Use these templates to correct inaccurate information, test improvements, maintain change records, qualify inquiries, and decide where to invest effort. Link them to Part C’s evidence logs and metric dictionary, Part D’s approved facts, Part F’s content records, and Part G’s permissions and professional reviews. Use Part E when a change requires technical verification.
Track what changed, what the evidence shows, and what management will do next. Set targets and review dates to suit the firm’s services, capacity, and sales cycle; these templates prescribe no universal performance benchmark.
Shared record fields
[Record ID; practice/service; owner; related asset, prompt, inquiry, experiment, and approval IDs; open date; status; next action; accountable person; due date; evidence location].
H.1 Correction request template
Record the error and its consequences
[Exact statement or omission; source URL or answer capture; date/time; affected firm/person/service; verified replacement fact; supporting evidence and effective date; audience; consequence; priority; requested remedy].
For an AI answer, attach the Part C evidence record: exact prompt and version, platform and mode, relevant settings and location, response, cited sources, and capture date. Record whether the issue appeared again; one capture establishes what occurred in that instance.
Prioritize errors by consequence and exposure. Examples include incorrect credentials, misleading service claims, wrong contact details, and disclosure of restricted material. Use Part G’s handling process for sensitive evidence. An unfavorable ranking, missing recommendation, or opinion is not automatically a factual error.
Choose the correction route
- Firm-controlled information: Correct the approved source and affected versions through the existing publication process. Update the relevant Part D or F record.
- Third-party page or listing: Identify the responsible publisher and its correction route. Supply the minimum evidence needed to verify the requested change.
- AI answer: Review any cited source, correct sources within the firm’s control, and use the provider’s available feedback route where appropriate. Keep the answer issue linked to any source issue.
Necessary factual corrections should proceed under existing authority without waiting for an experiment. Confirm the sender’s authority before requesting changes on behalf of the firm or a professional.
Suggested request wording
At [URL/location], the statement “[exact wording]” is inaccurate as of [date]. The verified fact is [replacement], supported by [public source or approved evidence]. Please [specific correction]. This request concerns [identified factual issue]. Please confirm the action taken or advise what further evidence is needed. Contact: [authorized contact].
Track submission and verification
[Recipient/route; submitted wording; submission date; acknowledgement; response; source-change evidence/date; answer-recheck evidence/date; follow-up owner/date; closure reason].
Use Logged / Evidence confirmed / Submitted / Source corrected / Answer rechecked / Closed with reason, as applicable. Record the result of each check, including whether an error persists. Submission, acknowledgement, source correction, and a satisfactory answer recheck are separate events.
Close the firm’s action when the agreed work is complete or another documented outcome applies. Retain unresolved external issues for proportionate monitoring; do not promise that one correction will change every future answer.
H.2 Experiment brief
Hypothesis and proposed change
[Experiment ID; decision to inform; audience and service; observed problem; proposed change; expected effect; explanation; owner; budget and staff hours].
Write the hypothesis as:
For [defined audience/questions], changing [specific element] is expected to improve [primary measure] because [reason], while maintaining [required quality and operational conditions].
Illustrative hypothesis: Adding an approved explanation of service scope and exclusions to selected service pages will reduce unsupported scope claims in a fixed sample of relevant answers.
Design before launch
Record the following before observing the result:
- Eligible units and comparison: [Pages, users, inquiries, or question families; selection method; treatment; comparison group; assignment method; baseline differences].
- Primary measure: [Part C definition; numerator/denominator; baseline; smallest improvement worth acting on; required evidence].
- Other conditions: [Accuracy, permissions, usability, inquiry quality, review effort, cost, and delivery capacity; thresholds that trigger a pause].
- Observation plan: [Start/end; review points; collection schedule; sample rationale; expected delay before effects could appear; missing-data treatment].
- Interpretation: [Analysis method; possible alternative explanations; spillover between groups; concurrent changes; decision rule].
- Release and recovery: [Asset/version; applicable approvals; verification; owner; procedure to restore an approved state or correct the change].
Microsoft’s guidance on controlled experiments emphasizes measurable hypotheses, suitable randomization units, adequate sensitivity, and checks for pre-existing differences. These principles inform the brief; the GEO study’s actual design determines what conclusions it supports.
Use random assignments where feasible and suitable. If the study compares selected pages or periods without randomization, document how comparability was assessed and the remaining limitations. Label a simple before-and-after result as observational.
Keep prompt wording, sampling rules, and evaluation criteria stable unless a documented change is necessary. Record model or product changes, campaigns, site updates, and other events that could affect results. Repeated answers to one prompt are not automatically independent experimental units.
Do not change the success measure after seeing favorable results. Pause for breached conditions; record any decision to extend or redesign the study.
Results and decision
[Actual exposure and sample; missing observations; primary result and uncertainty; other outcomes; costs/hours; deviations; alternative explanations; evidence strength; decision; owner; next review].
Choose Adopt / Revise and retest / Stop / Inconclusive. Explain why the evidence supports the action. An inconclusive result may still justify an inexpensive improvement on usability or factual-quality grounds; record that basis separately from any claimed GEO effect.
H.3 Change log
Maintain one chronological log across corrections, experiments, and routine maintenance.
[Change ID; related request/experiment; asset and destination; previous version; new version; exact change or difference reference; reason; expected effect; owner; approval basis; release date/time; verification result; recovery action; follow-up date].
Classify the change as Correction, Experiment, or Maintenance. Record actual implementation dates separately from approval and verification dates. Include changes to prompts, monitoring tools, metric definitions, qualification rules, and attribution methods when they affect comparisons.
Preserve the previous approved state and the evidence behind it. Recovery must restore an appropriate approved state; it should not reintroduce a known factual error or withdrawn permission.
Reopen only approvals affected by the change, using Part G. After release, verify the relevant destinations and link the result to Part C monitoring. If several changes occurred together, record the bundle so later reports do not assign the whole result to one component.
H.4 Inquiry qualification worksheet
Capture the inquiry and discovery evidence
[Inquiry ID; received date; organization/contact; practice/service; stated need; contact route; responsible professional; response due date; duplicate/related records].
Ask two separate questions:
- “How did you first hear about us?”
- “What helped you decide to contact us?”
Where relevant, ask which resource, referrer, search service, or AI answer helped. Record the prospect’s account separately from tracked visit or referral evidence. Allow multiple contributions and Unknown. Avoid prompting the prospect toward an AI explanation.
Collect only the information needed for routing and qualification. Use the appropriate confidential intake process before requesting sensitive matter details.
Qualification questions
Area | Questions and evidence to record |
Need and fit | What problem or decision needs support? Does it match an available service and the firm’s competence? |
Scope and context | What organization, sector, geography, jurisdiction, and deliverable are involved? What is still unclear? |
Timing and capacity | What deadline or trigger matters? Can the firm respond and deliver within the required period? |
Decision and funding | Who sponsors and approves the work? What procurement, budget, or funding process applies? Record unknowns. |
Professional acceptance | Which conflict, independence, competence, client-acceptance, or engagement checks apply, and who owns them? |
Next step | What information or conversation would establish a suitable next action? Who will arrange it, and by when? |
Qualification decision
[Qualified / More information needed / Nurture / Refer / Decline; reasons; unresolved checks; next action; owner/date; opportunity ID if created].
Define the firm’s qualification criteria before reporting the rate. Keep commercial qualification and professional acceptance as separate statuses. A qualified inquiry may still have acceptance checks outstanding; qualification does not establish an engagement.
For consulting, clarify the business decision and implementation role. For accounting, identify the service and any assurance or independence implications. For law, establish the matter type, jurisdiction, and appropriate conflict-check route. For engineering consulting, clarify the discipline, project stage, technical scope, and required authority.
Retain a reason for referrals, declines, and lost opportunities. Review recurring mismatches for changes needed in service descriptions, targeting, or capacity. Record follow-up preferences before adding prospects to ongoing marketing.
H.5 Management scorecard
Scorecard header
[Reporting period; inquiry/opportunity cohort; data cutoff; practice/service; metric-definition version; owner; comparison period; targets; material changes and limitations].
For each measure, complete [Current value; count and denominator; comparison; target; status; evidence source; interpretation; action; owner/date]. Use the Part C metric dictionary and retain the exact product, report, surface, filters, time zone, and aggregation for platform measures.
Measure | Definition and management use |
Representation quality | Open material factual errors by severity and age, plus accuracy results from the defined Part C sample. Use to prioritize corrections. |
Correction progress | Requests submitted, verified source corrections, and answer-recheck outcomes, reported separately. Show overdue and unresolved items. |
Experiment decisions | Studies completed by Adopt, Revise, Stop, or Inconclusive, with evidence strength, cost, and resulting action. |
Visibility | The selected platform measure or sampled-answer measure with its exact scope. Keep measures with different counting rules separate. |
Inquiry quality | Qualified inquiries divided by completed qualification assessments for the stated inquiry cohort and cutoff. Show pending assessments separately. |
Opportunity outcomes | Won divided by won plus lost opportunities for the stated cohort or closing period. Show open and no-decision cases separately and define each status. |
Commercial value | Report open opportunity value, signed fees, and recognized revenue separately, with currency and timing. Label any probability-weighted estimate. |
Resources and capacity | External spend, staff hours, professional review effort, response backlog, and delivery capacity relevant to the planned work. |
For example, Google’s Generative AI performance report currently describes impressions for AI Overviews and AI Mode, with aggregation rules that affect totals. Use the documented measure without relabeling it as clicks, inquiries, or revenue.
Interpretation and commercial decisions
Deduplicate inquiries and opportunities before counting them. Show counts beside percentages, Not applicable for zero denominators, and Unknown for missing evidence. Compare cohorts with similar time to progress; avoid strong trend claims from a few inquiries.
Keep tracked referral evidence, prospect-reported contribution, and inferred influence distinct. Do not allocate the same engagement’s full value to several channels and then add those values together. If management wants a single allocation, document the rule and its limitations.
Define included costs: suppliers, software, content production, professional review, technical work, and monitoring. An observed cost per qualified inquiry requires a stated cost scope, cohort, and contribution rule. Establish a credible incremental benefit and cost basis before presenting an ROI claim; ordinary attribution records alone do not establish causation.
Management decision record
[Continue / Expand / Modify / Pause / Retire; evidence and uncertainty; affected activity; budget and hours; capacity implications; accountable owner; completion date; next review].
Use the scorecard to decide which errors to resolve, which changes to retain, and which services or channels deserve further effort. Record the action and resource decision alongside the measure so the next review can assess what management actually changed.
