Guides · 24 October 2026

Measuring your brand in AI answers: a practical guide

More and more buyers ask ChatGPT, Gemini or Perplexity before they ask Google. Some brands get named in those answers, others do not. You can measure which side you are on, and whether it changes, with a handful of prompts and a little statistics.

Start with your buyers’ questions

Write down the questions a buyer asks before they know your name: “what is the best CRM for a small agency?”, not “is HubSpot good?”. Ten to twenty prompts cover most markets. Write them in your buyers’ language and keep the list stable from week to week, because every change to it breaks the comparison with earlier weeks.

Track your competitors along with your brand. A mention rate on its own says little. Next to three competitors it tells you who the engines prefer.

One answer is not a measurement

Ask the same question twice and you often get two different answers, with different brands in a different order. A single screenshot can show you at the top today and missing tomorrow, without anything having changed. So ask each prompt several times and count: “named in 4 of 6 answers” is a measurement, one answer is an anecdote.

Report a range with the rate. With 4 hits in 6 answers the true rate is likely somewhere between 30 and 90 percent (a 95 % Wilson interval). That sounds wide, and it is honest. Three samples per prompt are a good start, five or more when the numbers go into a report.

Tell real changes from noise

Last week 3 of 5, this week 1 of 5. That looks like a drop, but chance produces a gap like this about half the time (Fisher’s exact test, p = 0.52). With 10 samples, a fall from 9 of 10 to 3 of 10 is a different story: p = 0.02, and you can treat it as real. Run the same prompts weekly, compare each rate with the previous run, and react only to the changes that pass the test.

Count mentions with plain rules

It is tempting to let another model read the answers and judge them. That adds a second source of randomness on top of the first. Fixed rules work better and can be checked: whole words only, case and accents ignored (“Mueller” finds “Müller”), the brand’s other names and domains included. Keep cited links apart from mentions, because a source link to your site is a citation, not the engine recommending you. Where your brand appears matters too: first in a numbered list is worth more than a footnote.

Look at the sources

With web search on, the engines cite the pages they used. Those citations are the most useful part of the whole exercise: they show which review sites, comparison articles and directories shape the answers in your market. If the same three sites keep coming up and you are missing from all of them, you know where to work.

Measure through the official APIs and you measure the models themselves: the same questions, asked the same way, by a fresh anonymous user. Answers in someone’s app can differ because of chat history and personalisation, which is why the API gives the cleaner trend.

The short way: a weekly task

The AI Search Visibility Tracker does all of the above on Apify: it asks your prompts several times on ChatGPT, Gemini and Perplexity, counts mentions with fixed rules, reports every rate with its range, and marks changes as significant or not. Save this input as a task and give it a weekly schedule:

{
  "brand": "HubSpot",
  "brandDomains": ["hubspot.com"],
  "competitors": ["Pipedrive | pipedrive.com", "Salesforce | salesforce.com"],
  "prompts": ["What is the best CRM for a small marketing agency?"],
  "engines": ["chatgpt", "gemini", "perplexity"],
  "samplesPerPrompt": 3
}

Every answer becomes a row with the mentions, list rank, citations and the full text. The summary holds mention rate, citation rate and share of voice per engine. To read it in your own code, use the Apify Python client:

from apify_client import ApifyClient

client = ApifyClient("<YOUR_APIFY_TOKEN>")
run = client.task("<YOUR_TASK_ID>").call()
summary = client.key_value_store(run["defaultKeyValueStoreId"]).get_record("OUTPUT")["value"]
for engine, s in summary["engines"].items():
    rate = s["brandMentionRate"]
    print(engine, rate["hits"], "of", rate["samples"], rate["ci95"], rate.get("change"))

AI agents can call it directly as a tool through the Apify MCP server: https://mcp.apify.com?tools=layerwiselabs/ai-visibility-tracker

What it costs

From $0.02 per answer on ChatGPT, $0.025 on Perplexity and $0.03 on Gemini, plus only the web searches the engine really runs. 10 prompts on Perplexity with 3 samples each come to $0.75 per run. With your own OpenAI, Gemini or Perplexity key the Actor charges $0.003 per answer, and a free Apify account includes $5 of credit every month.