survive the monthly refresh, so a marked partner is never re-discovered as a competitor. The scorecard states who is excluded and why.

Click any prompt to exclude/restore; add custom prompts one per line. Applied from the next run.

The machine discovers competitors from real answers; a human qualifies them. Real case: AI listed Payoneer’s partners Upwork and Fiverr as competitors until they were tagged partner.

Agents generate verifiable facts, never experiences. First-person user-experience voice in any asset is treated as a fabricated testimonial and dropped.

Reddit, Quora, Facebook Groups and forums never get ghostwritten posts. They get an engagement brief: where to engage, what people ask, which verified facts to contribute — always from a disclosed official account.

“Trusted by millions” style claims with no source are deterministically flagged per item for operator review before anything is sent.

Wiki content ships only with the compliant path: paid-relationship disclosure ({{paid}}) → AfC draft review or Talk-page edit request → independent editors merge. Never direct posting.

Review content is never ghostwritten. The Report agent may recommend a genuine review-solicitation program (invitation links to real customers) — the customers write the reviews.

For regulated verticals the state frame adds jurisdiction-specific advertising constraints (e.g. bar rules on outcome guarantees, license-number display) directly into generation.