Skip to content
WaveWise

The closed loop, step by step

A shopper asks an AI which product to buy, the answer names someone else, and it gives a reason. WaveWise works on that reason — from what you already know, to a result stated at whatever grade the evidence supports, which is often very little.

It starts with what you already know

You know things about your products that no crawler can infer: which material you actually use, which claim you can stand behind, which page is out of date and which one you rewrote last month.

That is the input. Not an export of your customers, not access to your systems, not a feed. What you publish, and what you can say is true about it.

  • What you sell, described the way you would describe it to a buyer who asked.
  • Which of your public pages carry that description, and which of them no longer do.
  • What you are willing to change, and what is not on the table.

Watch the question, not the ranking

A buying question is a sentence a shopper types before they are ready to decide. We look at what an AI answers when it meets one, which product it names, and which reason it gives.

What we are reading is the reason. A position in a list tells you that something happened; the reason tells you what happened, and it is the only part of the answer you can act on.

Find the reason behind the answer

An answer leans on sources. Some of them are yours, some are not, and some are yours from two years ago. The diagnosis names which source the reason came from and where that source is stale, thin or wrong about you.

This is where most of the value is, and it is also where it would be easiest to overstate. A diagnosis is a reading of what a source says, not a claim about how any model weighs it.

One change, written out before anything moves

A diagnosis becomes exactly one proposed change. Written in full: which asset, which words, what it is meant to correct, and what would tell us afterwards whether it did.

One at a time, and reversible. A batch of simultaneous changes cannot be read afterwards — whatever happens, you will not know which change did it, and neither will we.

Nothing happens until you approve it

You read the mission and you decide. Approving it is the only thing that causes anything to happen, and what happens is exactly what the mission said — no wider, no extra assets touched, no scope discovered along the way.

If you approve nothing, nothing ships. That is a normal outcome of a review, not a stalled one.

Shipped, then seen, are two different questions

After a change ships we check two things separately. Did the approved action actually happen, in full, to the asset it named? And then: did the environment we care about have any chance to see it?

The first is ours to answer and we answer it plainly. The second often comes back unknown, and unknown is the honest answer, not a sign that something went wrong.

What the evidence supports, and what it does not

Every result is reported at a grade, and the grade is a ceiling on what anyone may say about it — including us, in a deck, later.

A change that shipped and was observed says something about that surface. It does not say a shopper chose you, and it never says money moved. Where the chain of reasoning stops, the report stops.

Then the loop closes

What was learned decides the next mission: a diagnosis that held gets extended, one that did not gets dropped rather than dressed up.

A loop ends with a decision and starts again from what it learned.

The same argument, in reference form

Nothing new below — the argument above, itemised. Each block states its own scope.

  1. Define the Outcome

    You say what result you want and what must not be touched.

    Not: Show me the data first.

  2. Observe

    We look at a real AI buying decision: who got named, on what reasoning, from which source, and where it was wrong about you.

    Not: Track a ranking.

  3. Diagnose

    We state a reason that can be proven wrong, with the evidence for it, the evidence against it, and what would falsify it.

    Not: An AI hands down a conclusion.

  1. Decide

    Together we pick one change, weighing value, control, reversibility and risk against what it would teach you.

    Not: Produce an optimisation checklist.

  2. Run One Controlled Mission

    You approve it and you ship it. One low-risk change you can roll back.

    Not: The system goes and edits things by itself.

  3. Verify Execution & Exposure

    We check two separate things: that the change really happened, and that the target environment really had a chance to see it.

    Not: It went live, so it is done.

  1. Measure the Result

    We report a direction, an evidence grade, the ceiling on what that grade lets anyone say, and the side effects.

    Not: Here is your score.

  2. Learn

    The result goes back to the reason we gave: it supports it, weakens it, refutes it, or cannot be attributed yet.

    Not: Cannot be attributed means it failed.

  3. Choose the Next Mission

    You choose what happens next: continue, hold, decline, or widen it.

    Not: It rolls straight into another round on its own.

What each grade permits anyone to say. The last column marks the only rungs anything on this page stands on.
GradeThe most it permitsShown here
E0The intended change happened.Yes
E1An offline or controlled evaluation improved.Not claimed
E2A real AI surface changed.Yes
E3A credible control supports a lift on an AI surface.Not claimed
E4A downstream user or revenue metric moved — never called incremental by default.Not claimed
E5Incremental economic value is demonstrated under guardrails.Not claimed

ANONYMOUS_FIXTURENON_LIVEEVIDENCE_MODE: PLACEHOLDER_INTENT_ONLY

When the surface moves

An anonymous consumer brand, one flagship product, one specific question shoppers keep asking before they buy.

One low-risk, reversible edit to the information on the product page.

RESULT_DIRECTION
POSITIVE
EVIDENCE_GRADE
E2
EXECUTION
COMPLETED
EXPOSURE
OBSERVED
CLAIM_CEILING
The most this permits: a real AI surface changed. It does not say anything about orders, conversion, revenue or profit.
LIMITATIONS
  • One run, one product, no control group.
  • Other explanations cannot be ruled out.
NEXT
A real choice: continue, widen the change, or test a different reason.

ANONYMOUS_FIXTURENON_LIVEEVIDENCE_MODE: PLACEHOLDER_INTENT_ONLY

When we cannot tell you yet

A comparable anonymous brand and product, a different buying question.

Again one low-risk, reversible edit.

RESULT_DIRECTION
INCONCLUSIVE / VERIFICATION_OPEN
EVIDENCE_GRADE
E0
EXECUTION
COMPLETED
EXPOSURE
UNKNOWN
CLAIM_CEILING
The most this permits: the intended change happened. The change being live is not evidence that anything saw it, so nothing beyond that is claimed.
LIMITATIONS
  • We could not confirm the target environment had a chance to see the change.
  • Because exposure is unknown, any effect and its absence are equally unexplained.
  • This is an open verification, not a failed mission and not a broken system.
NEXT
What the next evaluation would need before this question can be answered.
How neighbouring categories behave, and where each of them stops.
CategoryWhat it gives youWhere it stopsWhat we add
AI visibility trackingHow often and where your brand is mentioned in AI answers.The current state. It does not explain why, and it moves nothing.Turns the why into one change you approve and ship, then comes back and verifies it.
AI content generationMore descriptions, more FAQs, more assets.Output. Whether the target environment saw any of it is never checked.One change at a time, with execution and exposure verified separately and the result graded honestly.
Monitoring and BI dashboardsA wall of metrics.Handing it to you. Every action happens outside the system.Deciding and running are inside the loop, one step at a time, and reversible.
  • They tell you how things are. We change one thing with you, then come back and tell you how things are after.
  • They treat published as finished. We treat published and seen as two things that have to be verified separately.
  • Their conclusion always holds. We are willing to conclude that this one cannot be attributed yet, and to say what would settle it.

What this page is not claiming

  • Nothing here reads live data. This is a static explanation of how the loop works, not the product running.
  • We do not claim to know how any AI system ranks products, and we do not work on the parts of an answer you cannot control.
  • A verified execution is not a verified outcome. Where a result cannot be attributed, the report says so instead of rounding it up.