A screenshot showing a competitor in an AI answer can raise a useful question. It cannot tell you, by itself, whether your website has a technical problem, lacks relevant information, or simply did not appear in that particular response.
Before commissioning an audit labeled AI SEO, AEO, or GEO, agree what the provider will inspect and what you will receive. The following six deliverables turn the review into a practical decision about your website and ongoing marketing work.
1. A scope tied to real customer decisions
The audit should name the services, locations, pages, and search experiences being reviewed. “AI visibility” is too broad to define a useful test on its own.
Start with the questions a prospective customer needs answered: whether you offer the required service, work in their area, have relevant experience, and provide a suitable next step. Separate questions about your business by name from discovery questions that do not mention it.
For example, “Does this company repair commercial equipment?” tests a different issue from “How do I choose a commercial equipment repair provider?” The first concerns business facts. The second concerns useful discovery and selection information. An audit should not treat them as interchangeable.
Ask for the included platforms, page sample, research method, access required, and exclusions. If the review does not include server logs or account-level reports, it should say so. An untested area belongs under limitations, not under “passed.”
2. A page-level access and eligibility review
Request findings for important service and location pages, rather than one site-wide pass/fail score. The technical review should explain what was tested, when it was tested, and what the result means.
- Can the public URL load successfully without an unintended sign-in or challenge?
- Is essential service information available in the page content?
- Do crawl instructions, indexing directives, canonical signals, or platform controls conflict with the intended visibility?
- Are important pages discoverable through usable internal links?
- Do available crawl or server records support the finding, or is that evidence missing?
Platform requirements differ. Google's current generative AI search guidance connects eligibility to Search requirements, indexing and snippet eligibility, and the relevant Search Console inclusion setting. Meeting requirements still does not guarantee that a page will appear.
OpenAI's crawler documentation distinguishes OAI-SearchBot, which supports search discovery, from GPTBot, which concerns model training. An audit should identify the relevant control rather than assume every AI-related crawler serves the same purpose.
Do not approve a blanket removal of security protections as an “AI optimization.” Ask the website specialist to identify the specific issue, the affected legitimate requests, and a narrowly scoped remedy that preserves appropriate protection.
3. A dated answer and citation sample
A useful answer sample is reproducible enough for someone else to understand the observation. Ask for a log with:
- The exact question or prompt and why it represents a customer need.
- The platform and search mode used, with relevant location and session context where known.
- The date of the check and whether the same question was tested more than once.
- Whether your business was mentioned, linked, or cited as a source.
- The cited URLs, the claim associated with each, and a saved excerpt or screenshot where appropriate.
- Any missing context, unavailable answer, or uncertainty.
A mention without a link differs from a citation to your website. A citation to an outdated page differs from an accurate explanation of your service. These distinctions are more useful than a single unexplained “AI score.”
Treat the sample as a record of observed responses, not a census of what every customer sees. If the provider summarizes the sample as a percentage, require the numerator, denominator, question set, dates, and method. A changed question set should not be presented as a clean visibility gain.
4. A business-fact and content-gap review
Compare the information in sampled answers with your actual services, locations, hours, contact details, and supported claims. Then inspect the website and relevant public business information for conflicts.
The audit should name the incorrect or missing fact, identify the page you control that can clarify it, and distinguish that from third-party material you cannot directly edit. Changing your page does not guarantee that an outside answer will update immediately.
For content recommendations, ask for a specific customer question and the missing evidence needed to answer it. Useful inputs may include a real project example, an explanation from your specialist, a service limitation, or an original photograph. The provider should not invent experience or client outcomes to fill a gap.
Google's guidance does not require special AI text files or a unique schema format for generative AI search. Ask what a proposed technical addition accomplishes for the actual platform, and whether improving the existing page would be more useful.
5. A practical route from discovery to inquiry
Follow any sampled website links as a prospective customer would. Does the destination answer the question that brought the visitor there? Can someone tell whether your business is a fit? Is the next step clear and usable on a phone?
Record broken destinations, outdated offers, confusing contact paths, and forms that fail a delivery test. An accurate mention has limited business value if the visitor reaches the wrong page or cannot contact you.
The audit should also state what the reporting can establish. Separate platform visibility data, identifiable referral visits, inquiry events, and qualified opportunities. Unknown sources should remain unknown. A lead should not be attributed to an AI answer merely because the business appeared in a test response.
6. A prioritized implementation and retest plan
Every recommended task should have an affected URL or account, supporting evidence, expected benefit, owner, dependency, and completion check. Separate definite errors from ideas worth testing.
Illustrative audit finding: A service page describes installation but never explains whether maintenance is offered. The proposed task is to verify the actual offer with the business and revise that existing page. Completion means the approved explanation is live and readable; the later review checks relevant customer questions and inquiry patterns. It does not promise a citation.
This is a hypothetical example, not a Guardian case study. It shows the difference between a task the team can complete and an outcome it can only observe.
Prioritize problems that affect important customer journeys. Ask which fixes are included in the audit fee, which require a separate project, and which need ongoing review. A recurring engagement should include useful implementation and interpretation, rather than endlessly rerunning the same prompts.
Questions to ask before accepting the audit
- Can we inspect the underlying findings and sampled answers?
- Which conclusions are directly observed, inferred, or untested?
- Which changes improve the website even if AI citations do not increase?
- Who will implement and verify the priority work?
- How will the retest account for changes in platform, prompt, and context?
- What would justify continuing, changing, or stopping the work?
Be cautious if the proposed deliverable is only a score, if every missing mention becomes a demand for a new page, or if the provider guarantees inclusion. A useful audit helps you choose the next improvement and understand its limits.
Discuss an evidence-led website review
Guardian's SEO services connect technical access, service content, and measurement with the customers a business wants to reach. For scope and cost questions before commissioning work, see our SEO proposal checklist.
Talk with Guardian Solutions about your website's AI search visibility. Share your website, priority services, and the customer questions you want reviewed so the discussion starts with a defined problem and a practical next step.