What should an AI visibility audit measure?
Search rank and answer visibility are related but different observations. A page may rank well in conventional results while an answer engine cites a different source. An answer may also name a company without linking to it. Measure these events separately. Start with a stable set of questions a buyer might ask before choosing a product: problem questions, category questions, comparisons, and requests for proof.
For every question, save the wording, engine, date, market, answer, cited URLs, and whether the answer names your brand. Repeat the run under similar conditions. Responses can vary, so one screenshot is a lead for investigation rather than a trend. Keep the raw answers so someone else can reproduce your classification.
| Observation | What it means | What to inspect next |
|---|---|---|
| Named and cited | A claim is connected to your page | Does the page support the claim? |
| Named without citation | Attribution is unclear | Are supporting facts public and easy to find? |
| Competitor cited | Another page served the question | What evidence or detail does it supply? |
| No relevant answer | The prompt may be poorly served | Recheck the buyer intent before creating content. |
Improve the source page
Inspect the page a buyer should reach. Does it say who the product serves, what it does, how it works, and what proof supports the claims? Can someone reach the relevant passage without a gated PDF? Are headings and links descriptive? Can a crawler access the page? These questions are useful even if AI visibility never changes.
Do not assume a schema block or arbitrary word count guarantees inclusion. Google's guidance says its AI features build on core Search systems and do not require special schema markup. Structured data can still describe eligible content, but it must match what readers can see. Prioritize original material: a worked example, clear methodology, actual product behavior, or approved customer evidence.
Turn the audit into an editorial backlog
Group missing answers by buyer question rather than by engine. Write a one-line hypothesis for each gap and attach the query capture and relevant page. Prioritize questions that affect decisions and can be answered with evidence you have. After publishing, rerun the fixed set and review qualified visits and conventional search performance too. A mention alone is not a business outcome; the useful question is whether the content helped a buyer make progress.
Design a query panel that survives the next review
The hardest part of an answer-visibility program is defining the questions before seeing the results. If a team keeps replacing difficult prompts with favorable ones, its trend line becomes meaningless. Write down the buyer stage, intended audience, exact query, and reason for including each question. A balanced panel should contain some questions where your brand is expected, some where a competitor is likely, and some where no vendor recommendation is appropriate. Keep a separate exploratory list for new questions discovered during sales calls.
Set a reporting cadence and a consistent capture process. If the answer interface personalizes by account or market, use a documented testing environment. Record whether the engine cited your site, another owned property, a partner, or a third party. A citation to a partner page may be useful, but it is not the same as a citation to your current product explanation. Reviewers should classify the claim, not merely count links.
When a change is meaningful
Compare several runs before calling an apparent improvement a trend. One answer might change because the engine updated, the source page changed, a query was interpreted differently, or a new competitor article appeared. Keep these possibilities in a change log. When the panel is small, show counts alongside percentages: “three of twelve questions” is more honest than a percentage standing alone.
The editorial team should close each cycle by selecting a few specific page actions. Examples include publishing a transparent implementation section, correcting stale partner descriptions, or making product limitations visible. Assign an owner and review date. The next cycle should test whether the answer became more accurate and whether a buyer reaching the page could complete the intended task. This creates a useful feedback loop even when the answer engine does not cite the page.
Questions to settle before reporting visibility
How large should the question set be?
There is no universal number. Start with a panel small enough that someone can inspect every answer and large enough to cover the important buyer stages. Twelve carefully chosen questions can reveal content gaps, but they cannot estimate a market-wide citation share. As the panel grows, keep a stable core for comparison and a separate exploratory group for new language from buyers. Publish the question list and sampling conditions internally so a change in methodology is visible.
What counts as an improvement?
Define the hierarchy before seeing results. A linked, accurate answer to a relevant buyer question is different from a vague name mention. A correction to a previously inaccurate answer can be valuable even if the overall mention count stays flat. Pair the panel with qualified page behavior: did visitors reach the cited section, continue to documentation, or ask a better-informed sales question? No single metric captures all of that.
Copy this observation record
For each run, record question ID, exact wording, buyer stage, engine and configuration, date, market, full answer, brand mention, cited URLs, claim accuracy, and reviewer note. Add the owned page that ought to answer the question and a proposed improvement. Keep the raw answer separate from your interpretation. On the next review, compare like with like, flag changes to the prompt set, and close each editorial action only after checking the published page itself.