Make the first measurement useful
Start with the decision your team needs to make: where to invest, what questions customers ask or whether the brand is represented accurately. That determines the sample. A handful of interesting answers can reveal questions worth investigating, but it cannot establish your share of the entire AI-search market.
Design the sample before collecting answers
The prompt register separates discovery, comparison and validation journeys, and records why each prompt belongs. Keep branded and unbranded tests separate. Agree a stable comparison sample and a repetition plan, with exploratory questions recorded alongside it.
Keep the evidence behind every score
The log records the exact prompt, collection conditions, full answer reference, cited sources, competitors and factual accuracy. A mention, a recommendation and a citation answer different questions. Record unknown conditions and ambiguous answers explicitly.
Give the finding an owner and a test
The action template connects the observation to its evidence, alternative explanations, a proposed change and the person able to implement it. It includes acceptance criteria and a plan for repeat measurement.
Preview a finding
Illustrative example for a fictional business. This is a demonstration of the delivery format, not a client result.
- Observation
- In the saved answer to a comparison prompt, a competitor is recommended with a cited product guide; the fictional brand is absent.
- Evidence to retain
- Prompt ID, run date, model and search mode, full answer, citation URL and relevant source excerpt.
- Interpretation
- The cited guide answers a buying question. One answer does not establish that creating a similar guide will cause inclusion.
- Next action
- Check whether the question recurs in the sample and whether the brand has accessible, accurate evidence answering it. Assign a content owner if a gap is confirmed.
- Validation
- Verify the published improvement, then rerun the unchanged sample under recorded conditions and report variation alongside any change.
How to use the download
Open the Markdown file in a text editor, or copy the templates into your preferred document or spreadsheet. Duplicate the evidence record for each answer. Begin with a pilot to identify collection and scoring problems before committing to a broader programme.
This starter provides a structure for a pilot. A full baseline also needs research tailored to the business, a defensible sample, source investigation, technical evidence and human review. Platform behaviour changes; keep test conditions and limitations visible.
You can adapt the worksheets for internal work and client delivery. The download includes use terms and links to platform guidance. It contains no confidential client material.