Review citations and intelligence

Citations and Intelligence answer different questions. Citations shows what was observed when approved buyer questions were put to AI models; Intelligence points to opportunities supported by customer, demand, and storefront coverage evidence.

Use them together to choose a specific improvement for Northgate ESD Packaging, then verify that improvement through its page or product workflow.

#Read an observation in context

1. Select the intended account and storefront.
2. Open AEO & Agentic Storefronts > Citations.
3. Identify the run and prompt-set version behind the observation.
4. Read the buyer question and the recorded brand or source-URL evidence.
5. Compare observations only after checking their dates and prompt context.

The Citation Observatory uses approved, versioned prompt sets. Editing prompts creates a new draft version; approval replaces the previously approved version. This keeps an observation tied to the exact questions that produced it.

If Northgate appears in an answer to a question about ESD packaging suppliers, that is a point-in-time observation. It is not proof that Northgate appears for every buyer, every wording, or every future run. Likewise, an absent brand in one response does not establish that all relevant content is invisible.

#Keep three kinds of evidence separate

A citation is not a click. A click is not a form submission. A form submission is not necessarily a completed order. Preserve these distinctions when reporting progress to the Northgate team.

Changing a page and later seeing a mention can be useful evidence to investigate, but it does not by itself establish that the edit caused the mention. Prompt changes, timing, and model variation also affect comparisons.

#Use Intelligence to choose work

1. Open Intelligence.
2. Read an opportunity's evidence and source counts.
3. Open the linked page or product action.
4. Confirm that the evidence is relevant to the current offering.
5. Prepare a narrowly scoped improvement for review.

For example, repeated buyer questions about choosing packaging dimensions could justify clearer product guidance if the corresponding product lacks that answer. The responsible owner should supply the actual dimensions or selection rules. The opportunity identifies useful work; it does not supply permission to invent missing facts.

Intelligence draws on ideal customer profile, CRM, demand, and coverage evidence. Its purpose is to make the reason inspectable rather than ask you to trust an unexplained AI score. Read the evidence before treating the suggested priority as a business decision.

#Understand run availability

Observatory runs require an approved prompt-set version and enabled service controls. Runs reserve spend against the account's monthly AI budget before model calls and reconcile actual usage afterward. Do not start extra runs simply to refresh a screen; confirm the intended question set and account budget with the responsible operator.

An empty view can mean no eligible run has completed. It does not automatically mean zero visibility. Check whether prompts were approved, the service was enabled, and a run produced observations before drawing a conclusion.

#A useful weekly review

Pick a consistent question set, review the latest observations, select one evidence-backed gap, and hand it to the appropriate page or product owner. Record the actual content change separately from later visibility and traffic observations.

A strong Northgate update says which question was tested, what the run observed, and which verified improvement followed. Avoid a broad claim that Forge “increased AI rankings” when the evidence is a small set of model responses.

#Related articles

• Prioritize storefront work
• Read visitor journeys and attribution
• Review page evidence and JSON-LD