Brand Mentions in LLMs: Auditing What AI Says About Your Business

AI answers are the new first impression

Ask ChatGPT "is [business name] any good?" or "who should I use for [service] in Birmingham?" and the answer works like a search ranking, a review score and a sales pitch rolled into one. Unlike a rankings report, you cannot see it unless you go looking. Every answer is generated on the fly, drawing on training data and live web retrieval, and if it names three competitors and not you, or states that you are closed on Saturdays when you are not, you lose enquiries without ever knowing why.

The response is not panic but process. You audit your Google rankings on a schedule; AI assistants deserve the same treatment. What follows is a monthly protocol that takes about an hour once set up, produces a comparable log over time, and tells you exactly what to fix.

Build a stable query set

Write 15 to 25 queries and keep the wording identical every month, because consistency is what makes results comparable. Cover five types:

• Brand queries: "What is [business name]?", "Is [business name] reputable?", "Who owns [business name]?"
• Factual queries: "What are [business name]'s opening hours?", "How much does [business name] charge for [service]?", "Where is [business name] based?"
• Category queries: "Best [service] providers in [town]", "Recommend a [trade] for a small business in the West Midlands"
• Comparison queries: "[Your brand] vs [main competitor]", "Alternatives to [market leader]"
• Local queries: "[Service] near [town]", phrased the way a real customer would type or say them

The category and comparison queries usually matter most commercially, because they are the ones potential customers ask before they have ever heard of you.

Run the audit the same way every time

• Use a fresh chat for every query, logged out or in a private window where the product allows it, so personalisation and chat memory do not skew results
• Run the audit on the same day each month and record the date and whichever model name the interface shows
• Ask each query once and take the first answer; follow-up prompts change the response and break comparability
• Screenshot every answer and record any sources or links the assistant cites
• Cover the big five: ChatGPT, Google Gemini, Microsoft Copilot, Perplexity and Claude, and spot-check Google's AI Overviews by searching your category queries in a private browser window

Expect variation. The same assistant can word an answer differently on consecutive runs, and that is normal. You are not chasing a single perfect answer; you are looking for stable patterns across months: consistently absent, consistently wrong, or consistently recommended.

Score and log what you find

A simple spreadsheet beats specialist tooling to start. Use one row per query per assistant, with columns for query, assistant, mentioned (yes or no), accuracy of any details, sentiment, competitors named, sources cited, and action needed. Then score each answer:

• 0 — your business is not mentioned at all
• 1 — mentioned, but with wrong or outdated details
• 2 — mentioned accurately
• 3 — recommended, or cited with a link to your site

Total the scores per assistant per month and you have a trend line. The sources-cited column deserves special attention. When Perplexity or Copilot cites a directory, a review platform or a competitor's blog to answer a question about your category, it is showing you exactly which pages feed its answers, and therefore where you need to be present or corrected.

Correct wrong and missing answers at the source

You cannot edit an AI model, but modern assistants lean heavily on live retrieval, and you can influence what they retrieve. Work through the sources in order of impact:

• Your own website: make key facts unambiguous. A clear About page, service pages that state what you do and where, and an FAQ written in plain factual sentences give assistants quotable material
• Business profiles: keep Google Business Profile and Bing Places accurate, since factual queries about hours and location often draw on them
• Structured data: Organization schema with sameAs links to your official profiles helps systems connect your name to the right entity
• Third-party citations: trade directories, professional bodies, local press and review platforms are precisely the pages assistants cite for "best X in Y" queries
• Publish category content: if you want to be the answer to "how to choose a [service] in the UK", write the page that deserves to be cited

For persistent fabrications, use the feedback controls in each product and publish a page that states the correct facts plainly. Retrieval-based answers typically update within weeks of the underlying sources changing, so corrections do pay off.

Make it a habit, not a project

Book a recurring hour, assign one named owner, and file the screenshots in a dated folder. After three months you will know which assistants matter for your category, which queries you are losing, and whether your corrections are landing. Pair the audit with GA4 referral tracking for ChatGPT, Perplexity and Copilot so you can connect what assistants say with the clicks they send. If you would rather have the audit run and acted on for you, our team can fold it into a monthly SEO retainer.