Move from occasional screenshots to a comparable evidence register.
Marketing teams can manually check AI answers, but ad-hoc prompts and screenshots are difficult to compare. Verify Presence freezes the question set, retains the answer context and ties proposed changes to public evidence.
The operating problem
One surprising answer can trigger reactive work without showing whether it is representative.
- Prompts change between checks.
- Screenshots omit model, date or source context.
- Teams cannot distinguish factual error from non-mention.
- Corrections lack approval history.
Example prompt set
Questions are agreed with the brand and buyer journey in mind.
- Category discovery questions
- Problem-and-provider questions
- Location or segment variants
- Accuracy checks for material brand facts
Evidence reviewed
The system uses approved public sources rather than internal assertions by default.
- Brand, product and service pages
- Structured organisation information
- Approved proof and case examples
- Public policies and contact routes
Likely gaps
Results are prioritised by evidence and buyer impact.
- Inconsistent category language
- Unsupported proof point
- Old domain or rename residue
- Missing explanation for a key use case
Deliverables
The register can fit an existing content and governance process.
- Prompt and answer history
- Accuracy and mention classification
- Evidence-backed intervention backlog
- Approval and remeasurement record
Collaboration boundary
Verify Presence can work alongside an agency or internal team without claiming control of AI outputs.
- Existing publishing ownership can remain
- Sensitive claims retain human approval
- Results stay bounded to the measured set
What to know before you begin
Can our agency use the output?
Yes. Access and approval roles should be agreed so the accountable business retains control.
Does the system publish every recommendation?
No. Sensitive or material public changes require sufficient evidence and explicit authority.
Can prompt sets evolve?
Yes, but comparability requires versioning. New prompts should not be silently mixed into an earlier baseline.