How to Run an AI Visibility Audit for a B2B Brand (Prompts, Sources and Repeat Tests)

If you lead marketing in a B2B business, you can feel the shift already: buyers are asking AI tools for vendor shortlists, implementation advice, comparisons, and “what should I do next?” guidance before they ever reach a search results page.

An AI visibility audit gives you a structured way to measure what these tools say about your category and whether your company appears as a credible option. It is not about “optimising for AI” in the abstract. It is about repeatable measurement: prompts in, responses out, logged evidence, and actions that improve how consistently you are mentioned, cited, and recommended.

One important caveat before we start: results change. Generative systems update frequently, personalise in subtle ways, and may answer differently across sessions. That is exactly why you need a repeatable audit method and a retrieval log rather than one-off screenshots.

What an AI visibility audit is (and is not)

An AI visibility audit measures how AI answer environments represent your market and whether (and how) they surface your company: mentions, accuracy, positioning, citations, and recommendations.

A technical SEO audit checks crawlability, indexing, site performance, internal linking, structured data implementation, rendering, and other issues that affect how search engines access and understand your website.

You often need both, but they answer different questions. The AI visibility audit asks: “When a buyer asks an AI tool for help, do we show up, and does the model point to reliable sources?” The technical SEO audit asks: “Can search engines reliably find, interpret, and rank our pages?”

Practical insight: If you have limited time, do not start by testing 100 prompts. Start by making your evidence reliable: a smaller prompt set, consistent test conditions, and a clean logging sheet. That gives you trend data you can trust when you repeat the audit next month.

Audit setup: scope, tools and a simple logging sheet

To make the audit repeatable, define your scope and standardise how you test.

Pick a scope you can actually maintain

• Markets: UK only, or UK + a second region?
• Offer area: one core service line (recommended) rather than “everything we do”.
• Persona: one buyer role (for example, Head of Operations, IT Manager, Compliance Lead).
• Platforms: 2-3 to start (for example, ChatGPT, Perplexity, Google AI Overviews/Gemini).

Tools you can use without heavy setup

• Spreadsheet for logging prompts and results.
• Private browsing / fresh sessions (or a dedicated browser profile) to reduce contamination.
• Screen capture for record-keeping. If your organisation allows it, also paste response text into the sheet for analysis.

Create a retrieval log (the minimum viable version)

Set up columns like:

• Date and time (include timezone)
• Platform (and model/version if shown)
• Location/VPN setting (if used)
• Prompt ID and prompt text
• Did the answer cite sources? (Y/N)
• Sources cited (domains + specific URLs if shown)
• Your company mentioned? (Y/N)
• Competitors mentioned (free text)
• How you were described (positioning notes)
• Recommendation type (shortlist / “top providers” / “steps to choose” / none)
• Accuracy check result (accurate / partly / inaccurate / not applicable)
• Follow-up actions (what you will change)

Practical insight: Treat this like an experiment. If two people on your team run the same prompt and get very different outputs, that is not a failure. It is a signal you need tighter test conditions and more repetitions before you draw conclusions.

Step 1: Choose representative buyer questions (without using your brand name)

The biggest mistake in AI visibility testing is starting with “Tell me about [Company Name]”. That measures branded awareness, not discovery. Most B2B growth happens earlier, when buyers do not know who to ask for yet.

Your prompt set should represent real buyer questions across the journey, written in the language a buyer would use. Keep your company name out of the initial prompts so you can measure whether the model introduces you organically.

Where to find representative questions quickly

• Sales and pre-sales: ask for the top 20 questions from discovery calls and objections that delay decisions.
• Customer success/support: gather implementation and “what do I do next?” questions.
• RFPs and procurement: extract requirements and evaluation criteria.
• On-site search / internal search logs: if you have them, they are gold.
• Existing content performance: pages that already attract problem-aware traffic often translate well into AI prompts.

Build a balanced sample (not a biased one)

A useful starter set for a time-pressed Head of Marketing is 12-20 prompts split across intent types:

• Problem framing (what is happening, what risks exist)
• Solution approaches (in-house vs outsourced, tool categories)
• Vendor evaluation (criteria, due diligence, pitfalls)
• Comparison (approach A vs approach B)
• Implementation (timeline, prerequisites, stakeholders)

Prompt templates you can adapt (non-branded)

Category discovery prompt
“I’m a Head of Marketing at a UK B2B company. We need help with [problem]. What types of providers should I consider, and how do I evaluate them?”

Evaluation criteria prompt
“Create a vendor evaluation checklist for [solution category]. Include red flags, must-have questions, and what good evidence looks like.”

Shortlist prompt
“List reputable providers for [solution category] that work with UK B2B firms. For each, explain what they are best for and cite sources.”

Alternatives prompt
“What are alternatives to using an agency for [activity]? Compare trade-offs for a small team with limited internal bandwidth.”

Once you have the non-branded prompt set, create a separate branded check to run later (same log sheet, different tab). That isolates brand-recall behaviour from discovery behaviour.

Branded check (run separately)
“What does Rebell Way do, who is it for, and how does it differ from a traditional SEO retainer? Cite sources.”

Sampling limitation to acknowledge: 12-20 prompts will not represent your entire market. That is fine. Your goal is not statistical certainty; it is a repeatable baseline that reflects your commercial priorities.

Step 2: Run platform tests and capture sources, mentions and recommendations

Now you run the exact same prompt set across your chosen platforms and record outputs consistently. If you do this casually, you end up with opinions. If you do it systematically, you get decision-making data.

Standardise your test conditions

• Use a fresh session (private window or a clean browser profile).
• Keep prompt wording identical (copy/paste from your sheet).
• Record the date/time and any visible model/version info.
• If you use a VPN for “UK-only” testing, keep it consistent and note it.

What to capture from each response

• Mentions: Is your company named? Is it a direct recommendation or a passing mention?
• Positioning language: How are you categorised (agency, consultancy, software, content partner)? Is that accurate?
• Citations and sources: Which domains are cited? Are they credible? Do they include your website, partners, directories, review sites, industry publications?
• Comparative context: Which alternatives are listed next to you (competitors, adjacent categories)?
• Factual issues: Incorrect claims, wrong locations, mismatched offerings, invented features.

Platform-specific notes (so you interpret results correctly)

• Perplexity: Often surfaces explicit citations. Great for learning which sources the system thinks are authoritative for your category.
• ChatGPT: Depending on configuration, it may provide fewer visible citations. Your audit focus becomes “does it mention you, and is it accurate and consistent?”
• Google AI Overviews/Gemini: Answers can be highly query-dependent and may vary based on location and personalisation. Treat it as a “does Google choose to synthesise and attribute information about this?” check rather than a single definitive result.

Practical insight: If the tool gives you an answer without citations, do not guess where it came from. Add a follow-up prompt like: “What sources are you using for this? Provide URLs.” Log whether it complies. The willingness (or refusal) to cite is itself a useful signal for how you should validate and act on the output.

Step 3: Repeat tests, score results, and track change over time

AI visibility is volatile. The only way to see progress is to repeat the same tests and score them.

How many repeats do you need?

For a lightweight but meaningful process:

• Run each prompt 3 times per platform in separate sessions (ideally on different days).
• Repeat the full audit monthly or quarterly depending on how often you publish and update content.

What to score (simple and decision-friendly)

Use a 0-2 scale so it stays manageable:

• Mention score: 0 = not mentioned, 1 = mentioned, 2 = recommended/shortlisted
• Accuracy score: 0 = inaccurate, 1 = partly accurate, 2 = accurate
• Citation score: 0 = no citations, 1 = citations but weak/irrelevant, 2 = credible citations (and ideally includes your site or strong third-party sources)

Sample results table (example structure)

This table is intentionally simple. You are trying to answer: (1) do we show up, (2) are we described correctly, (3) what sources does the model rely on when it talks about our space?

Sampling limitation to acknowledge: even with repeats, you are observing a moving system. Use scores to compare your own baseline over time, not to claim market share with statistical certainty.

Step 4: Turn findings into service pages, supporting articles and independent sources

The audit only matters if it changes what you publish and how you are referenced across the web. When you review the logs, bucket findings into three action types:

1) Service page improvements (high leverage)

Service pages often act as the “ground truth” for how AI systems and humans interpret what you do. Use the audit to tighten:

• Category clarity: state what you are (and are not) in plain language.
• Who it is for: roles, company types, and constraints (for example, “small in-house team”).
• Proof structure: explain process, deliverables, and how results are measured (avoid hype).
• FAQs: address evaluation questions you see in prompts (timelines, inputs required, what success looks like).

2) Supporting articles that answer the prompts you tested

If your audit prompts include “vendor evaluation checklist” and the AI output cites competitor content or generic directories, that is a signal: you need your own strong, specific content that earns citations.

Prioritise articles that:

• Reflect buyer language from sales calls
• Include concrete decision frameworks (checklists, comparisons, trade-offs)
• Reference verifiable source materials (your own documentation, publicly available standards, or clearly explained methodologies)

3) Independent sources and entity signals (often the missing piece)

If AI tools repeatedly cite third-party sites and never your domain, you may have a credibility distribution problem, not just a content gap. Without inventing “PR campaigns”, you can still take practical steps:

• Make sure your company details are consistent across credible profiles (name, location, description).
• Publish material that can be referenced: original frameworks, clear definitions, and documented processes (not empty thought leadership).
• Where appropriate, pursue legitimate mentions in industry publications or partner ecosystems that your buyers already trust.

Practical insight: When you are short on internal resources, the fastest path is usually not “more content”. It is the right content in the right order: improve the core service page first, then add 2-3 support pieces that match your highest-intent prompts, then expand from what starts getting cited.

Where Rebell Way fits if you need a scalable process

If you want to run this audit once, you can do it with a spreadsheet and discipline. The harder part is turning findings into a repeatable content system that your small team can maintain alongside approvals, campaigns, and sales support.

This is where Rebell Way typically helps:

• Through an Organic Growth Audit / Website Potential Assessment, to identify the best starting point and where visibility is realistically achievable.
• By building an SEO & GEO Content Growth System that turns SME knowledge, buyer questions, and audit findings into service-page improvements and supporting articles - then measures performance using real search data.
• For teams that need speed without generic output, SEO/GEO Article Generation can help produce structured drafts grounded in company context and source materials, ready for editorial review.

If you want support implementing a measurement-driven content process (not just a one-off report), explore how Rebell Way works at Rebell Way.

FAQ

How is an AI visibility audit different from a technical SEO audit?

An AI visibility audit measures how AI tools mention, describe, and cite your brand for buyer-style questions. A technical SEO audit checks whether search engines can crawl, index, and interpret your website effectively. They are complementary, but they do not replace each other.

Which AI platforms matter most for a B2B visibility audit?

Start with the platforms your buyers are most likely to use for research. Common choices are ChatGPT, Perplexity, and Google AI Overviews/Gemini. Use 2-3 platforms initially so you can repeat tests consistently.

How many prompts do I need for a representative audit?

A practical starting point is 12-20 non-branded prompts that reflect real buyer questions across the journey, plus a separate set of branded checks. The key is repeatability, not volume.

How often should I repeat the audit?

Monthly is useful if you publish or update content regularly. Quarterly can be enough for smaller sites or slower publishing cycles. Repeat the same prompt set and keep test conditions consistent so scores are comparable.

What should I do if the AI response is wrong about my company?

Log the inaccuracy, then fix what you control: clarify positioning on your core pages, add supporting content that addresses the exact buyer question, and strengthen credible third-party references where appropriate. Re-test to see if the representation improves over time.