To track your brand's visibility in AI answers, write a fixed set of 20–40 questions your customers actually ask, run them in ChatGPT, Perplexity, Gemini and Google's AI results each month, and log whether your brand is mentioned, cited with a link, described accurately, and which competitors appear instead. Repeat each prompt a few times, because answers vary.
There's no Search Console for AI assistants yet, so you have to build your own panel. The good news is that a spreadsheet and two hours a month will tell a small business most of what it needs to know. This guide shows exactly how to set it up, and it plugs into the wider strategy in our guide to AEO, GEO and ranking in AI answers.
What exactly should you measure?
Measure presence, prominence, accuracy and competition, not just "did we show up". A brand that's mentioned but described wrongly has a different problem from one that's never mentioned at all.
| Metric | What it means | How to calculate it |
|---|---|---|
| Mention rate | How often your brand name appears in answers | Answers mentioning you ÷ total answers checked |
| Citation rate | How often your website is linked as a source | Answers citing your domain ÷ total answers checked |
| First-mention position | Whether you're named first, middle or last in lists | Average position when mentioned |
| Accuracy | Whether facts about you are correct | Answers with errors ÷ answers mentioning you |
| Sentiment | Whether you're described positively, neutrally or with caveats | Manual label per answer |
| Competitor share | Which rivals appear when you don't | Count of mentions per competitor |
Citation rate and accuracy are the two that most directly affect enquiries. A link sends visits; an error can send them elsewhere.
Step 1: How do you build a prompt set?
Write questions in your customers' words across the stages of their decision, then freeze the list so month-on-month results are comparable. Twenty to forty prompts is a good size for a small business; fewer and the results swing wildly, more and you won't keep it up.
Cover five types of question:
- Category questions: "What are the best study-abroad consultants in Hyderabad?"
- Problem questions: "How do I get a student visa for Australia from India?"
- Comparison questions: "Is it better to use a consultant or apply for a visa myself?"
- Pricing questions: "How much do study-abroad consultants charge?"
- Branded questions: "Is [your brand] good?" and "What does [your brand] do?"
Pull wording from real sources: enquiry form messages, WhatsApp chats, sales call notes and the queries in Google Search Console. If your CRM stores the first message of every lead, that's a goldmine of natural phrasing. Avoid writing prompts that include your own brand except in the branded group; otherwise you'll overstate visibility.
Step 2: How should you run the checks?
Use consistent conditions every time: the same platforms, a fresh session, the same location, and several runs per prompt. AI answers change between sessions, so a single check can mislead.
Practical rules:
- Platforms: ChatGPT (with search), Perplexity, Gemini, and Google results with AI Overviews or AI Mode. Add Microsoft Copilot if your buyers are corporate.
- Fresh sessions: use a logged-out or new private window so past chats don't personalise the answer.
- Location: for local businesses, check from the city you serve or set location where the tool allows it.
- Repeats: run each prompt three times and record how many of the three mention you. "Mentioned 2 of 3" is far more useful than a yes or no.
- Timing: same week each month. Avoid checking right after you've made big site changes and expecting instant shifts.
Consistency matters more than volume. A small panel checked the same way every month beats a large one checked randomly.
Step 3: What should your tracking sheet look like?
One row per prompt per platform per month, with columns for each metric and a notes field. Keep it simple enough that whoever runs it can do so in under two hours.
Suggested columns:
- Month, prompt, prompt type, platform
- Mentioned (0–3 out of three runs)
- Cited with link (0–3)
- Position when mentioned
- Accurate? (yes, partly, no) and what was wrong
- Sentiment (positive, neutral, negative)
- Competitors named
- Sources cited (domains), which shows you which pages and directories the AI trusts
That last column is quietly the most valuable. When an assistant cites the same review site, directory or industry publication again and again, you've found a place your brand needs to be.
Step 4: How do you turn results into a monthly scorecard?
Summarise each month into four numbers per platform and one short list of actions. Owners don't need the raw sheet; they need the trend and the next move.
A one-page scorecard might show:
- Mention rate across all prompts, by platform.
- Citation rate, by platform.
- Accuracy issues found this month, listed.
- Top three competitors by mention count.
- Two or three actions, each tied to a finding.
Pair it with the traffic side covered in our guide to measuring clicks from AI Overviews: visibility tells you whether you're in the answer, analytics tells you whether that's bringing visits and enquiries.
Should you track manually or use a tool?
Start manually so you understand the data, then consider a tool once you know which prompts matter. Several SEO platforms now offer AI visibility tracking, and each has trade-offs.
| Approach | Typical effort | Strengths | Weaknesses |
|---|---|---|---|
| Manual spreadsheet | 1–3 hours a month for 20–40 prompts | Free, you see real answers, catches nuance and errors | Small sample, depends on discipline |
| SEO platform AI-tracking add-on | Set up once, reviewed monthly | More prompts, more frequent checks, trend charts | Subscription cost; methods vary and aren't always transparent |
| Custom script using AI provider APIs | Developer time up front | Large samples, your own logic, cheap to run | API answers can differ from what users see in the consumer apps |
Whichever you choose, keep a small manual check alongside it. Reading actual answers is how you spot a wrong phone number or an outdated service description that a dashboard reduces to "mentioned: yes".
What do you do with what you find?
Match each finding to a specific fix, because different visibility problems have different causes.
- Never mentioned: you likely lack content that answers these questions clearly and lack mentions on the sites AIs trust. Start with our guide to getting your brand recommended by ChatGPT.
- Mentioned but not cited: the assistant knows of you but prefers someone else's page as a source. Restructure your relevant page so it answers the question directly; see how Perplexity chooses its sources.
- Described inaccurately: find where the wrong fact comes from. It's often an old directory listing, an outdated page on your own site or inconsistent details across profiles. Fix the source, not the symptom.
- Competitor dominates: read the pages cited for them. Note their structure, specifics and where else they're mentioned, then close the gap honestly.
- Visibility drops suddenly: check whether you've blocked a crawler, changed URLs without redirects or removed a page that used to be cited.
Give changes time. Some assistants pick up page updates within weeks; others reflect broader reputation that shifts over months.
Frequently asked questions
How many prompts do I need to track?
For most small businesses, 20–40 prompts gives a useful picture without becoming a chore. Cover category, problem, comparison, pricing and branded questions, and keep the list fixed for at least six months so trends mean something.
Why do I get different answers each time I ask the same question?
Language models generate answers with some randomness, and search-backed assistants may retrieve different pages each time. That's why running each prompt several times and recording a score out of three is more reliable than a single check.
Should I check AI answers while logged in?
Use logged-out or fresh sessions for tracking, because logged-in assistants may personalise answers based on your history. Your own past interest in your brand can make it appear more often than it would for a new customer.
Can I track AI visibility through Google Search Console?
Only partially. Search Console includes AI Overview and AI Mode data within normal Web results without separating it, and it says nothing about ChatGPT, Perplexity or Gemini. You need a separate prompt panel for those.
How quickly will visibility improve after I make changes?
It varies by platform and by the type of change. Fixing crawl access or restructuring a page can show up in search-backed assistants within weeks, while building broader reputation through mentions and reviews tends to take several months.
Want a visibility baseline without the busywork?
We can build your prompt panel, run the first audit and set up a simple monthly scorecard tied to your enquiries, then help fix what it uncovers. Explore our services or tell us which questions matter to your buyers and we will quote a fixed price.