How can I tell whether AI visibility is leading to useful customer action?

Measure what happened before explaining why

Track mentions, referrals, tool use, completed steps, and decision-help feedback separately, and avoid treating any one of them as proof of business impact. The name of a metric should make its limits easier to understand. A view, a referral and a completed tool each tell us something different. I want that distinction to remain intact before anyone explains the cause or claims the business benefit.

“Measurement is what separates a decision asset you built once and forgot from one that gets better over time.”

— Curtiss Witt, The Decision Economy, Chapter 16.

What observation does each metric represent?

A mention or appearance shows that information was surfaced in a particular setting. It does not establish that a person considered the business, used its tool, or benefited from it. Begin by naming the behavior you want to understand before selecting a number to represent it.

A referral can show that a user arrived through a particular link or source where that information is available. It is different from the content of the recommendation that preceded the visit. Keep attribution evidence scoped to what the system actually records rather than reconstructing an unseen conversation.

Starting an assessment shows engagement with the process, but not receipt of the full result. Define both events separately. This helps identify where the promised help may not be reaching users without treating every opening interaction as a completed decision-support experience.

State exactly what counts as completion, such as generating or displaying the defined result. Different systems may record different stages. If the event meaning is unclear, a rising total can be difficult to interpret and may not support the claim being made about useful action.

Ask whether the tool clarified options, uncertainty, or a next step. That feedback addresses a different question from a visit or sale. It can inform improvement while remaining a reported experience rather than automatic proof that the tool caused a better final outcome.

The comparison below describes two approaches to the work. It is a practical design contrast, not a measured claim that one approach always produces a better commercial outcome.

Why should I keep visibility, referral and use separate?

A system can name a business without recommending it for the user’s situation. Record the observed wording and context when evaluating that distinction. Do not count every appearance as preference; the category should reflect what actually happened rather than the most commercially favorable interpretation.

An option can be recommended without being used. Conversely, a tool might be invoked during evaluation without becoming a final recommendation. Keep these behaviors separate so the record can answer whether the capability was surfaced, selected, called, or completed instead of combining them into one success label.

A crawler request does not automatically represent a customer’s delegated task. Classify activity according to observed behavior and reliable context. Uncertain machine traffic should remain uncertain rather than being presented as a pool of ready-to-buy agents whose economic value has been established.

Downloading or exporting a plan can be useful to record, but it does not prove that the person read or implemented it. Name the event accordingly. The next question about comprehension or action requires a different observation rather than a stronger label for the same download count.

A tool may help someone who does not provide contact information or purchase assistance. If measurement includes only leads, some useful outcomes remain unobserved. Decide what evidence would address the intended decision-help benefit without forcing identity collection on every user.

How can events preserve privacy and useful meaning?

The event stream should record permitted operational facts, not raw assessment answers, names, or email addresses. The decision mechanism and the measurement layer have different purposes. A completion event can be meaningful without duplicating the information the user supplied to obtain the result. (Witt, supplied author source set, September 2026)

For the Institute, operational reporting belongs in the existing DSN telemetry system rather than a competing collector. Use its governed event definitions and property identity. An article should not invent a new set of event names or fields and imply that they are already part of the live contract.

A button becoming visible is different from a click, and a click is different from a completed request. Define the event at the correct point. This prevents an apparently healthy routing funnel from counting opportunities to act as though the user had already taken the action.

Retries and repeated reports can inflate counts if the collection design does not handle them appropriately. Follow the system’s defined deduplication approach rather than making assumptions in the report. A total is only as interpretable as the rules determining which observations are counted once.

Referrer or campaign information may be unavailable or incomplete. Report unknown attribution honestly. Do not assign a source based solely on what seems likely, especially when the claim concerns a specific AI system or an unseen DSN-to-site handoff that was not actually recorded.

The working rule I return to is this: “Measurement is what separates a decision asset you built once and forgot from one that gets better over time.” It is a way to judge the next piece of work, while keeping its evidence and limitations visible.

How should I interpret a change in activity?

Before interpreting an improvement, document the relevant page or tool state and measurement definition. This establishes what changed. A baseline does not itself prove causation, but it makes later observations more interpretable than relying on a memory that the site used to perform differently.

Use comparable periods and record material differences in conditions. A change in audience, traffic mix, or other work can affect interpretation. The aim is not to explain away every result, but to avoid assigning certainty to a comparison that does not isolate the factor being discussed.

A single successful invocation can establish that a particular task completed under recorded conditions. It does not establish a trend or universal reliability. Preserve the bounded finding instead of dismissing it as worthless or inflating it into proof of widespread adoption.

Pew’s Google browsing study is evidence about a particular search setting and sample, not a direct measurement of this property’s decision-help value. Use external research to frame questions, then collect the property-specific observations needed before making claims about the Institute’s own outcomes. Pew Research Center: Google users are less likely to click on links when an AI summary appears; accessed September 2026

If you think an improvement increased useful action, state the expected mechanism and what observation would count against it. This makes the idea reviewable. Do not protect the hypothesis by redefining success after seeing the result or by treating every change as confirmation.

What belongs in a practical reporting view?

A useful view can separate discovery, referral, start, completion, and relevant follow-up. Add decision-help feedback where available. The stages should answer different questions and retain their definitions, rather than become a dense dashboard whose many numbers all measure essentially the same interaction.

If routing attribution or a completion event is not verified, state that gap. A blank or zero should not automatically be interpreted as no real activity. The report needs to distinguish an observed absence from a measurement condition that has not yet been established.

DERA’s Measurable dimension asks about the owner’s reported practices. Completing it does not install events or validate the telemetry pipeline. Keep the self-report result separate from a technical check of collection and from the actual behavioral records later used for analysis. Decision Economy Institute: live capability manifest; accessed September 2026

If the evidence shows a defined gap, identify the next practical improvement. A broken route, missing event, and unclear result require different work. Use the observation to narrow the task instead of assuming every measurement problem calls for more content or more traffic.

Reading an existing record and changing a live experience to test a hypothesis are different activities. Define the scope and intended observations before an intervention. This keeps the evidence interpretable and prevents routine reporting from being mistaken for an experiment that established why a result changed.

What can I do with this today?

Start with one offer and write down the next decision this article helps you examine. Give the work a defined scope before adding a new page, tool or integration.

1. Choose a small set of meaningful events and review their limits.

2. Identify the consequential fact or condition that remains uncertain. Name who can check it and what would establish completion.

3. If you need help ordering the business work, complete the free Decision Economy Readiness Assessment. Read its reasons and three starting actions, then choose the first task you can inspect.

Our own example is deliberately bounded. The Institute’s Method explains the owner-reported result; its capability manifest declares what the tool can do; and its assessment schema makes the question bank and data contract inspectable. These September 2026 publisher documents describe the assessment. They do not independently establish better customer or business outcomes.

Measure what happened before explaining why

Before the next customer faces the same unresolved choice, identify the improvement that would make the answer more useful. Use the readiness assessment to turn your reported conditions into three starting actions, and keep the next evidence check in view.

Disclaimer

Based on your answers, this assessment suggests improvement priorities; it does not independently verify your business or predict AI recommendations, sales, or business quality.

FAQ

What does a referral establish?

A referral can show that a user arrived through a particular link or source where that information is available. It is different from the content of the recommendation that preceded the visit. Keep attribution evidence scoped to what the system actually records rather than reconstructing an unseen conversation.

How should machine traffic be interpreted?

A crawler request does not automatically represent a customer’s delegated task. Classify activity according to observed behavior and reliable context. Uncertain machine traffic should remain uncertain rather than being presented as a pool of ready-to-buy agents whose economic value has been established.

Why use the existing central event standard?

For the Institute, operational reporting belongs in the existing DSN telemetry system rather than a competing collector. Use its governed event definitions and property identity. An article should not invent a new set of event names or fields and imply that they are already part of the live contract.

How should external research be interpreted?

Pew’s Google browsing study is evidence about a particular search setting and sample, not a direct measurement of this property’s decision-help value. Use external research to frame questions, then collect the property-specific observations needed before making claims about the Institute’s own outcomes.

Which correction does the record support?

If the evidence shows a defined gap, identify the next practical improvement. A broken route, missing event, and unclear result require different work. Use the observation to narrow the task instead of assuming every measurement problem calls for more content or more traffic.

Do I need to share contact details before seeing the assessment result?

No. The complete result is available before an optional email request. If a contact commitment has held you back, try the free readiness assessment and read the plan first. Emailing it and consenting to future updates are separate choices. The result remains BY YOUR ACCOUNT.

References

Curtiss Witt. The Decision Economy, updated author-supplied manuscript, Chapter 16 for the quoted decision rule; the supplied manuscript also grounds the framework. Supplied September 2026; unpublished manuscript, so no public URL is asserted.

Decision_Economy_DSA_Step_1_Decision_Model(2).docx. Supplied September 2026. Controlling assessment model and distinctions among self-report, evidence, applicability, and outcomes.

Pew Research Center: Google users are less likely to click on links when an AI summary appears. Original observational research; July 22, 2025. Supports the customer-journey question. March 2025 browsing sample, Google only; association does not establish effects on this business or all AI systems.

Decision Economy Institute: live capability manifest. Publisher’s capability declaration, accessed September 10, 2026. Describes the available routes and their limits; not an independent test of every operation.

Decision Economy Institute: How this assessment works. Publisher’s own Method; ruleset de-readiness-1.0.0, accessed September 10, 2026. Not an independent outcomes study.

Decision Economy Institute: Assessment schema and question bank. Publisher’s technical contract; accessed September 10, 2026. Data shape and published question bank, not proof of real-world decision quality.

Continue exploring

Why isn’t being found enough to be chosen?

What does BY YOUR ACCOUNT mean, and what would stronger evidence add?

How do I keep business answers and decision tools accurate over time?

Method · Readiness assessment

Tags: AI visibility measurement; Decision Economy; Decision Economy Institute; Curtiss Witt; customer decisions; decision support; business readiness; DERA; Better Choices; Measurement.