To track your brand in AI assistant answers, build a fixed list of realistic buyer questions, ask them across several assistants on a regular schedule, and record whether you are mentioned, how you are described, whether it is accurate and who else appears. Because replies vary, read the results as trends, not exact scores, and compare them with enquiries.
- Use a fixed question list so results are comparable over time.
- Record mentions, accuracy, sentiment and competitors, with dates.
- Replies vary by session, so look at patterns rather than single answers.
- Pair findings with enquiry and branded search data for the real picture.
Why is tracking AI answers different from tracking rankings?
A search ranking is a position you can look up. An assistant reply is a piece of generated text that can change between sessions, wordings and product updates. There may be no list of links, no impression count and no click data, so the usual dashboards do not apply.
That does not make measurement impossible, just different. The aim is to understand three things: whether you appear, whether what is said about you is correct and how you compare with alternatives. Consistency in method matters more than precision in any one result.
How do I build a useful question set?
Write the questions your buyers would really ask, phrased naturally. Include category questions such as who provides this service in my city, comparison questions such as what should I look for when choosing a provider, and specific questions about your own brand. Draw wording from sales calls, WhatsApp chats and enquiry forms.
Keep the list modest, perhaps a couple of dozen questions grouped by theme, and do not change it casually. Adding new questions is fine, but keep the original set so you can compare like with like. Include a few in Hindi or other languages if your customers use them.
- Category questions without your brand name.
- Comparison and selection-criteria questions.
- Questions about your brand, services and reputation.
- Local and language variants your customers use.
What should I record from each answer?
Capture the date, the assistant, the exact question and the full reply. Then note whether your brand was mentioned, in what position, with what description, and whether any facts were wrong. Record which competitors were mentioned and whether sources were cited, as these show where the assistant is getting its information.
A simple spreadsheet is enough. Add columns for sentiment and for the source named, if any. Save screenshots or copied text, because replies change and you may want to look back.
- Date, assistant and exact question.
- Mentioned or not, and how it was described.
- Factual errors or outdated details.
- Competitors named and sources cited.
How do I handle the variability?
Ask each question a few times, at different times, ideally from clean sessions without chat history that could shape the reply. Treat a single appearance or absence as weak evidence. A repeated pattern across weeks is far more meaningful.
Be cautious with paid tracking tools too. Many sample a limited set of prompts and report a score that looks more precise than it is. They can be useful for scale and convenience, as long as you understand what they sample and keep your own manual checks as a reality test.
What should I do with what I find?
Act on errors first. If an assistant states a wrong address, an old service or a confused description, look for the likely source, such as an outdated directory, an ambiguous page or a mismatched profile, and correct it there. Then look at omissions: questions where you should reasonably appear but do not, which suggest content or reputation gaps.
Prioritise fixes that help customers regardless of AI: clearer service pages, better answers, fresh reviews and consistent facts. Re-test after changes, but allow time, since updates reach assistants unevenly.
How do I connect this to business results?
Combine your log with the numbers you already have: enquiries, calls, branded searches and direct visits. Add a how did you hear about us question to forms and chats, including an option for AI assistants. Some customers will say they asked an assistant, which is direct evidence that no tool provides.
Review quarterly. If mentions improve and enquiries do too, you have encouraging signals; if neither moves, revisit the fundamentals rather than chasing the tracking itself.
Step by step
- Write the question set. List realistic buyer questions by theme, including local and language variants.
- Choose assistants. Pick the assistants your customers are likely to use.
- Run and log. Ask each question on a schedule and record the full replies with dates.
- Analyse patterns. Note mentions, accuracy, sentiment and competitors across repeated runs.
- Fix and re-test. Correct sources of errors, fill content gaps and check again later.
Frequently asked questions
Can I rely on paid AI visibility tools?
They can help with scale, but most sample limited prompts and present approximate scores. Use them alongside your own manual checks.
How often should I run the checks?
Monthly is enough for most small businesses, with extra checks after major website or reputation changes.
Why does the assistant give different answers each time?
Generated replies vary by wording, session, model updates and retrieval. That is why repeated runs and trend-reading matter.
What if an assistant says something wrong about my business?
Find and correct the likely source, such as an old listing or unclear page, and re-test later. There may also be feedback options within the product.
Does tracking require special software?
No. A spreadsheet and a consistent routine are enough to begin.
Need help with this? See our GEO — Generative Engine Optimisation service or talk to Yash Parikh.