How Do I Know If Ai Assistants Are Talking About My Brand

• Start with a fixed list of 10 to 20 prompts, split into three types: branded prompts ("Is Fastlook any good?"), category prompts ("best AEO tools for B2B SaaS") and comparison prompts ("Fastlook vs…
• AI answers vary between runs, between engines, and over time, so a single check of a prompt can mislead in either direction.
• Weight your prompt set toward buying intent rather than vanity queries.
• A mention and a recommendation are not the same thing, and conflating them inflates your results.
• A sheet with one row per prompt per engine per run captures what matters. For instance, according to BrandIgnity, generative engine optimization (GEO) was being searched 9,200 times a month at the time of its article.

Start with a fixed list of 10 to 20 prompts, split into three types: branded prompts ("Is Fastlook any good?"), category prompts ("best AEO tools for B2B SaaS") and comparison prompts ("Fastlook vs Profound"). According to AI Brand Report, that three-part mix is a sensible starting baseline. Run each prompt in ChatGPT, Gemini, Perplexity and Claude, then record every brand named, the order it appears in, and any links shown. Do this in fresh chats so earlier conversation context does not steer the answer. The result is a plain record of whether you appear at all, and where you sit against competitors. Because a single reply is only one sample, treat the first pass as a signal to investigate, not a verdict.

Use the same prompts every cycle so results stay comparable. According to BrandIgnity, ChatGPT holds roughly 80% of the AI search market, with Gemini, Perplexity, Claude, and Copilot splitting most of the rest.

AI answers vary between runs, between engines, and over time, so a single check of a prompt can mislead in either direction. Technology.org makes this point directly: asking each question several times in fresh conversations shows whether a brand appears consistently, occasionally, or never. GEOly recommends the same discipline and notes that each question should be run more than once because answers can change. Classify each prompt by its hit rate across repeats:

• Consistent: the brand appears in most runs.
• Occasional: the brand appears in some runs.
• Never: the brand appears in no runs.

For example, a brand that appears in two of ten runs has a visibility problem worth tracking. A one-off check would have shown either a false win or a false loss instead. According to BrandIgnity, a baseline client check can be run in under an hour by pulling 10-15 commercial-intent queries, running them through a tracking tool, capturing every brand named, and calculating share of voice against the top three competitors.

Weight your prompt set toward buying intent rather than vanity queries. Category and comparison prompts reveal consideration, which is where AI answers now influence shortlists. BrandIgnity describes a baseline check of 10 to 15 commercial-intent queries. Each query is run through a tracking method that captures every brand named.

• Commercial-intent queries reveal shortlist decisions.
• Tracking captures every brand named in each answer.
• Share of voice is calculated against the top three competitors.

Specifically, that share-of-voice figure is the most useful single number for a first audit. Write prompts the way your buyers type them, including the constraints they add, such as budget, team size, and integration needs. Vague prompts produce generic answers that flatter nobody.

A mention and a recommendation are not the same thing, and conflating them inflates your results. A mention means your name appeared in the answer. A recommendation means the assistant advised a buyer to choose you.

GEOly stresses this distinction and notes that a product card is not a sale. It also advises logging paid placements and unpaid mentions separately. Tag each appearance with one of these labels:

• Recommended
• Listed among options
• Mentioned critically
• Mentioned in passing

Sentiment and accuracy need the same care. An assistant that names you but cites outdated pricing has produced a visible error, not a win.

A sheet with one row per prompt per engine per run captures what matters. Record whether the brand was mentioned, its position in the list, the sentiment, whether the facts stated are accurate, which competitors appeared, and which sources were cited.

Venture Magazine frames the core metrics as mention frequency, how AI describes the brand, competitor mentions, citations and sources, and trends over time. Source capture matters most: when a product or service page is cited, you know which pages shape the answer. According to Apple Gazette, product and service pages account for about a quarter of AI citations, which is a reason to audit those pages first.

A scoring rubric turns anecdotes into a trend. Score each answer on four axes:

• Mention: 0 absent, 1 named, 2 shortlisted.
• Accuracy: 0 wrong, 1 partly right, 2 correct.
• Sentiment: 0 negative, 1 neutral, 2 positive.
• Citation: 0 none, 1 uncited brand reference, 2 linked to your own page.

A total out of eight gives a comparable monthly figure. Verify accuracy against your own current product facts, not against memory. Repeat the audit monthly and compare totals, because the movement matters more than any single score.

A manual audit costs time but teaches you how assistants frame your category. MarketBetter recommends a manual audit of 10 buyer prompts across ChatGPT, Gemini, Perplexity and Claude, scored in a spreadsheet and repeated monthly. Expect roughly an hour for a baseline pass, longer once you add repeat runs.

Automated tools such as Fastlook's Citation Analytics and Brand Memory, which tracks citations across six engines, remove the copy-and-paste work and make repeat runs practical at scale. Profound, Peec.ai, Otterly.ai and Goodie.ai are alternatives in this category; their pricing and engine coverage are not publicly documented here, so compare them directly before deciding. Manual work suits a small prompt set; automation suits agencies or teams tracking many brands.

Fixing a wrong AI description means making the correct version easier to find and cite. Publish clear, dated pages that answer the exact buyer questions where the error appears, with plain statements of current features and pricing. Add structured data and keep an llms.txt file so crawlers can read your facts without guessing.

Fastlook's Page Engine generates and publishes such pages to WordPress, Webflow, and Shopify, and AI Feed keeps content fresh for crawlers. No publishing step guarantees that an engine updates its answer. Therefore, re-run your prompt set after each change and track whether the description shifts over the following cycles.

A monitoring method shows what assistants say today, but it cannot prove why they say it or guarantee a stable position. Hit rates can move with model updates that you do not control. Sample sizes of 10 to 20 prompts are a baseline, not statistical certainty.

Some engines personalise answers or change sources without notice. However, the audit still works as an early-warning system. It shows when your visibility falls, when a fact goes stale, or when a competitor starts winning category prompts. Those signals justify deeper work.

The specific figures and claims on this page are grounded in the following sources, reviewed at the time of writing:

• How to Track Whether AI Assistants Mention Your Clients & Brands
• How to Check What AI Says About Your Brand: Free 60-Second Audit 2026
• Do AI Assistants Recommend Your Brand? 5 Checks to Run
• How to Check Whether AI Assistants Actually Know Your Brand - Apple Gazette
• Discover What AI Says About Your Brand: 6-Step System
• Does ChatGPT Mention Your Brand? How to Check Your Visibility in AI Assistants - Technology Org

Related guides

• Software That Tracks What AI Assistants Say About Your Brand
• Check What AI Tools Say About Your Brand: 2026 Method
• AI Search for Brands: What You Need to Know Now
• Tools That Show Whether AI Assistants Mention Your Brand
• Track AI Assistant Brand Mentions: A Practical Workflow