What you will achieve
You will turn a campaign's evidence, content, publication and visibility records into three prioritised actions. Each action will have an owner, a reason and a clear completion check. The review is designed to move work forward without treating every observation as a new content request.
Use the relevant records in AdCyte alongside your existing website, analytics and project records. Keep the decision log in the system your team already uses. This guide does not assume a single dashboard contains every source of evidence.
Before you begin
- Define the campaign, products, markets and review period.
- Gather reviewed findings, approved outputs, public URLs and any unresolved factual questions.
- Bring AI visibility samples with their exact questions and methods, if these were collected.
- Include customer or website measures only where they are actually available and appropriately scoped.
- Invite the people who can make or own the likely decisions.
If you need to design the report itself, read What should a useful VideoGEO report tell you?. This guide starts with the records and ends with a work queue.
1. State the decision context
Write one sentence describing why the review is happening. For example: “We need to decide which three changes will most improve the usefulness and reliability of the UK storage campaign's product answers.”
Confirm what is in scope. A review of one campaign should not quietly become a judgement on the entire brand's AI visibility. Note any major change in product availability, campaign activity or measurement method that affects interpretation.
Agree the people who can resolve product questions and publication issues. A review without decision owners tends to produce recommendations addressed to a mysterious colleague called Someone.
Check: everyone is looking at the same campaign, period and decision.
2. Separate the record into four layers
Organise what you know into evidence, content, publication and observations. Evidence includes reviewed video findings and approved facts. Content includes drafts, approvals and unresolved wording. Publication includes actual destinations and checked live versions. Observations include sampled AI answers and separately measured customer behaviour.
Use a short record for each item: what happened, what supports the statement, its scope and any uncertainty. Do not collapse a draft into a publication or a citation into a referral visit.
This structure helps locate the cause of a problem. A weak answer may reflect a missing fact. A complete answer may still be unpublished. An accurate public page may not appear in a particular sample. Those situations call for different actions.
Illustrative campaign review
Keep the four layers distinct

Drafts, publications and AI observations support different conclusions.
3. Identify material gaps before counting activity
Review the priority customer questions. For each, ask whether there is a supported answer, whether it has been reviewed and whether the intended reader can reach it. Note gaps in product identity, qualifications, source quality or page usability.
Treat factual errors and misleading current information as a priority according to their actual consequence. A missing decorative image and an incorrect included-components statement should not compete solely on how easy they are to fix.
Then consider efficiency issues: repeated reconstruction of the same evidence, unclear approvals or external copies without an owner. These may justify a process change that prevents several future problems.
Check: the review can describe the underlying gap, not merely the metric or symptom that drew attention to it.
4. Inspect visibility observations without overreacting
Open the sampled answer behind any important observation. Check the exact question, provider, method, date and cited source where available. Determine whether the issue is an incorrect statement, an absent brand, an irrelevant source or simply a different answer to a broad question.
Compare only compatible observations. If the question wording or method changed, record that limitation. Keep failed checks outside eligible sample counts and avoid interpreting them as brand absence.
A citation to an old page may suggest a source-maintenance task. A missing brand may justify reviewing the question and available evidence before creating new copy. Repeating the same prompt until it says something agreeable is a poor substitute for an observation method.
Use the AI visibility signals article when the team needs to distinguish mentions, recommendations, citations and traffic.
Example priorities, not measured campaign outcomes
Choose a small work queue

Choose fewer than three actions if that is all the evidence justifies.
5. Turn each gap into a candidate action
Write actions as a change to a specific object. “Improve visibility” is an objective. “Review the current product answer's compatibility wording against the approved reference” is an action someone can complete.
For each candidate, record the evidence, expected benefit, dependency and owner. State how completion will be checked. Keep the benefit proportionate: correcting an answer can improve its accuracy; it does not guarantee that an external AI response will change.
| Observation | Candidate action | Completion check |
|---|---|---|
| Current page names the wrong variant | Correct the product answer after factual review | Live page matches the approved variant record |
| Useful demonstration has no written explanation | Add a reviewed explanation to the existing page | Reader can understand the sequence and inspect the source |
| Several outputs await the same factual decision | Assign and resolve the shared question | Approved decision recorded and affected drafts reviewed |
| AI sample cites an outdated public source | Inspect that source and correct owned information where appropriate | Publication change verified; later samples recorded separately |
6. Choose three actions by consequence and usefulness
Prioritise actions that resolve material inaccuracy, remove a blocking dependency or answer an important unmet customer question. Consider effort and ownership, but do not let easy cosmetic tasks displace a consequential factual correction.
Choose a mix that fits the situation. A campaign with inaccurate pages may need three corrective actions. A healthy pilot may benefit from one content improvement, one process change and one further observation. There is no requirement to fill a particular category for appearances.
If only two actions are justified, record two and explain why. The number three is a useful limit on focus, not a demand to invent work. Keep lower-priority candidates in a backlog with their rationale so the discussion does not have to restart next time.
7. Assign owners and review triggers
Each selected action needs one accountable owner, any contributing reviewers, a practical target date and a completion check. Use dates the team has agreed, not dates generated merely to make the table look finished.
For an observation task, specify the question and method to repeat. For a publication task, specify the destination and required approval. For a source task, specify the factual uncertainty that must be resolved.
Set the next review around completion or a meaningful new observation. A recurring meeting can help coordination, but a calendar invitation does not by itself create new evidence.
Worked decision record
The following example shows a possible review outcome, not a reported campaign result.
| Priority | Action | Owner role | Reason | Completion evidence |
|---|---|---|---|---|
| 1 | Correct the included-components answer for the current storage box | Product content owner | Current page could create a wrong expectation | Approved wording and checked live URL |
| 2 | Add a clear explanation of the folding sequence | Campaign content editor | Useful reviewed footage is absent from the product explanation | Published section linked to the correct demonstration |
| 3 | Recheck a defined question using the same supported observation method | Discovery analyst | Establish a comparable later sample | Full answer, method, date and source record |
Keep the third action separate from the first two. A later change in the answer may be relevant, but the sequence alone does not prove the page edits caused it.
Troubleshooting
| Problem | Action |
|---|---|
| The meeting produces twenty priorities | Select the most consequential few and retain the rest in a backlog |
| Nobody can inspect the underlying observation | Request the sample or source before drawing a strong conclusion |
| Every issue becomes a new article request | Check whether an existing page, reference or process needs correction |
| An action has several owners | Name one accountable owner and list contributors separately |
| The team wants a single success score | Preserve the distinct measures and decide what each can support |
Completion checklist
- Scope, period and decision are explicit.
- Evidence, content, publication and observations are separated.
- Material gaps have been inspected at source.
- Selected actions name a concrete change and a responsible owner.
- Completion checks are observable and proportionate.
- Uncertainty and lower-priority work remain recorded.
Next step
Explore AdCyte's AI Visibility workflow for inspectable observations, or use the content-update guide to complete a corrective action identified in the review.
FAQs
Should every review produce new content?
No. The most useful action may be correcting a fact, resolving an approval, improving an existing page or changing the observation method. Commission new content when it answers a distinct need with adequate evidence.
How often should we run this review?
Choose a rhythm that matches the campaign's activity and the rate of meaningful change. Review after important publications, product changes or a planned observation period. Avoid repeating the meeting when there is no new information or completed action to assess.
Can we call an action complete if AI answers have not changed?
Yes, when the action was a verified source, content or publication improvement. Record external observations separately. A completed correction does not guarantee that every service has retrieved or used the updated information.



