Start with the buying journey
An AI visibility audit should begin with buyer questions, not with your brand name.
Branded prompts are useful for checking accuracy, but they do not tell you whether the business appears during discovery. A buyer who already knows your name is at a different stage from someone asking which type of provider they need, how two approaches compare or who is credible in their market.
Choose five to ten prompts across three stages:
- Problem recognition: “Why is our website traffic no longer converting?”
- Category understanding: “What is AI search visibility?”
- Provider discovery: “Which UK agencies help B2B brands appear in AI answers?”
- Comparison: “What is the difference between GEO and SEO?”
- Decision support: “What should I ask an AI visibility consultancy before hiring them?”
Use the language a real buyer would use. Internal service terminology can hide the questions the market is actually asking.
Test more than one answer engine
Different systems can produce different answers to the same prompt. They may use different sources, interpret the question differently or present recommendations in a different format.
For a lightweight audit, test the same prompts in at least two systems. Keep the prompt wording identical on the first pass so the comparison remains useful.
Record:
- Whether your brand appears
- How it is described
- Which competitors appear
- Which sources are cited
- Whether the answer is accurate
- Whether the system names a clear category leader
- Which concepts or phrases repeat across answers
Do not reduce the result to a single yes or no. A brand can be mentioned inaccurately, appear for an irrelevant category or receive a weak description that gives the buyer no reason to investigate further.
The quality of the mention matters as much as its presence.
Separate visibility from accuracy
The audit should score two different problems.
Visibility asks whether the brand appears when it should.
Accuracy asks whether the answer describes the brand correctly.
These require different responses. Low visibility usually points toward missing authority, weak category alignment or a lack of useful supporting content. Poor accuracy often points toward inconsistent company descriptions, outdated pages or unclear entity signals across the web.
A simple recording table might use four outcomes:
| Outcome | What it means |
|---|---|
| Visible and accurate | The brand appears in the right context and is described correctly |
| Visible but inaccurate | The system knows the brand but has weak or conflicting information |
| Absent but competitors appear | The category is understood, but your evidence is not strong enough |
| No clear brands appear | The prompt may be too broad, too early or poorly served by existing sources |
This distinction prevents a common mistake: publishing more content when the real issue is that the existing information is contradictory.
Inspect the cited sources
The sources inside the answer show what the system currently considers useful.
Do not copy them. Study why they work.
Look at the pages that receive citations and ask:
- Do they answer the question near the top?
- Are they more specific than your equivalent page?
- Do they include definitions, examples or comparisons?
- Is the publisher’s expertise clear?
- Does the page connect to a broader cluster of relevant content?
- Is the information current and internally consistent?
You may find that the cited page is not the most polished or the longest. It is often simply the easiest one to use as evidence for the answer being assembled.
The purpose of this step is to identify the missing information shape. Your site may need a clearer definition, a more useful comparison, a better proof page or a stronger explanation of how your service works.
Find the first useful action
An audit is only valuable when it produces a sensible next step.
Group the findings into four types:
- Positioning: The market or system cannot tell exactly what category you belong to.
- Content: The site does not answer an important buyer question clearly enough.
- Authority: Useful claims exist, but they are not supported by sufficient evidence.
- Consistency: Names, descriptions, locations or service information conflict across sources.
Choose one high-value prompt and trace the issue back to the smallest meaningful fix.
That fix might be rewriting the opening of a service page, publishing a focused comparison, strengthening an About page, adding first-hand examples or aligning company descriptions across major profiles.
Avoid turning the audit into a list of fifty tasks. The aim is to find the strongest constraint, improve it and test again.
Repeat the same test
AI visibility is not measured well by occasional screenshots. Answers change, sources change and your market continues publishing.
Keep the original prompts and repeat them on a regular schedule. Monthly is enough for many small B2B brands. More active categories may need a more frequent check.
Use the same baseline prompts, then add a small number of new ones as buyer language develops. That gives you a stable comparison without freezing the audit in time.
Over several rounds, patterns become more useful than individual answers:
- Which competitors appear consistently?
- Which pages are repeatedly cited?
- Which descriptions of your category are becoming standard?
- Where has your brand moved from absent to present?
- Has accuracy improved after your changes?
A 30-minute audit will not explain every ranking or retrieval decision. It will do something more practical: show you what buyers are being told today and reveal the clearest place to improve it.