A GEO audit needs a repeatable observation method and an actionable output. A screenshot showing your brand once is a useful observation, but it is not a baseline. This checklist helps a team scope the research, retain evidence and decide what to change.
This is a practical method for conducting a limited audit. To commission one, use our AI visibility audit service overview to compare scopes and deliverables.
1. Write down the business decision
Choose one product or service, a target market and the buying decisions that matter. For example, a SaaS team may want to understand discovery during product comparisons; a professional-services firm may care about location and service suitability. Define the question before choosing a monitoring tool.
2. Build a prompt sample you can explain
Collect questions from sales conversations, customer support, site search and keyword research. Group them by discovery, comparison, suitability and purchase preparation. Remove near-duplicates and label questions that explicitly mention your brand.
A practical pilot might use 20 questions in one market. That number is a manageable research choice, not a claim about statistical representativeness. Keep the list stable during a measurement period and record any additions separately.
3. Define the observation conditions
Record platform, mode, language, available location settings, date and whether a conversation was fresh or continued. Save the original answer and citation URLs. Include failed checks and answers that do not cite sources. If a setting cannot be controlled, record the limitation.
Repeat observations using the same protocol. More runs can help reveal variability, but repeated outputs are not automatically independent samples of what all customers see.
4. Use clear metrics
| Metric | Calculation or rule | Limitation |
|---|---|---|
| Mention rate | Valid observed answers mentioning the brand ÷ valid observed answers | Describe the sample and keep branded questions separate |
| Website citation rate | Valid answers citing the domain ÷ valid observed answers | A citation does not guarantee a recommendation or visit |
| Factual accuracy | Review product or service statements against approved facts | Requires human judgement and a record of corrections |
| Recommendation context | Record the use case, limitations and alternatives named | Do not reduce nuanced answers to an unexplained score |
If a brand appears in 8 of 20 valid observed answers, its mention rate for that sample is 40%. That is not 40% of all AI searches or 40% market share. State the denominator every time.
5. Inspect competitor and citation evidence
Apply the same observations to agreed competitors. Inspect the URLs cited alongside them: product pages, comparisons, directories, reviews or editorial coverage. Identify what those pages actually say. Do not infer that buying a mention on a cited site will reproduce the result.
Use the distinction between AI mentions and citations to keep recommendations and source links separate. Record irrelevant or incorrect citations too.
6. Review the pages you control
- Can readers understand the offer, intended audience and limitations?
- Are important claims supported by product facts or documented examples?
- Are relevant pages connected by crawlable internal links?
- Are valuable pages isolated from the rest of the site? Check for orphan pages.
- Are facts consistent across the website and relevant business profiles?
7. Turn findings into a work queue
| Finding | Action | Verification |
|---|---|---|
| Product capability described incorrectly | Review public descriptions and cited evidence; correct owned information | Approved page facts, followed by repeat observation |
| Buying question unanswered | Add the answer to the best existing page | Editorial review and working internal links |
| No identifiable referral path | Review acquisition reporting and landing pages | Documented measurement coverage and limits |
Assign an owner, effort estimate and review date to each action. Prioritise factual errors and commercially relevant gaps. Avoid treating every absent mention as a defect.
8. Report outcomes without overstating attribution
Keep implementation progress, observed visibility and business outcomes in separate report sections. Use AI traffic measurement and qualified enquiries where available. A visibility improvement does not prove that an editorial change caused it.
What the finished audit should contain
A scope statement, prompt inventory, dated evidence, metric definitions, competitor observations, a prioritised roadmap and a clear description of uncertainty. Agree what the audit budget includes before work starts.
How often should the audit be repeated?
Choose a cadence that fits your decision cycle and resources. Repeat after substantial product or content changes, while keeping comparable observations available.
Move from findings to implementation
Use our guide to building content hubs when the audit reveals disconnected answers across your website. Our GEO services explain the implementation work that can follow an audit. Visit Crescendo to discuss the scope and priorities for your business.