RosoAI SEO/GEO Squad V3: what changed on July 24, 2026

From Claude Projects to Claude Code or Codex

V2 loads into a Claude Project and follows a guided conversation. V3 works directly in a folder with Claude Code or Codex. This lets the engine create, check and resume project files without manually rebuilding the state.

The user opens the kit folder and pastes a starting sentence with the site URL. No API key needs configuring. The engine uses Python 3.10 or newer; if a dependency is missing, the agent should explain it and ask permission before installation.

21 agents, without launching all 21 just for volume

The first eleven agents cover collection, positioning, pages, demand, competitors, content, technical checks, authority, generative visibility, QA and reports.

Ten specialists add writing, express audits, single-page audits, Delta re-audits, international work, llms.txt, implementation, monthly monitoring, source acquisition and measurement imports.

The Master Orchestrator launches only the roles needed for the scope. A single-page audit therefore follows a different process from a full audit, monitoring or international task.

Dated evidence before every important finding

V3 keeps an explicit chain: evidence, fact, finding, action, result. Each important element should trace back to a source, date, method and scope.

Values use measurement statuses. Data may be observed, client-reported, inferred, unmeasured or unknown. Missing data is not turned into zero to make the report look cleaner.

An adversarial QA step then tries to challenge the audit. It can block delivery if evidence is missing, files contradict one another or a report no longer matches validated data.

Five dimensions instead of one opaque score

V3 separates Foundations, Observed Visibility, Opportunity, Execution and Measurement Quality. These dimensions have different meanings and are therefore not blended into one average.

Each result shows its coverage and confidence. Poor measurement quality limits the precision of conclusions, even when some figures are available.

AI engine visibility is also segmented. ChatGPT, Claude, Gemini, Perplexity and generative search features are not grouped into one universal rate.

A persistent project for each site or client

Each task can keep its Digital Twin, evidence, findings, actions and events in a separate folder. An agency or consultant can resume a client’s work without rereading the entire previous conversation.

The Delta Re-audit compares two states only when scope, markets, methods and contexts are compatible. If the comparison is invalid, the engine disables it instead of displaying artificial progress.

A local Control Center can then bring together several projects, their open actions and measurements. It is a locally generated static file, not a SaaS subscription or live dashboard.

PDF reports and tools for taking action

The full task produces a strategic audit, implementation plan and appendices. Reports can be generated in Markdown and professional PDF formats after data validation and output review.

The plan links actions to impact, effort, confidence, dependencies, owner and acceptance criteria. Sensitive changes include approval, testing and rollback.

V3 can also export actions as CSV for Jira or Notion and import GSC, Bing or GA4 files. These imports work from local exports and should not be confused with real-time API connectors.

White labeling, pricing and access

Reports can use the provider’s or client’s identity, logo and colors. White labeling does not remove methodological provenance, limitations or evidence needed to understand the deliverable.

The kit costs $89 as a one-time purchase. French-edition owners who bought before July 24, 2026 received V3 through their existing access. English files are being prepared.

The kit helps audit, prioritize, prepare changes and track measurements. It does not guarantee rankings, citations in an answer engine or revenue.

Frequently asked questions