Before buying, agree the market, platforms, question set, repeat checks and pages to be reviewed. Also ask whether the work ends with research or includes improving the website. Those are different deliverables.
Start with a buying decision worth investigating
Choose questions a plausible customer asks before selecting your product or service. Separate recommendations, comparisons and practical selection questions from branded prompts that already contain your company name.
For a local service business, the city and service availability matter. For a software company, integrations or a specific use case may matter more. The question set should reflect the offer you can actually fulfill.
Ask where the questions came from: your customer conversations, search data, observed related questions or the researcher's proposed test set. A provider's chosen prompts are not automatically a sample of real private conversations in ChatGPT. Broad AI keyword estimates are also not a count of your prospective buyers.
Check exactly what the audit covers
A Google AI Overview, Google AI Mode response, ChatGPT Search answer and Gemini app conversation are separate observations. Ask which products and modes will actually be tested. An API response is not automatically the same experience a customer sees in the product.
The scope should identify the target country and language, observation dates, question count and repetition plan. Record the available session conditions and any limitations. If no AI Overview appears, that is different from your brand being absent from an answer that did appear.
Confirm how unavailable checks are handled and whether the delivery contains raw observation records, a summary and page-specific recommendations. A short repeat sample can reveal variation; it should not be described as long-term market share.
Read one finding from evidence to proposed work
This is an annotated report example. It is not an actual model answer, a measured citation or a result achieved for a client. A real report would include the exact observation time, answer record and source URLs.
Expect someone to open and evaluate the cited pages
A list of competitor domains is not enough. The reviewer should inspect a relevant selection of cited pages, explain what information they offer and compare that with your existing pages. Useful findings can involve missing buying answers, unclear business facts, weak evidence or difficult access.
The proposed work should name the page and the question it will answer better. Sometimes that means improving an existing service page; sometimes it means a distinct supporting guide. Do not commission a new article for every variation of a prompt.
Access should be considered separately from answer quality. OpenAI's crawler documentation distinguishes search crawling from model-training controls. Allowing a crawler does not prove that your brand will be selected or recommended.
Do not buy a guaranteed mention or an unexplained score
A provider may use a visibility score to summarize its own sample. Ask for the denominator, weighting and treatment of unavailable checks. The score describes that method; it is not a universal AI ranking or your share of all customer conversations.
Be cautious about claims that a special schema type or AI file is required. Google's AI features guidance states that ordinary search eligibility and SEO fundamentals remain relevant; it does not require special AI markup.
You can still test useful ways to present original information, clearer comparisons or machine-readable material. The audit should explain the hypothesis and what can actually be observed, rather than turn an experiment into a guaranteed citation technique.
Connect visibility with the next business decision
An audit can identify where the brand was mentioned, linked or recommended in the selected checks. Analytics may show identifiable AI referral visits, subject to the tracking and referral data available. Sales or inquiry records are needed to understand whether those visitors became suitable prospects.
Search Console's available reports have their own scope. Review the actual reports available to your property; do not treat every Web Search click as an AI click. Preserve the date, platform and metric definition when comparing results.
If your main need is to establish ongoing measurement, the AI visibility measurement guide covers that workflow. This buying guide is about deciding whether the research and recommendations you receive are useful.
Ask how research becomes a page improvement
The handoff should contain the observation sheet, a plain-language summary, the source review and a short list of recommended changes. Each change should connect a customer question with a relevant page and a way to verify the completed work.
Agree whether writing, design, publication and follow-up checks are included. The research-only option should not be presented as if the pages have already been improved. Ongoing monitoring, digital PR and additional markets are separate decisions.
My AI search research and page-improvement scopes set out the questions, platforms, dates and review limits. One option covers research and a plan; another adds defined page work. You can also read the dated local-business AI source study for an example of how I keep the question set and observation limits visible.
Check the audit evidence
- Record the tested prompts, platforms, dates, language, and location settings so the same sample can be checked again.
- Separate brand mentions from linked citations; open cited URLs and compare the answer with the current source text.
- Assign each recommended change to an editor or developer, with an affected URL and a repeatable follow-up test.
Questions about AI visibility audits
What exactly will our team receive from an AI visibility audit?
A useful handoff includes tested buyer questions, platform and test settings, captured answers, brand mentions, and linked citations. Each finding should explain what the cited page actually supports and distinguish your own pages from third-party sources. The action list should identify affected URLs, priority, ownership, and how each change will be checked.
Is an audit useful if we have almost no AI referral traffic?
Yes. Referral analytics show some visits, but not unclicked brand mentions or every AI-assisted buying journey. A structured prompt sample can reveal whether your brand appears, which sources are cited, and whether business details are accurate. Interpret the results within the tested prompts, platforms, and dates rather than as a complete market-share score.
Will the audit tell us whether we need content changes or developer work?
It should separate the two. An editor may need to clarify service facts, substantiate claims, or update outdated source material. A developer may need to fix rendering, crawl access, or indexing issues. Training access and answer-retrieval access serve different purposes, so crawler recommendations should name the affected agent and its function.