Next steps · No. 08 / 11
Stage 5: How should you structure a 90-day AI visibility pilot?
Last updated
Use 90 days as a decision window, not a promised timetable for recommendations or revenue. Limit the programme to a small number of testable interventions.
Recommended pilot schedule
| Period | Work | Required output | Gate |
|---|---|---|---|
| Days 1–14 | Confirm buying questions, inspect existing reporting, collect baseline and review accuracy | Prompt register, evidence archive, issue log and pilot charter | Do not expand if the commercial question or measurement method remains unclear |
| Days 15–30 | Prioritise and deploy the first approved changes | Change log, acceptance tests and deployment dates | Pause if implementation access or approvals are unavailable |
| Days 31–60 | Repeat comparable observations and check implementation quality | Updated evidence, failure report and progress against hypotheses | Repair the method or revise the hypothesis if results are uninterpretable |
| Days 61–90 | Review matched evidence, cost and commercial signals | Continue/change/stop memo and exportable handover | Renew only against an explicit next decision and agreed scope |
What belongs in the pilot charter?
Complete these fields before work starts:
- Decision: what the budget owner will decide at the end.
- Hypothesis: the specific intervention and expected observable change.
- Scope: product, customer segment, surfaces and fixed prompts.
- Primary measure: for example, recommendation frequency on the agreed requirements prompts.
- Secondary measures: accuracy, owned citations, implementation completion and relevant analytics.
- Comparison: untreated pages or prompt groups where practical, with their limitations recorded.
- Minimum worthwhile change: agreed before seeing post-intervention results and assessed against baseline variability.
- Guardrails: factual accuracy, conventional-search performance, site usability and approved claims.
- Owners: budget, analysis, product facts, publishing and validation.
- Cost cap and exit: maximum commitment, review dates and export rights.
If the same change affects both the intervention and comparison groups, do not call the latter a clean control. If the collection method changes, show separate series rather than silently joining them.
How much should you budget?
Build the budget from the scope; do not adopt an unverified “typical GEO retainer”.
Use:
Pilot cost = diagnostic + monitoring and add-ons + implementation + internal time + validation + applicable taxes and contingency.
One current pricing example illustrates the importance of add-ons: OtterlyAI lists its Standard monthly plan at $189 for 100 prompts, with Google AI Mode as an additional $59 monthly on that tier. That combination is $248 per month, or $744 over three monthly billing periods, before other charges. This is a supplier-listed monitoring example checked on 16 October 2026—not an endorsement, implementation budget or all-market price benchmark. (otterly.ai)
For your budget worksheet, obtain a quote against the agreed surfaces and collection design. Confirm whether the package can support the required repetitions, exports and review workflow; a prompt allowance alone is not a measurement specification.
Sources cited on this page
- otterly.ai otterly.ai/pricing