How to Run Your First AI Visibility Audit: A Practical Guide for Brands and Agencies

To run your first AI visibility audit, define the brand and competitors you are assessing, test a fixed set of buyer questions across named answer surfaces, record mentions, recommendations and citations separately, and check whether your key pages can be crawled and indexed. Keep raw answers and dates attached to every finding.

An initial audit gives brands and agencies a dated baseline. It records where your company is absent and where competitors are recommended, then helps you investigate possible content, evidence or access gaps. It sits within Generative Engine Optimisation (GEO) and Answer Engine Optimisation (AEO).

Key takeaways

• Separate technical website readiness from observed brand appearances in AI answers.

• Test a fixed cohort of unbranded buyer questions alongside a smaller set of branded accuracy checks.

• Count mentions, positive recommendations, website citations and factual accuracy as four distinct outcomes.

• Preserve the exact prompt, date, collection surface and denominator for every baseline observation.

What an initial AI visibility audit can and cannot prove

A baseline audit answers two practical questions: can search and answer systems access clear information on your website, and how does your brand currently appear when a realistic buyer asks for options in your category? Use both sets of findings to form testable explanations rather than assuming why a competitor appeared.

Be clear about the boundaries of your sample from the start. Recording answers to fifteen or twenty buyer questions establishes what happened under those specific conditions on that date. It does not measure total AI query volume, represent every personalised consumer conversation or prove that one page edit will guarantee a future citation.

Treating the audit as structured diagnostic evidence keeps client and stakeholder expectations realistic. You can verify whether an important page blocks search crawlers, whether a pricing page omits service limits, or whether an answer cites an outdated directory. Each finding leads to a concrete, testable fix.

Step 1: Define the brand, entities and competitor scope

Start by writing down the exact entities you are auditing. Ambiguous company names, regional trading names and product sub-brands often cause false matches or missed detections if they are not defined before you collect responses.

• Primary brand and domain: record the legal name, common trading name, primary domain and any legacy brand names that still appear online.

• Core products or services: list the specific offerings that belong in this audit rather than auditing every division at once.

• Target geography and buyer profile: state the market and customer type, such as B2B software for UK agencies or commercial legal services in South Wales.

• Direct competitors: select three to five named competitors that prospective buyers genuinely compare during sales conversations.

Keeping the scope narrow makes the first audit manageable. A multi-service company or an agency onboarding a new client should audit one commercial category well before expanding to secondary product lines.

Step 2: Select a small, representative buyer question set

Choose ten to twenty questions that reflect how real customers evaluate the category. Avoid testing single keywords such as “best CRM” or “SEO agency”. Instead, write full questions that include the job to be done, the buyer’s constraints and the relevant market context.

Divide your baseline questions into four clear intent groups so you can see where visibility breaks down:

1. Problem and approach questions: how a buyer frames the operational task before naming vendors.

2. Category shortlist questions: unbranded questions asking which providers suit a specific audience, budget or location, such as “Which AI visibility platforms suit a digital marketing agency in Wales managing multiple client domains?”

3. Comparison and criteria questions: prompts asking how to evaluate options or compare two approaches for a stated requirement.

4. Branded verification questions: direct questions about your company’s capabilities, pricing model, integrations or service boundaries.

Keep unbranded discovery prompts and branded verification prompts in separate groups when you calculate results. Supplying your brand name in the prompt tests whether the engine describes your company accurately; it does not test whether the engine recommends you to an undecided buyer.

Step 3: Record dated observations with consistent collection settings

Run your baseline questions under consistent, documented conditions. An answer collected through a consumer web interface, a search-enabled API response and a direct model response without web retrieval are different observation surfaces. Label which surface you used and keep it consistent across the baseline.

For every prompt run, capture a complete observation record:

• Prompt identifier and exact text: preserve the exact wording, spelling and language submitted.

• Engine and collection surface: record the provider, product surface or API mode, model name where known, and whether web search was enabled.

• Timestamp and location context: note the date, time and any geographic setting applied to the check.

• Run status: distinguish completed answers from timeouts, rate limits or provider errors so failed runs are never counted as brand absences.

• Full response and source records: save the complete answer text alongside any attached citations, consulted source lists or printed URLs.

Google Search Central explains that AI Overviews and AI Mode may use different models and techniques and can issue multiple related searches to build a response. Preserving the exact surface and full answer text ensures your team reviews what was actually returned rather than relying on memory.

Step 4: Score brand mentions, recommendations, citations and accuracy

Review each completed response and score four outcomes separately. Combining them into a single pass-or-fail mark hides the difference between being named in passing and being recommended with verifiable proof. Our guide to interpreting AI visibility evidence covers these definitions in detail.

• Brand mention: did the answer name the audited brand or product, and was the reference about the correct entity?

• Positive recommendation: did the answer present the brand as a suitable option for the buyer’s stated requirement, rather than listing it with a disqualifying caveat?

• Website citation: did the answer attach a source link to the brand’s own domain, and which specific URL and claim did it support?

• Factual accuracy: did the answer state the brand’s capabilities, location, pricing structure or service scope accurately when checked against current approved facts?

Record the same mention, recommendation and citation counts for your named competitors across the unbranded prompts. Always state the denominator alongside any summary figure, for example: “Recommended in 3 of 12 completed unbranded shortlist answers on the defined surface.”

Inspect the URLs cited when competitors are recommended. Classify them as official product pages, documentation, independent directories or industry articles. A citation records the source link presented; check that the linked page supports the relevant claim before treating it as supporting evidence.

Step 5: Check technical and content readiness on priority pages

Answer observations record what happened in a sample of prompts. A technical readiness check helps identify crawl access, indexing and content clarity issues. Keep the findings distinct: access alone does not guarantee that a page will be indexed, cited or recommended.

Start with your homepage, primary product or service pages, pricing page and core comparison or educational guides. Verify four technical foundations:

• Search indexing and snippet eligibility: for Google's AI features, confirm that priority pages are indexed and eligible to appear in Search with a snippet. Also check the property's Search generative AI control in Search Console, including inherited settings, to see whether the site is excluded from those features. No special AI files or special schema are required.

• Purpose-specific AI crawler rules: check your robots.txt file and firewall rules against official provider documentation. OpenAI documents separate user agents for search surfacing (OAI-SearchBot), potential model training (GPTBot) and user actions (ChatGPT-User). Confirm you have not blocked search crawlers while attempting to restrict training crawlers.

• Readable HTML content: check that essential product facts, audience fit, pricing boundaries and geographic coverage appear in clean, crawlable HTML rather than hiding behind client-side scripts or unlabelled tabs.

• Aligned structured data: ensure any Organization, Product, Service or Article schema matches the visible text on the page.

Use NeuraCite's free GEO readiness report for the website diagnostic portion. It checks crawl access, indexing signals, structured data and content, with prioritised findings across up to 500 pages, subject to access and availability. No account or card is required; your private report link expires seven days after your request.

Because the free report diagnoses website readiness rather than running live LLM prompts, pair it with your prompt observations or Answer Engine Insights to build a broader baseline.

Adapting the audit for an agency prospect pitch

Agencies frequently use a compact version of this audit during new-business discovery or pitch preparation. A prospect audit is most persuasive when it is transparent, bounded and immediately actionable rather than alarmist.

For a pitch, limit the scope to five to eight high-intent unbranded buyer questions and two branded verification questions relevant to the prospect’s main revenue line. Combine those dated answer observations with a technical readiness check of the prospect’s top commercial URLs.

Show the prospect the exact questions asked, the saved answers, the cited competitor URLs and the specific information missing from their own site. Do not present this small prompt sample as total market share. Instead, propose a thirty-day plan: resolve technical blocks, close commercial page gaps and establish a recurring baseline cohort.

A fictional example: baseline audit for a Welsh service brand

This example is fictional. It illustrates how to combine prompt observations and technical checks in an audit summary, rather than reporting a real customer or NeuraCite platform benchmark.

Suppose an agency in Newport audits a fictional Cardiff-based B2B payroll firm, “Severn Payroll Partners”. The agency tests twelve unbranded buyer questions about outsourced payroll for UK and Welsh employers, plus three branded verification questions, on one defined search-enabled surface. All fifteen checks complete.

• Unbranded observations (12 completed answers): Severn Payroll Partners is named in two answers, recommended as a suitable option in one answer, and receives zero website citations. Two national competitors are recommended in nine of the twelve answers, often supported by citations to their dedicated employer-size and pricing pages.

• Branded observations (3 completed answers): the firm is named in all three answers, but one answer states an outdated minimum employee threshold taken from an old directory listing.

• Technical and page readiness findings: the firm’s main service page is indexed and allows Googlebot and OAI-SearchBot, but it combines all employer tiers onto one thin brochure page with no clear explanation of regional support, software integrations or onboarding steps.

This audit has found two actionable issues: incomplete commercial page content and an outdated third-party listing. The checked pages show no crawler block, but these observations do not establish why an engine omitted or recommended a brand. Prioritise those fixes and repeat the saved prompt cohort. For deeper gap diagnosis, review our guide on why AI engines recommend your competitors.

A reusable AI visibility audit record and checklist

Copy these eight fields into your audit spreadsheet or documentation before starting your first baseline:

1. Audit scope: brand name, primary domain, included products, target geography and three to five named competitors.

2. Prompt cohort version: numbered list of unbranded discovery prompts and branded verification prompts, with buyer stage labels.

3. Collection protocol: engine or provider, collection surface, model details, search setting, location context and audit date.

4. Run accounting: total prompts attempted, completed answers analysed and any failed runs excluded from percentages.

5. Visibility tally: separate counts and denominators for brand mentions, positive recommendations, website citations and factual accuracy.

6. Cited source log: exact URLs cited in the answers, categorised into owned pages, competitor pages and third-party publications.

7. Technical readiness log: status codes, robots.txt rules for search vs. training bots, indexing/snippet eligibility, HTML clarity and schema alignment on priority URLs.

8. Prioritised action register: specific page or technical fixes, assigned owners, verification steps and the date scheduled for the repeat check.

Planning your first repeat check

Freeze your baseline prompt cohort and observation records before making changes to the website. Start by fixing confirmed technical access or indexing barriers, then update the existing product, service or pricing pages that failed to answer core buyer questions. Create new educational articles only where a distinct instructional question has no natural home on your site.

When the updates are live and indexed, repeat the exact same prompt set under the same collection settings. For deeper measurement architecture, read our technical guide to measuring brand visibility in AI-generated answers, or explore NeuraCite Answer Engine Insights and Visibility Reports to structure your baseline workflow.