You open an AI visibility report and find a lot of numbers. Mentions. Citations. Competitors. What should your team do on Monday?
A useful audit connects those observations to specific work: correct an inaccurate product page, answer an important customer question, or investigate a page that crawlers can't reach. Start with the decisions your audience needs to make, then inspect the answers and sources behind the totals.
This walkthrough gives you a manageable first audit and a clear way to prioritize the findings.
What are you trying to learn?
Choose a scope before running checks. For example: “How do AI assistants describe our product to small marketing teams, and do we appear for the three buying needs we care about most?”
Write down the company, website, audience, competitors, languages, and locations in scope. Choose the AI tools and available modes you'll check. Record anything you can't control or identify. An API sample should be described as an API sample, rather than a replica of every consumer conversation.
Start small enough that you can read the answers. A carefully chosen set of questions will teach you more than a huge export nobody reviews.
Build three question groups
Use customer conversations and search data to choose the needs. Keep these groups separate:
- Discovery: questions about solving a problem or choosing a category of tool.
- Comparison: questions about requirements, alternatives, or tradeoffs.
- Brand facts: questions about your company's current product, audience, pricing, and capabilities.
A hypothetical example: “How can a small agency report client results?” belongs in discovery. “Does ExampleCo support automated report delivery?” belongs in brand facts. A strong result on the second doesn't mean people will discover the company through the first.
Keep the exact wording and a note explaining why each question matters. The prompts are a research sample you chose; they don't reveal private conversations or establish market-wide demand.
Read the responses behind the metrics
Save completed answers with their dates, setup, and sources. Record failed runs separately. Look for whether your company was named, recommended for the stated need, linked as a source, and described accurately.
Don't collapse those into one success flag. An educational article can receive a citation while a different vendor gets the recommendation. A product can also be recommended with an incorrect price.
Read examples from each group, including ambiguous cases. Check brand-name variants and make sure another company with a similar name isn't being counted as yours. Open cited URLs and confirm that they support the relevant statement.
WildSEO can help you inspect tracked answers, competitors, and citations. Before using a summary number, read our AI share-of-voice guide so your team agrees on what it counts.
Check whether important pages are accessible
Pick the pages that should answer your priority questions. Confirm that they load publicly, contain the current answer, and have relevant internal links pointing to them. Check for accidental access restrictions, broken links, misleading canonicals, or indexing controls that conflict with your goals.
For Google's AI search features, Google says supporting pages must be indexed and eligible for a search snippet. A good page isn't guaranteed inclusion, but access and eligibility are sensible things to verify before commissioning more copy.
Be precise about bots. OpenAI documents separate controls for OAI-SearchBot and GPTBot: search access and model-training access serve different purposes. Don't treat a training-bot visit as proof that an answer retrieved or cited your page.
If crawler policies need changing, review the intended access with the site owner. Record the decision and verify it afterward. A blanket “allow every bot” instruction isn't a useful audit finding.
Compare the page with the question
Can a reader answer the question using the page? Check whether it explains the relevant capability, requirement, price, or limitation. Look for current evidence rather than simply counting words.
For a fictional integration page, “connect your tools effortlessly” leaves a lot unanswered. A useful version names the supported connection, explains which records move, shows a tested setup, and states what doesn't sync.
Record the gap in that level of detail. Prefer improving an existing relevant page to publishing several overlapping articles. Link to supporting documentation where it helps the reader complete the task.
Finish with owners and a review date
Prioritize inaccurate information and technical blockers first, then important unanswered customer questions. Leave speculative formatting experiments until you have a clear reason to test them.
For each finding, record the evidence, affected URL, proposed action, owner, and acceptance check. “Update pricing explanation and verify the live page” is actionable. “Improve AI authority” isn't.
After the changes, rerun the same question set over time and compare the underlying answers. Also check relevant visits and conversions. Keep the audit's visibility observations separate from claims about what caused a sale.
The audit is finished when the team knows what to fix, who owns it, and how to check it. Keep that list short enough to act on.
