GEO Audit TemplateTurn AI-answer evidence into client-ready work.

GEO Audit Template for Local SEO Agencies

GEO Audit Template for Local SEO Agencies

Last updated: July 15, 2026

A GEO audit template is a controlled framework for reviewing how a business is mentioned, described, recommended, and cited in generated answers, then planning evidence-backed corrections, source improvements, and retests. This template gives a local SEO agency the intake sheets, classifications, evidence log, decision rules, QA checks, and client handoff needed to run that review consistently.

What this audit does—and does not do

Use this audit to answer four practical questions:

1. Does the client appear for the local buyer questions in the approved sample?
2. Is the client represented accurately when it appears?
3. Which competitors and sources recur in the same answers?
4. Which corrections or authority-building actions are supported by evidence?

The audit is a dated sample, not a census of every possible AI response. Results can change with prompt wording, provider, model, mode, location, personalization, account state, source availability, and time. The agency can control the test method and the work it delivers; it cannot guarantee a mention, citation, recommendation, rank, response order, lead, or revenue result.

Copy and complete the audit workbook

Create one working document or spreadsheet with the sheets below. Keep raw evidence separate from the client summary so reviewers can trace every conclusion.

If software will support collection or reporting, evaluate it against the workflow in AI visibility tools for local SEO agencies. Name only the systems actually tested.

Step 1: complete intake and lock the audit boundary

Do not run prompts until the agency and client agree on the decision the audit must support. A prospect snapshot, one-time diagnostic, and recurring monitoring baseline need different depth.

Intake sheet

Scope decision table

Gate: If the entity facts, tested systems, or evidence permissions are unresolved, pause collection and label the audit blocked. Do not fill gaps with assumptions.

Step 2: build the entity truth set

The truth set is the standard used to judge factual accuracy. It must contain client-approved facts with a source, verification date, and owner. A directory listing is not automatically authoritative merely because it ranks or appears in an answer.

Entity truth-set worksheet

For a service-area business, do not publish or infer a hidden address. Do not add license, insurance, certification, award, review, longevity, pricing, or availability claims unless the client can verify them.

Source-of-truth order

Use the most direct approved evidence available:

1. client records and authorized statements;
2. current official pages and controlled profiles;
3. issuing bodies for licenses, certifications, or memberships;
4. current independent sources with clear provenance; and
5. directories or aggregators as conflict signals, not automatic truth.

Entity conflict log

Resolve high-risk contradictions—such as a wrong phone, address, availability claim, or regulated credential—before treating visibility growth as the primary goal.

Step 3: design and freeze the prompt panel

A prompt panel is the repeatable set of buyer questions used for the audit. Start with 20–30 prompts for a focused single-market baseline, then add prompts deliberately for extra services or locations. More prompts are not automatically better if the team cannot review them consistently.

Prompt families

Prompt-panel worksheet

Control sheet for comparable runs

Freeze the approved core panel for trend reporting. New exploratory prompts can be added, but report them separately until they have a comparable baseline. A material change to prompts, markets, providers, modes, or classification rules creates a series break and must be disclosed.

Step 4: run the panel and preserve evidence

Use one row per prompt-provider-run combination. Do not overwrite a prior answer when retesting.

Run log

Evidence log

Preserve the full answer when terms and permissions allow. A cropped screenshot can omit qualifying language, source context, or competitors. If full capture is not permitted, save the most complete authorized reference and document the limitation.

Step 5: classify answers and sources consistently

A single label cannot describe every dimension of a generated answer. Assign one discovery outcome, then record recommendation, citation, and accuracy separately.

Answer classification rules

Key distinctions

• Branded-only: The client appears because the prompt names it. Do not count this as unprompted discovery.
• Unprompted mention: The client appears without being named in the prompt.
• Recommended: The answer clearly presents the client as a suitable choice. A directory-style list without endorsement may be a mention only.
• Source-backed recommendation: Record this only when the answer recommends the client and exposes a source that supports the relevant statement. It does not prove the source caused the recommendation.
• Competitor displacement: A named competitor is recommended where the client is absent or discussed less favorably. Keep the underlying answer evidence.
• Inaccurate representation: Any material client fact conflicts with the truth set. Log the specific fact ID and severity.
• Cited: A source is visibly attached to the answer. A citation is not the same as a brand mention, and a mention is not the same as a recommendation.

Do not force a first/second/third “rank” unless the interface returns an explicit ordered list. Narrative answers, maps, carousels, and unordered recommendations need their own format labels.

Source classification register

Use these source classes:

When an answer exposes no source, write “no exposed citation” rather than inventing a source trail. A relevant uncited page discovered during follow-up belongs in the gap analysis as a hypothesis, not as a proven influence.

Step 6: calculate transparent summary metrics

Report counts and rates by provider, family, market, and service where the sample permits. Always show the valid-run denominator.

Optional outcome index

If stakeholders require a summary index, define it openly and keep the underlying metrics beside it.

Audit index = earned points ÷ maximum possible points × 100. Exclude branded-only, failed, blocked, and non-applicable runs according to the documented rule.

Scoring cautions

• The index is a sample summary, not an AI ranking or probability of future recommendation.
• Do not blend providers, locations, or prompt families without also showing their separate results.
• Do not treat small changes from a small denominator as meaningful without context.
• Do not backfill a missing baseline with estimates.
• Do not change weights after seeing results.
• Do not assign citation points when the provider exposes no citation.
• Do not compare periods across a series break without a visible warning.
• A higher score with inaccurate facts can be worse for the client than lower visibility with correct facts.

Step 7: compare competitors and identify source gaps

Use the same applicable prompts, collection window, providers, and classification rules for every named competitor. The audit can document recurring patterns; it cannot prove why a model selected one business.

Competitor comparison

Source-gap analysis

Use precise language:

• Observed: “Competitor A was recommended in 8 of 20 applicable runs and Source X was cited in 5 of those answers.”
• Supported gap: “The client is absent from Source X while two recurring competitors have complete profiles there.”
• Hypothesis: “Improving the client’s verified profile may strengthen corroboration; retesting is required.”
• Unsupported claim to avoid: “Source X makes the model rank Competitor A.”

Step 8: prioritize work with evidence, impact, confidence, and effort

Correct critical factual risks before chasing broader visibility. For every finding, create a task only when the evidence supports a practical next step.

Priority method

1. Assign Impact from 1–5 based on the number and commercial importance of affected prompts or facts.
2. Assign Effort from 1–5 based on total delivery time and dependencies.
3. Calculate Base priority = Impact × (6 − Effort), producing a range of 1–25.
4. Assign Confidence as High, Medium, or Low based on evidence quality.
5. Apply overrides for critical misinformation, regulated claims, privacy, security, or accessibility risks.

Opportunity queue

Action decision table

Step 9: prepare the client handoff

The handoff should be shorter than the working audit and traceable to it. Include:

1. Scope: prompts, families, markets, services, competitors, systems, modes, dates, and valid-run count.
2. Entity risks: any inaccurate or unverifiable business facts and their severity.
3. Visibility findings: separate unprompted presence, recommendation, accuracy, citation, and competitor displacement.
4. Source findings: exposed citations, relevant profile or editorial gaps, and clearly labeled hypotheses.
5. Top actions: three to seven tasks with evidence, owner, timing, and retest prompts.
6. Limitations: variability, unavailable sources, access gaps, sample size, and series breaks.
7. Next decision: correction project, audit extension, monitoring, implementation retainer, or scheduled recheck.

Client-ready summary template

What we tested: [N] valid runs from [N] approved buyer prompts across [SYSTEMS / MODES] from [DATES], covering [FAMILIES] in [MARKETS].

What we observed: [CLIENT] appeared without being named in [X]% (N/N) of applicable runs and was clearly recommended in [Y]% (N/N). [COMPETITOR] most often displaced the client in [FAMILIES]. We found [N] inaccurate facts and [N] source gaps supported by the evidence log.

What should happen next: Prioritize [ACTION 1], [ACTION 2], and [ACTION 3] because they address [FACT RISK / HIGH-VALUE PROMPTS / RECURRING SOURCE GAP]. Each action has an owner and defined retest prompts.

How progress will be reviewed: Complete the approved work, rerun the unchanged core panel on [CADENCE], and compare the same metrics while disclosing provider or methodology changes.

Limitations: This audit is a controlled sample of variable generated answers. It does not guarantee a mention, citation, recommendation, rank, lead, or revenue outcome.

Use how to sell GEO services to present the commercial decision without overstating the evidence. If recurring work is justified, compare GEO retainer packages, then adapt the GEO proposal template. If the audit finds only a finite correction, quote a project rather than forcing a retainer.

Handoff pack checklist

• [ ] Executive summary with exact denominators.
• [ ] Scope and methodology statement.
• [ ] Entity truth set and unresolved conflict list.
• [ ] Prompt-panel appendix.
• [ ] Provider- and family-level result tables.
• [ ] Competitor and source-gap analysis.
• [ ] Prioritized opportunity queue.
• [ ] Evidence index with permissions or redactions.
• [ ] Limitations and series-break notes.
• [ ] Named client and agency owners.
• [ ] Agreed next decision and approval date.

Step 10: run final audit QA

A second reviewer should complete this checklist before the report reaches the client.

Method QA

• [ ] Intake, purpose, market, service, provider, mode, dates, and settings are complete.
• [ ] Entity facts have approved values, sources, dates, and owners.
• [ ] Core prompts are frozen and prompt IDs are unique.
• [ ] Every attempted run has a status; invalid runs are excluded transparently.
• [ ] Branded-only appearances are not reported as unprompted discovery.
• [ ] Mentions, recommendations, citations, accuracy, and ordering are separate fields.
• [ ] Narrative answers are not converted into invented ranks.
• [ ] Provider-level and prompt-family results remain visible.

Evidence QA

• [ ] Every number can be recalculated from the run log.
• [ ] Every client-facing finding references evidence IDs.
• [ ] Exposed sources are distinguished from inferred opportunities.
• [ ] Screenshots and exports preserve enough context for review.
• [ ] Confidential information is authorized, minimized, or redacted.
• [ ] Competitors were tested under comparable conditions.
• [ ] Unsupported causality language has been removed.

Decision QA

• [ ] Critical inaccurate facts are prioritized before visibility expansion.
• [ ] Every production task has evidence, an owner, a due date, and retest prompts.
• [ ] Low-confidence ideas are labeled investigation, not promised fixes.
• [ ] The scoring formula, denominator, exclusions, and limitations are visible.
• [ ] No guaranteed placement, traffic, lead, or revenue claim appears.
• [ ] The client handoff recommends the smallest appropriate next scope.

Frequently Asked Questions

What is included in a GEO audit?

A GEO audit includes a defined scope, verified entity facts, a stable buyer-prompt panel, preserved answer evidence, separate presence and recommendation classifications, accuracy checks, source and competitor analysis, a prioritized action queue, QA, limitations, and a retest plan.

How many prompts should a local GEO audit use?

A focused single-market baseline can begin with 20–30 prompts across recommendation, service, location, cost, trust, comparison, urgency, branded, and follow-up families; add prompts deliberately when extra services or markets create a distinct decision need.

How should an auditor classify a brand mention versus a recommendation?

Classify a mention when the business is merely named or discussed, and classify a recommendation only when the answer clearly presents it as a suitable choice; also record whether the appearance was branded-only or unprompted.

What counts as a source in a GEO audit?

Count a source as exposed evidence only when the tested answer shows or links it; other relevant pages found during follow-up can be logged as source-gap hypotheses, but they should not be described as causes of the answer.

Can a GEO audit score be called an AI ranking?

No. A GEO audit score summarizes outcomes from a defined sample and should always show its formula, denominator, provider-level results, and limitations; it is not a universal rank or a probability of future visibility.

How should competitors be compared in a GEO audit?

Compare competitors with the same applicable prompts, providers, modes, dates, and classification rules, then report observed presence, recommendations, prompt-family strengths, and recurring sources without claiming to know why a model selected them.

How often should the prompt panel be retested?

Retest on a cadence that matches implementation and reporting—often monthly for active work—while preserving the approved core prompts and documenting any change to providers, modes, locations, or classification rules as a possible series break.

What should the client receive after the audit?

The client should receive a concise scope and methodology summary, key entity and visibility findings, competitor and source gaps, prioritized actions with owners and retests, an evidence index, limitations, and a clear decision between a project, monitoring, implementation, or a later recheck.

Turn the audit into the right next scope

Use the opportunity queue—not a generic package—to estimate the actual prompts, locations, analyst time, implementation, reporting, approvals, and margin. Then use the GEO retainer calculator and copy the GEO Proposal Template only if the evidence supports a paid next step.