Michele Eccher
On this page
To audit AI search visibility, choose the questions your customers actually need answered, test the relevant search experiences, and save the answers, brand mentions and cited URLs. Compare those sources with your own and competitors’ pages, check for technical barriers, then prioritise the work. The outcome is a documented diagnosis, not a fixed position or a score that applies across platforms.
A question that names your brand may return a convincing answer while hiding the real gap: does the brand appear when the question does not name it? An audit starts with that distinction, records answers under defined conditions and follows the sources a reader can actually open.
This guide focuses on the diagnostic process. The separate guide to measuring AI search visibility explains ongoing KPIs and reporting. A technical SEO audit goes deeper into site-wide crawling, rendering and indexation.
Scope
Define markets, audiences, decisions and platforms before writing prompts.
Observe
Record answers, mentions, citations, sources and competitor presence consistently.
Diagnose
Trace gaps back to access, content, evidence, entities or third-party sources.
Prioritise
Turn findings into testable actions with owners, confidence and review dates.
Begin with the commercial decisions the audit must support
Start with the decision you need to understand. For B2B software, it might be a procurement team choosing a supplier; for a clinic, someone exploring a treatment in one city. Specify the market, the people involved and the platforms they are likely to use. Test each language or market separately: an answer in English does not tell you what a customer in Italy will see.
Write the scope before collecting answers
- Priority products, services and topics
- Countries, languages and local market assumptions
- Customer roles and stages of the decision
- AI search experiences included in the review
- Named competitors and alternative ways to solve the problem
- The period, device, account state and collection method
- The business question the findings should help answer
Build prompts around decisions, not keyword permutations
Collect real questions from sales calls, support requests, search queries and interviews. Group them by task: understand the problem, find alternatives, compare them, verify a brand and act. “Which platform handles orders in multiple currencies?” tests a capability; “Which one suits a team with two-step approvals?” adds a constraint that may change the answer. Avoid turning every question into a keyword variation.
| Decision stage | Prompt purpose | What the audit observes |
|---|---|---|
| Understand the problem | Explain a need, risk or unfamiliar category | Whether the topic and its important distinctions are represented accurately |
| Create a shortlist | Find providers, products or approaches that meet criteria | Brand inclusion, competitor inclusion and the reasons given |
| Compare options | Evaluate trade-offs, capabilities, price or suitability | Framing, factual accuracy, evidence and cited sources |
| Validate a brand | Check experience, reputation, proof or limitations | Whether first- and third-party sources support a coherent description |
| Take action | Choose, contact, buy, visit or implement | Whether the answer creates a relevant and traceable next step |
Separate prompts that name the brand from those that do not. If the system describes a company correctly only after you name it, you have tested description accuracy, not inclusion in category searches. Keep both groups in the log and do not average them into one percentage.
Capture the answer, the sources and the conditions
For each test, save the exact wording, platform and search mode, country, language, date, device and account conditions. Retain the answer within the platform’s rules. Mark brand mentions, links to its site, description and competitors separately. For each cited URL, record the exact page and source type: company site, review or independent publication. Illustrative example: an answer to the two-step approval question names your brand without a link and cites a competitor’s documentation. The log says brand mentioned: yes; own site cited: no; competitor URL and source type recorded.
Presence
Is the brand absent, mentioned in context or presented as a relevant option?
Description
Are its category, capabilities, location and limitations represented accurately?
Citation
Does the answer link to the brand, an independent source or neither?
Source quality
Is the evidence primary, independent, current, specific and accessible?
Competitive context
Which alternatives appear, and what evidence appears to support their inclusion?
Next step
Can a user verify the claim or continue to a useful page?
Official reports add a different layer of evidence from manual tests. Search Console documents content appearing in generative Google Search and Discover experiences; Bing Webmaster Tools shows citations, cited pages, grounding queries and trends. Compare data within each platform rather than turning it into one ranking across AI products.
Open cited sources and compare the evidence behind competitors
An answer may name your brand but link only to a competitor’s documentation. Open that page and check what it establishes. In the approval example, the competitor’s documentation might explain the feature while your page says nothing about it. That is a verifiable gap, not proof of why the system selected that source. Look for supporting or contradictory results in other answers before assigning a cause.
| What you observed | Hypothesis to test | Useful check |
|---|---|---|
| Competitors are cited from their own sites | Their primary documentation answers the task more precisely | Compare page scope, evidence, freshness, accessibility and internal links |
| Competitors appear through independent publications | Third-party corroboration is stronger or more relevant | Review source quality, context, recency and whether the coverage is genuinely earned |
| The brand is cited but described incorrectly | Entity information or source consistency may be weak | Compare official naming, profiles, structured data and current page copy |
| No supplier is cited | The answer may be educational rather than commercial | Do not force a provider page into an intent that does not require one |
The existing guide on Digital PR and AI visibility explains how credible third-party sources can support discoverability without guaranteeing selection or citation.
Check whether the website contains an answer worth retrieving
For each priority group of questions, identify the page that should help the reader. If there is none, first ask whether an existing page can be improved without overlapping another. In the two-step approval example, check whether the product actually supports the feature and whether the page explains it accurately. Do not add a claim until the product team has confirmed it.
Review the source page
- One clear purpose that matches the customer decision
- Direct answers supported by examples, data or first-hand expertise
- Precise names for the company, product, people and concepts discussed
- Visible authorship, review responsibility and meaningful update dates
- Primary sources for facts that can change or carry material risk
- Useful headings, tables and media that clarify rather than decorate
- A canonical, indexable URL connected through relevant internal links
Verify access separately for each search system
Check the exact URL expected to provide the answer: status code, access without login, essential content, canonical and robots directives. Google Search Central requires a page to be indexed and eligible for a Search snippet for its generative features, without guaranteeing inclusion. OpenAI distinguishes OAI-SearchBot for ChatGPT search from GPTBot for potential model training. Perplexity separately documents PerplexityBot for search. Check rules for the relevant crawler before attributing a missing mention to a technical block.
Fetch
Check status codes, robots.txt, meta robots, authentication and crawler-specific rules.
Render
Verify that essential facts, links and navigation are available without a failed script or user interaction.
Consolidate
Confirm canonicals, redirects, duplicate variants, language alternates and sitemap inclusion.
Maintain
Keep important URLs, evidence and update signals stable after campaigns and site changes.
Record access problems as eligibility constraints, not proof that fixing them will produce a citation. The dedicated guide to AI crawlers and search visibility provides the technical implementation checklist.
Turn observations into hypotheses with owners and limits
Turn each finding into a record: observed evidence, affected URL or source, hypothesis, proposed change, owner and follow-up check. Separate business importance from confidence in the diagnosis. An incorrect price or availability claim deserves prompt correction even with a small sample; one absence from a broad prompt needs more observations before a large project is justified.
| Finding | Action | Verification |
|---|---|---|
| High-value comparison prompts omit the brand | Improve the comparison source with explicit criteria, proof and limitations | Recheck the prompt set and monitor page-level visibility over a defined period |
| The brand appears with outdated capabilities | Correct official pages and the most visible inconsistent sources | Confirm discovery, recrawl and description changes without promising timing |
| Important pages cannot be accessed | Resolve the relevant technical or crawler policy constraint | Retest access and index eligibility for the exact URL |
| The brand is mentioned but its site is not linked | Review cited sources and improve the relevant page only where verifiable information is missing | Repeat the test and log mentions, cited URLs and visits separately |
In the example, if the product team confirms the feature but the page omits it, the content owner can update the page. The team then repeats the sample and records what changed. Assign other work to product, technical SEO, analytics or communications in the same way. Findings without an established cause remain questions for the next check.
Leave a baseline that can be repeated responsibly
A useful audit package includes
- A protocol recording questions, platforms, markets and conditions needed to repeat the test
- A dated answer and citation log with source types and sample limitations
- Finding records with evidence, hypotheses, owners, action order and review dates
The evidence log and action plan should let someone else repeat the test without reconstructing its conditions from memory. If you need to move from diagnosis to implementation, Qreativa’s AI SEO service connects content, technical access and measurement.
AI search visibility audit FAQs
How many prompts do you need to start?
There is no universal threshold. Start with priority decisions and include questions with genuinely different constraints. A justified set of 30 questions may be more useful than 500 near-duplicates; document what the sample leaves out.
Can you run a first audit without a paid tool?
Yes. Choose a small set, test the search experiences you can access manually, and log answers, URLs and conditions in a shared sheet. Keep the limits of manual collection and any accounts used visible in the report.
What if an answer mentions the brand but does not link to its site?
Record the mention and the missing link separately. Open the sources that are actually cited and check whether they describe the brand accurately. A mention does not prove that the system used a page on your site.
Should you repeat the same prompt?
Yes, when you need to see whether an observation holds up. Repeat the test under the same conditions and date each result. Do not count repetitions as independent customer searches or a stable ranking.
Can the audit explain why a competitor appears?
It can reveal verifiable differences between sources and content and suggest hypotheses to test. It cannot expose the system’s internal reasoning, so a source appearing alongside an answer does not prove why a brand was selected.
When should you repeat the full audit?
Repeat it when the offer, site, market or platforms under review change materially. Between full audits, monitor the most important questions while retaining the original sample and recording each change so comparisons remain meaningful.
