How to measure your brand’s visibility in ChatGPT and Perplexity (without fooling yourself)
A practical, honest way to track how often AI assistants recommend your brand, what they say about you, and whether it's moving the business.
Key takeaways
- AI answers vary from run to run, so one screenshot proves nothing. Measure across many prompts and repeated runs.
- Build a fixed prompt set based on how real shoppers ask, and record mentions, accuracy, sources and competitors every time.
- Track rates and trends, not rankings. “Mentioned in a given share of runs” is far more honest than “we’re #1.”
- Connect visibility to outcomes with referral data, branded search trends and post-purchase surveys, while being clear about what each can and can’t prove.
Why this is harder than rank tracking
Traditional SEO gave you a tidy number: your page ranks fifth for a keyword. AI assistants don’t work that way. Answers are generated fresh each time and can shift with wording, location, conversation history, personalization and whatever sources were retrieved in that moment.
That means the most common way brands “measure” AI visibility, someone on the team asking ChatGPT a question and screenshotting the result, is close to useless. It might be a lucky run. It might be an unlucky one. Either way, it’s a sample of one.
The fix isn’t complicated. It’s just disciplined.
Build a prompt set that reflects real shoppers
Start with the questions your customers actually ask. Pull from support tickets, sales chats, reviews, on-site search and your own understanding of the category. Then group them by intent:
- Category discovery: “best clean sunscreen for sensitive skin,” “sustainable sneakers under $150.”
- Attribute-specific: “fragrance-free mineral SPF that doesn’t leave a white cast.”
- Comparison: “[your brand] vs [competitor],” “is [product] worth it.”
- Alternatives: “cheaper alternative to [category leader].”
- Brand-direct: “what is [your brand] known for,” “is [your brand] good.”
Keep the set large enough to smooth out noise and stable enough to compare over time. If you change prompts every month, you can’t tell whether visibility changed or your questions did.
Be careful with brand-direct prompts. They’re useful for checking accuracy, but naming yourself in the question will obviously inflate mention rates. Report them separately from unbranded discovery prompts.
Decide what to record
For each prompt and each run, capture the same fields:
- Mentioned or not. Did your brand or product appear at all?
- Prominence. Were you the lead recommendation, one of several, or a passing mention?
- Description. How did the assistant describe you, and which attributes did it highlight?
- Accuracy. Were price, ingredients, materials and availability correct?
- Sources cited. Which pages did it link to or reference? Perplexity shows sources openly; others do so when they browse.
- Competitors. Who else appeared, and how were they described?
The description and sources columns are where the real insight lives. If assistants keep citing one outdated roundup, or describing you with an attribute you dropped two years ago, you know exactly what to fix. Often the answer is on your own product pages, or in the third-party sources assistants lean on.
Run it the same way every time
Consistency is what turns anecdotes into data.
- Use fresh sessions or logged-out access where possible, so past conversations don’t steer the answer.
- Note the assistant, the mode (such as web search on or off) and the model where it’s visible.
- Keep location consistent. For an LA brand, decide whether you’re measuring local or national answers.
- Run each prompt more than once per cycle, since a single run can mislead.
- Pick a regular cadence, such as monthly, and stick to it.
Specialist tracking tools can automate much of this. They’re useful, but ask how they collect answers and how many runs sit behind each number before you trust a chart.
Practical tip: Log a short change note alongside every measurement cycle: new product pages, a press hit, a feed fix, a model update you’ve heard about. Months later, it’s the only way to make sense of the trend line.
Metrics that mean something
Once you have consistent data, a few simple metrics go a long way:
- Mention rate: the share of runs, across your unbranded prompts, in which your brand appears.
- Share of recommendations: your mentions as a portion of all brand mentions across those answers, compared with key competitors.
- Accuracy rate: the share of mentions where the facts about you were correct.
- Source mix: how often answers cite your site, retailers, reviews, press or forums.
Report these as trends by assistant and by prompt group. A rising mention rate on attribute-specific prompts tells you something concrete. A single “visibility score” with no explanation behind it usually doesn’t.
Connect visibility to business outcomes
Visibility is a means, not the goal. To link it to revenue, combine several imperfect signals:
- Referral traffic. In your analytics, look for sessions referred from assistant domains such as chatgpt.com and perplexity.ai, and for UTM tags some assistants add to outbound links. Traffic from Google’s AI features generally isn’t broken out separately from regular search in standard reports.
- Branded search and direct traffic. Shoppers who hear about you from an assistant may search your name later. Watch the trend, knowing many things drive it.
- Post-purchase surveys. Add “an AI assistant (ChatGPT, Perplexity, etc.)” as an option in “How did you hear about us?” It’s self-reported, but it’s direct.
Be honest about attribution. Many AI-influenced purchases leave no clean trail, and correlation isn’t causation. Triangulate, and resist claiming more than the data shows.
How not to fool yourself
The biggest risks in AI visibility measurement are human, not technical:
- Celebrating one great screenshot, or panicking over one bad one.
- Quietly adding prompts you already win and dropping ones you lose.
- Mixing branded and unbranded prompts into one flattering number.
- Crediting your latest campaign for a jump that coincided with a model update.
Good measurement is boring on purpose. Same prompts, same method, same fields, every cycle.
Measure, then improve
Measurement only matters if it drives action. Use what you learn to tighten product pages, fix inconsistent data and earn better third-party proof, then measure again. If you’re new to how assistants choose products, start with how AI assistants decide which products to recommend.
Want help setting up a measurement program you can actually trust? Explore our services or book a free strategy call.
