A website can rank for relevant searches and still be absent from an AI-generated answer. One missing mention does not identify a technical fault, and one successful prompt does not establish reliable visibility. A useful GEO audit starts by defining which questions matter and collecting evidence across access, content, sources, and business outcomes.
GEO usually means generative engine optimization; AEO refers to answer engine optimization. You may also encounter AIO used for AI optimization or AI Overviews. The labels are less useful than a clear objective: make your website a relevant, accessible, and credible source for the questions your customers ask.
Choose questions that represent real decisions
Build a short question set from support conversations, sales enquiries, site search, and existing search queries. Include several stages of a decision:
- Understanding: “What causes condensation inside a garden office?”
- Comparing: “How do ventilation and insulation solve different moisture problems?”
- Choosing: “Which ventilation options fit a small office without a window?”
- Checking: “What must I measure before installing this particular unit?”
These are illustrative questions. Replace them with language from your own customers. Avoid a test set consisting only of your brand name: it cannot reveal whether your useful expertise is discoverable before someone already knows you.
Keep the observation conditions consistent
Record the product, displayed model or mode, date, language, region, prompt, and whether web search was used. Save the answer and its source URLs. Use fresh conversations for the main comparison and keep follow-up questions in a separate track.
Repeat a manageable set over time. Label the results as observations under those conditions rather than universal rankings. An answer can change with wording or context, and a system may answer without consulting the live web.
Separate three outcomes: a brand is mentioned, a URL is cited, and the cited source actually supports the statement. These are different findings. A citation to the wrong page can expose an information problem even when a dashboard calls it a visibility win.
Verify technical eligibility before rewriting
Google explains that its generative search features build on core Search systems and that ordinary SEO practices remain relevant. Its AI optimization guide does not prescribe a special AI schema or an llms.txt ranking shortcut.
For each target page, check the final response, crawler access, indexing directives, canonical URL, internal links, and the availability of the main content. Use the crawler access audit to keep search access separate from training preferences. Resolve a demonstrable access problem before interpreting prompt results as a verdict on the writing.
Compare the source, not just the brand
When another site is cited, inspect the exact cited page. What question does it answer? Does it provide measurements, a clear definition, a compatibility explanation, or a documented procedure? Note the useful element without copying its wording or assuming you know why the system selected it.
For example, a competitor's installation page might include minimum clearance measurements while yours says only “easy to install.” The practical improvement is to obtain and publish verified measurements with their conditions. Repeating the competitor's brand-neutral introduction would not close that information gap.
Keep comparison notes factual. “This page includes a dated test method” is observable. “The model rewards this exact heading length” is an unverified theory.
Improve one answerable section at a time
Choose a page that already belongs to the question. Put a direct answer near the relevant heading, then explain the conditions and exceptions. Include a worked example when it helps readers apply the information. Attribute externally sourced facts and identify who is responsible for maintaining your own data.
A useful section on ventilation would explain the room size, occupancy assumptions, measurement units, and when professional advice is needed. It would not promise a universal recommendation from a single number. Precision makes a page useful even if an AI product never cites it.
Connect the section to appropriate supporting pages using ordinary descriptive links. Our internal linking guide covers how to make those relationships navigable. Do not create dozens of disconnected question pages when one maintained guide serves the reader better.
Prioritize the audit findings
- Access failure: an important page returns a challenge or cannot be retrieved. Assign a technical owner.
- Missing information: a customer question has no accurate answer on the site. Assign a subject expert.
- Conflicting information: product facts or company details disagree across pages. Assign a data owner.
- Presentation problem: the answer exists but is difficult to find or interpret. Assign an editor.
- Uncertain visibility: the page is sound but absent from a small prompt sample. Continue observation before making speculative changes.
End each audit cycle with a short change log and the same repeatable question set. Track visits and useful enquiries where measurable. GEO work is easier to evaluate when every change fixes an identifiable problem instead of chasing a promise of guaranteed mentions.