Begin with customer questions.
Record the prompt, date, answer, and cited URLs when shown. Treat the result as a snapshot, not a stable ranking.
Choose questions that matter to your services, not vanity prompts such as “Who is the best agency?” A useful set might ask how to plan a website redesign, when a business needs local SEO, or what a website care plan should include. Include early research and decision-stage questions. The audit should reveal where your explanations are useful or missing.
Write down the exact wording. A small wording change can produce a different answer. Run the same questions again on a schedule only if someone owns the comparison and will act on what it shows.
Inspect the cited pages and your own site.
If a competing page is cited, examine what it actually answers, which facts it supports, and how it structures the explanation. Do not copy its wording or assume its appearance proves every claim is correct. Then inspect your closest page. Does it answer the question early? Is the information current and readable? Can a visitor find the relevant service or deeper guide?
Google’s AI features documentation says conventional SEO best practices remain applicable and inclusion is not guaranteed. That makes page usefulness, accessibility, and discoverability sensible priorities, not a proprietary “AI ranking formula.”
Separate AI answer visibility from business value.
An AI answer may mention your brand without sending a visitor. A citation may appear for a broad educational query that never leads to a qualified inquiry. Record those outcomes separately. Look at referral traffic when available, but do not assume all AI-influenced visits carry a clear referrer.
The audit should answer a business question: which customer decisions does your site explain well, and where does a person still need a better source? Citation count alone is not enough.
Correct facts at the source.
If AI answers repeat an outdated service description, confirm the current offer and update your own website and governed public profiles. If your page lacks a direct answer, add one with supporting context. If a claim needs proof, secure the evidence or remove it. Do not create thin pages for every prompt variant.
A search engine optimization plan can connect these findings with crawl access, content ownership, and measurement. That is more durable than chasing a single model response.
Make the next audit comparable.
Save the URL you tested, not only a screenshot of the answer. Note whether the page was live and materially updated before each test. If a tool cites a page that no longer says what the answer implies, record the mismatch. A later retest should compare the same question with the same market and account conditions where possible. Even then, results can vary. Use the audit to choose a clearer explanation or correction, not to declare that one model has established your market position.
Start with ten questions from real inquiries.
Choose ten questions from actual inquiries. Save baseline answers from the tools you can access, note citations and factual errors, and identify three pages to improve. Retest after those changes are live and indexed. Do not promise a citation or position based on a draft.
