VercatusAI Visibility Audits for Agencies: A Monthly Client Workflow
Run a client audit from authorization and baseline to evidence-backed tasks and a monthly review. Keep access, citations and business results separate.
By Norbert Szabo, founder of Vercatus
Published · Updated · 6 min read
A useful monthly AI visibility audit gives a client three things: a reproducible baseline, a short implementation queue and evidence of what changed. A dashboard screenshot without the questions, dates and next actions leaves the client unable to assess progress.
In short
- Agree access, scope and ownership before the first run.
- Keep one stable question panel per market and compare like with like.
- Turn findings into scoped tasks with an owner, an acceptance test and a recheck.
- Report implementation, verification, answer visibility and business outcomes separately.
Use this workflow as an operating template. It is not a customer case study, and the sample task is fictional.
Start with authorization and access boundaries
Verify each client domain before running a full analysis. Record the responsible owner, supported markets, business goals and who may receive reports. Decide whether clients belong in separate workspaces or have site-limited access in a shared workspace.
Vercatus supports site restrictions for analysts and viewers, and a client role confined to selected sites. Admins and owners have broader workspace access. An unrestricted viewer should not be described as “client-only.” Review the team and access guide before inviting anyone. White-label branding changes presentation, not permissions.
Keep a stable question panel per market
Collect customer buying questions, support questions and observed search queries. Label branded and non-branded questions, funnel stage, language, country and intended source page. Record competitors explicitly.
Start with a manageable panel your team can review. Keep its wording and observation conditions stable when comparing months. New questions can join a separate discovery cohort; silently replacing the original panel makes a score change hard to interpret.
Use the citation measurement guide to define citations, mentions, unanswered requests and denominators. A retrieved page without an observed citation is a different result.
Turn findings into scoped work
Inspect access failures before recommending more articles. If a page is readable, identify the missing answer, contradictory product fact or weak source evidence. Name the canonical page that should own the question.
A fictional handoff might read:
| Task field | Example |
|---|---|
| Reader question | Does the product support invoices in the buyer’s currency? |
| Target | Existing billing help page |
| Evidence | Current page explains payment methods but omits supported currencies |
| Change | Add the verified currency list and exceptions beside the billing explanation |
| Owner | Product editor, with billing-team review |
| Acceptance | The answer exists in initial HTML and agrees with the checkout policy |
| Follow-up | Recheck the page and the unchanged billing question panel |
Keep the evidence attachment and implementation date with the task. Vercatus’s ticketing workflow turns a play into an issue, with its evidence and acceptance check, rather than requiring the agency to transcribe the result.
Review four outcomes separately
- Implementation: Was the requested change actually published?
- Verification: Did a later run confirm the relevant access or content fix?
- Answer visibility: Did comparable observations cite or mention the client differently?
- Business outcome: Did attributable visits or qualified enquiries change?
A confirmed technical fix is progress even if citation observations remain unchanged. A citation increase alongside several campaigns is not proof that one edit caused additional revenue. State unavailable data and changes in sample size.
Present a report the client can act on
Open with the decisions: what was fixed, what remains blocked and which two or three tasks deserve the next month’s budget. Show the measurement period, markets, engines, successful observations and evidence links alongside any share.
Use white-label reports where the plan supports them, but retain the methodology and source references. A branded report should still explain how its figures were calculated. The archived audit of our own site is Vercatus’s own-site evidence, not a claim about an agency customer’s results.
Close the month with ownership
Agree which tasks the client will implement and which the agency owns. Record decisions to leave a finding unchanged. Check report-link access and revoke unused keys or memberships when an engagement ends.
A three-month cadence
| Month | Focus | Deliverable |
|---|---|---|
| 1 | Baseline: verify domains, set the panel, read the first measurements, fix access | A short report with the panel, the evidence and the first queue |
| 2 | Implement the top tasks; recheck acceptance; start separating discovery questions from the fixed panel | A report that says what was published, what was verified and what is blocked |
| 3 | Compare like with like; review outcomes against the business objective | A report with decisions for the next quarter |
The point of the cadence is that each month ends with a decision someone owns.
A monthly audit checklist
- Authorisation. Domains verified; owners named; who may receive reports.
- Panel. Questions listed with market, language, funnel stage and intended owner page; competitors named.
- Access. Crawler access, delivery and readable content checked per host.
- Findings. Each one has evidence, a target page and a proposed change.
- Tasks. Each has an owner, an acceptance check and a due date.
- Verification. The acceptance checks were run and recorded.
- Observations. The same panel was asked again; failures are counted.
- Report. Decisions first, evidence behind them, methods stated.
How to run the client relationship
- Separate implementation from outcome. “The page was published” and “the answer changed” are different statements, and the client should never have to guess which one you mean.
- Use roles. Analysts and viewers can be limited to selected sites, a client role is confined to selected sites, and branding does not change permissions (team and access).
- Hand off cleanly. Plays can be handed to the people who ship them as tickets or to a coding assistant, with the evidence and the acceptance check attached (ticketing).
- Close the engagement. Revoke unused keys and memberships, and expire shared report links.
Common mistakes
- Replacing the question panel in the middle of a quarter.
- Reporting a share without its denominator.
- Promising citations instead of fixes.
- Putting all clients in one workspace without site restrictions, or calling an unrestricted viewer “client-only”.
- Leaving no record of a finding the client chose not to act on.
Questions people ask
Should each client get a separate workspace?
Use separate workspaces where that is the boundary the client requires; use site-limited roles where a shared workspace is acceptable.
How many questions should a panel have?
Enough to represent the buying journey and few enough that your team reviews them. Say what you chose.
What do I do when a month shows no change?
Report it. A verified fix without a change in citations is still progress, and an unchanged panel is information.
The evidence behind the checks
- Access comes first. An answer engine can only use pages its fetchers can retrieve, and the CDN counts as much as robots.txt (crawler access, grade A).
- Check the raw HTML. The AI crawlers measured in a December 2024 analysis did not execute JavaScript, so content added by script may be missing for them (raw HTML for AI crawlers, grade C).
The agency setup covers the cross-site workflow; the ROI guide helps build a separate business-outcome model. Together, they support a monthly review that distinguishes completed work from observed visibility and commercial impact.