Direct questions · No. 28 / 31
28. What should a defensible 90-day AI visibility pilot look like?
Last updated
Build the pilot around one commercial hypothesis, a fixed baseline and a small set of implementable changes. The end-of-pilot decision should be whether to continue, revise or stop—not whether a chart can be made to look positive.
The following is a proposed management design, not a guaranteed-results timetable:
| Period | Work | Required output | Decision gate |
|---|---|---|---|
| Days 1–14 | Validate buying questions; inspect first-party reporting; establish collection conditions | Prompt register, baseline and failure log | Is there a repeatable, relevant problem? |
| Days 15–30 | Diagnose access, content, accuracy and source gaps | Prioritised backlog with owners and hypotheses | Is the proposed work justified by evidence? |
| Days 31–60 | Publish approved changes; maintain comparable measurement | Deployment log and source-level verification | Was the work completed correctly? |
| Days 61–90 | Reassess equivalent windows; inspect commercial observations | Validation report and next-step recommendation | Continue, revise or stop? |
Before launch, specify:
- The customer segment and purchase situation.
- The primary visibility measure and accuracy guardrails.
- The prompt families that must remain fixed.
- The conditions that would make the comparison invalid.
- A practical improvement threshold worth paying for.
- A spending cap and internal time allowance.
Do not declare failure solely because a long sales cycle has not completed. Equally, do not declare success solely because more pages were published or more prompts were tracked.