AI Visibility Audit Checklist: Evidence Log and Action Plan
An AI visibility audit is a bounded review of how a business appears in selected AI answers and whether the available information supports those answers. A useful audit produces retained evidence, clear limitations and a prioritized action list. It does not produce a universal visibility score or a guarantee that a business will be recommended.
This checklist is for business owners and marketing teams who need a starting assessment. It separates three jobs: checking information people can find, observing selected answers and deciding what to improve. Ongoing tracking belongs in a separate measurement process.
1. Define what the audit will cover
Start with the business decision. Are you checking whether customers can understand a new service, whether an assistant describes the business accurately, or whether a category comparison overlooks important facts? Choose a scope you can document and review.
- Business: use the correct name, public website and service being reviewed.
- Audience: identify the buyer and the decision they are making.
- Markets and languages: state which location and language each question represents. Singapore English, Malaysia English, Simplified Chinese and Traditional Chinese are not interchangeable evidence.
- Products and modes: record the AI service and whether a visible search or research mode is used.
- Dates: define the observation period. A later answer may differ.
- Limits: state what the audit does not test, including unavailable accounts, markets or features.
Use only public or otherwise approved information in a simple initial audit. Do not upload customer contracts, identity documents or private business records merely to obtain a visibility opinion.
2. Build a customer-question panel
Write questions a potential customer could ask before making contact. Use conversations and genuine service questions as a starting point. Do not fill the panel with dozens of near-identical prompts just to increase the count.
| Decision | Illustrative question pattern | What to check |
|---|---|---|
| Understand a service | What does this service cover, and who is it for? | Clear scope, audience and limitations. |
| Compare approaches | Should our team choose training or tailored advice? | A useful distinction between the options. |
| Research providers | Which providers should we investigate for this need? | Names, context, reasons and cited sources where shown. |
| Check a named business | What should I confirm before booking with this provider? | Accurate facts and questions that can be verified. |
There is no universal minimum number of prompts that makes a sample representative of every buyer. Cover your priority decisions, report the panel size and explain why those questions were selected.
3. Check the website information that should answer those questions
Review the public pages a customer should use. Can someone identify the service, prerequisites, contact details and next step? Does a page contain visible text that answers the question, or only a vague claim and a form?
Check whether important pages can be discovered and read through the relevant search systems. For Google AI search features, ordinary Search eligibility and SEO foundations apply. Google says no special AI file or schema is required. Where access is uncertain, ask the person responsible for the website to check the actual crawl and indexing evidence.
Record the result you observed, not an assumption based on a filename. A robots.txt file being present does not by itself prove that a page is indexed. Structured data should match the visible content; it does not guarantee a citation.
- Check the intended page URL and whether the public page loads.
- Confirm the main answer is visible and understandable.
- Check factual consistency across pages you control.
- Look for genuine evidence behind important claims.
- Check internal links and the actual conversion destination.
- Record technical questions for an appropriate administrator.
4. Run the observations and preserve the full answers
For each question, record enough context for another reviewer to understand the observation. Keep the exact wording, date, product, language, stated location and relevant mode. A fresh conversation helps reduce unrecorded context, but it does not make the result universal or perfectly reproducible.
Retain the full answer and any visible source links. Mark separately whether the business was named, whether a page was cited and whether the description was accurate. An assistant mentioning a brand is different from citing a particular URL.
5. Copy this audit evidence log
Copy the table into a spreadsheet. Add one record for each observation. Keep an evidence identifier that points to the retained answer capture; do not replace the evidence with a score alone.
| Field | Your record |
|---|---|
| Observation ID and date | ________________ |
| Audience, decision and service | ________________ |
| Exact question, language and stated location | ________________ |
| Platform, visible mode and conversation context | ________________ |
| Full answer capture location | ________________ |
| Business mentioned? Which exact wording? | ________________ |
| Source URLs shown | ________________ |
| Claim checked and supporting public source | ________________ |
| Finding and evidence limitation | ________________ |
| Action, owner and follow-up date | ________________ |
6. Verify claims before deciding what to change
Open the sources shown, locate the specific claim and compare its wording, context and scope. Record whether the source supports, partly supports or does not support the claim. A link to the right domain is not enough if the cited page does not contain it. Distinguish a wrong business fact from a valid difference in opinion or an incomplete sample.
Also check your own assumptions. A business absent from one answer is not proven absent from the market. A competitor named in an answer is not automatically a direct competitor for every customer. Avoid reporting an anecdote as a population statistic.
7. Turn findings into a bounded action plan
| Finding | Suitable action | Reason for priority |
|---|---|---|
| Your own course page states an outdated prerequisite. | Correct the existing page and related information you control. | Customers may make an unsuitable booking using a factual error. |
| A key customer question has no useful owner page. | Improve an existing relevant page or commission a distinct answer where needed. | The information gap affects the buyer’s decision. |
| A claim appears in an answer without adequate support. | Document the discrepancy and investigate the apparent sources. | The issue may concern a third-party response rather than a missing keyword. |
| Your business is absent from one sampled answer. | Retain the observation and compare a consistent panel before deciding on work. | A single answer gives limited evidence for a large change. |
Connect each finding to its evidence, owner, next action and follow-up date. Correct factual information you control before considering wider content work. A missing keyword alone is not a reason to commission another article.
In a hypothetical audit, an assistant says a provider offers a local classroom course, but the provider’s registration page only describes a different format. The finding is the mismatch, with both sources retained. The next action is to correct any stale information the provider controls and check the apparent source. The report should not promise that the assistant’s answer will change on a particular date.
8. Keep audit, strategy and measurement separate
An audit gives a starting assessment and action list. A strategy chooses what the business will improve. Recurring measurement checks selected signals over time. Use our practical recommendation strategy guide for content work and our AI visibility measurement guide for repeat observations, referrals and enquiries.
Report changes carefully. A later improvement in sampled answers does not isolate the cause. Platform updates, different questions and wider marketing activity may affect what you observe.
Get support appropriate to the task
The 1-hour Word of AI workshop introduces business AI visibility. Afterward, a 30-minute discovery conversation can help explore fit. A 1-hour advisory session provides a focused discussion of priorities.
This DIY checklist is not a statement that a formal audit or implementation project is included in an Academy session. Any further deliverables and work must be agreed separately.
Common questions
Does an audit require a paid tracking platform?
No. You can begin with an approved manual process and a spreadsheet. Paid tools may help scale observations, but the scope, evidence and review still matter.
Should we combine all languages into one visibility percentage?
Keep the language, location and sample design visible. Combining unlike observations can hide important differences and overstate what the result means.
Should every missing keyword become another article?
No. Improve the page that should answer the question first. Commission a separate article when a materially different task needs its own complete answer.