What should we do next?

Evidence backed opportunities and strategies now, governed execution and verification later.

Agent is the action layer. At launch it is recommendation only: it detects opportunities, investigates the evidence and recommends a strategy, each with a defined measurement. Approval, execution, verification and experiments arrive in V3, with policies that control what can happen.

Get an AI Audit

All product areas

Agent

What should we do next?

Opportunities
Strategies
Actions and Approvals
Verification
Experiments and Impact
Activity

Quick answer: Agent

Agent is the action layer of BrandWater AI. At launch it is recommendation only: it detects opportunities, investigates the evidence and recommends a strategy, each with a defined measurement. Approvals, execution, verification, experiments and an activity timeline are coming in V3, governed by policies. Nothing is published to your site without approval.

The agent lifecycle

Recommend first. Execute only with approval, later.

State 1At launch
Detected
The opportunity engine finds a gap or anomaly.
State 2At launch
Investigating
Prompts, answers, sources and competitors are traced.
State 3At launch
Recommended
A structured strategy is produced, with evidence.
State 4V3
Approval
A permitted person approves, edits or rejects.
State 5V3
Executing
The approved change is applied through a connected system.
State 6V3
Verifying
The target state is checked and affected prompts re-run.
State 7V3
Completed
The outcome is stored with its evidence.
State 8V3
Monitoring
Recurring checks watch for change.

At launch, you act on recommendations yourself. From V3, approved actions are executed, verified and measured.

What Agent gives you

6 capabilities, each answering a specific question and labelled with its release stage.

Available at launch

Opportunities

Where are the gaps worth acting on?

Evidence backed opportunities, each with the problem, the evidence, the scope, the impact signal, a recommendation and how success will be measured.

Available at launch

Strategies

What is the plan to close the gap?

A recommended plan for each opportunity, expressed as concrete actions across content, technical, structured data, authority, competitive and localisation classes.

Coming in V3

Actions and Approvals

How do approved changes get made safely?

An approval center, an action library and agent policies. Every external change needs explicit approval unless a policy you set allows auto execution for a low risk class.

Coming in V3

Verification

Did the change really happen, and did it help?

Automatic checks that the target changed as expected, followed by a targeted re-run of only the affected prompts to produce a before and after result.

Coming in V3

Experiments and Impact

What was the measured effect of our work?

Before and after experiments with explicit windows and decision rules, and impact tied to outcomes, with the causal caveats stated.

Coming in V3

Activity

What did the agent do, and why?

A chronological, auditable timeline of every trigger, action, result and verification, with rollback and daily or weekly summaries.

How Agent measures things

Every metric has a definition, so a number always means the same thing.

Metrics in Agent
Metric	Definition	Capability
Impact signal	Why it matters, for example high demand and a high competitive gap.	Opportunities
Confidence	How strong the evidence is.	Opportunities
Owner and status	Detected, investigating, recommended, and later stages.	Opportunities
Action class	The kind of change recommended.	Strategies
Affected scope	Pages, topics, intents, engines and markets involved.	Strategies
Expected measurement	The rates that will be compared after the change.	Strategies
Approval record	Who approved, when, the payload and the scope.	Actions and Approvals
Policy level	The permission tier for the workspace or brand.	Actions and Approvals
Guardrail events	Blocked actions, logged with the reason.	Actions and Approvals
Verification outcome	Verified, failed or unverified.	Verification
Affected scope	The topics and prompts re-run.	Verification
Delta	Change in mention, recommendation or citation rate with sample size.	Verification
Baseline and current	The two windows compared.	Experiments and Impact
Decision rule	The rule that decides the result, stored in advance.	Experiments and Impact
Outcome	Measured change with sample size and caveats.	Experiments and Impact
Event	One state change with actor, time and payload.	Activity
State	Detected, investigating, recommended, approval, executing, verifying, completed, monitoring.	Activity
Audit trail	The immutable record for privileged actions.	Activity

How Agent connects

At launch, you act on recommendations yourself. From V3, approved actions are executed, verified and measured.

Comes afterIntelligenceWhy is it happening?

Hands off toPerformanceWhat is happening?

Questions it answers

Where are the gaps worth acting on first?
What is the plan to close each gap?
How will we know whether it worked?
What did the agent do, and who approved it? (V3)

What it is not

Autopilot at launch: The first release is recommendation only. You make the changes yourself.
Promised outcomes: An opportunity is a hypothesis backed by evidence, not a guarantee.
Working without evidence: Every opportunity states the problem, evidence, scope and expected measurement.

Agent questions

Does the Agent change my website?

At launch the Agent recommends and you make the change. Execution is coming in V3 and will need explicit approval within a policy you set.

How will results be verified?

From V3, the changed page is checked and only the affected prompts are re-run for a before and after comparison.

Which CMS will be supported?

One will be chosen from early customer demand.

See Agent on your own brand.

Start with a first audit. We show where you appear, who is named instead, and what to fix first.

Get an AI Audit