How to track whether AI answers mention you

To track your AI visibility, write down 20 to 30 questions your buyers ask, run each one several times in ChatGPT, Perplexity and Google’s AI features on a fixed schedule, and log whether your brand is named, which of your URLs is cited and which other sources appear. Your share of answers, the percentage of runs that name you, is the number to watch.

One screenshot of one answer tells you very little, because the same question gives a different answer from one run to the next. A fixed prompt set, run the same way every time, turns that noise into a trend you can act on.

This post covers whether you are named in answers at all. For the visits those answers send, see how to track ChatGPT traffic in GA4.

The short version

• Build a prompt set of 20 to 30 real buyer questions, grouped by topic, and freeze it.
• Run each prompt several times per engine, on the same day each week or month, from a clean session.
• Log five things per run: engine, date, whether you were named, which of your URLs was cited, and which other domains were cited.
• Report share of answers per topic and per engine, not a “rank”. SparkToro found AI lists almost never come back in the same order (SparkToro).
• Use Search Console for Google’s side: its generative AI reports show impressions in AI Overviews and AI Mode (Google).
• Read the cited sources. They are the pages to improve, pitch or answer.

Step 1: build a prompt set

Start from what buyers ask, not from keywords you want to rank for. Good places to look:

• Sales call notes and demo requests. Copy the questions word for word.
• Support tickets and onboarding emails.
• Your own search queries in Search Console, rewritten as full questions.
• Comparison questions: “best X for Y”, “X vs Y”, “alternatives to X”.

Aim for 20 to 30 prompts. Fewer than that and one odd answer swings the whole number. Many more and the weekly run becomes a chore that gets skipped.

Group the prompts into three to five topics, such as category questions, comparisons and how-to questions. Then freeze the wording. If you edit a prompt, treat it as a new prompt and start its history again, or you will not know whether the answer changed or the question did.

Real people phrase things in many ways. In SparkToro’s study, 142 people wrote prompts for the same need and almost no two looked alike, yet the brands in the answers stayed fairly consistent (SparkToro). So you don’t need to guess every phrasing. You need prompts that cover the intent.

Step 2: run and log the prompts

Pick the engines your buyers use. For most B2B SaaS teams that means ChatGPT with search, Perplexity and Google’s AI Overviews and AI Mode. Add others if your buyers use them.

Keep the conditions the same each time:

• Use a logged-out or fresh session where you can, so your own history does not shape the answer.
• Use the same plan and the same model setting each run, and note them.
• Run each prompt more than once per engine. Three runs is a practical minimum by hand. More runs give a steadier number: SparkToro suggests 60 to 100 runs to really know which brands an AI tends to recommend (SparkToro), which is where tools help.

For each run, log one row:

Named and cited are different things. An answer can name your product with no link, or link your docs without naming you in the text. Track both.

Step 3: report share of answers

Share of answers is the number of runs that name you divided by the total runs, per topic and per engine. If you are named in 9 of 30 runs on comparison prompts in Perplexity, that is 30% for that cell.

Report it as a trend over time, with the competitors’ shares next to yours. A single week means little. Several weeks of the same prompts start to show a direction.

Don’t report a position. SparkToro asked 600 volunteers to run 12 prompts through ChatGPT, Claude and Google’s AI 2,961 times. There was less than a 1 in 100 chance of the same list of brands twice, and about 1 in 1,000 for the same list in the same order (SparkToro). The same study concluded that visibility percentage across many runs is a reasonable metric, while “ranking position in AI” is not. The research was run with an AI tracking company, Gumshoe, which the post discloses, and the data was collected in November and December 2025.

Why results vary run to run

Three things move the answer even when nothing about your site changes:

• The model samples. These tools generate a new answer each time, so the list and its length shift.
• The search step changes. ChatGPT search often rewrites your question into one or more search queries and may send more after reading the first results (OpenAI). Google’s AI Overviews and AI Mode may run several related searches, and the two may use different models, so their links differ (Google).
• Context. Location, memory and account settings can change what a search returns. OpenAI says ChatGPT may use approximate location and saved memories when it rewrites a query (OpenAI).

This is why the method matters more than any one result. Same prompts, same engines, same schedule, several runs each.

Check you can be found at all

Before you read a low share as a content problem, check access. OpenAI says sites that opt out of OAI-SearchBot will not be shown in ChatGPT search answers, though they can still appear as navigational links (OpenAI). Perplexity recommends allowing PerplexityBot so your site appears in its search results (Perplexity). For Google, a page must be indexed and eligible for a snippet to appear as a supporting link in AI Overviews or AI Mode, with no additional technical requirements (Google).

The free AI crawler checker shows which AI crawlers your robots.txt lets in. AI crawlers and robots.txt explains which ones to allow.

Google’s side: Search Console

Google gives you first-party numbers that the other engines don’t. Search Console’s generative AI performance reports show how often your URLs appeared in AI features in Search and Discover, by page, country, device and date. Google started them on 3 June 2026 and says they reached all websites on 31 August 2026 (Google).

The same data also stays in the main Performance report, under the Web search type (Google). Use the report for Google and your prompt log for the engines that don’t report to you.

For visits, GA4’s default channel group now has an AI Assistant channel for sources like ChatGPT, Gemini, Deepseek, Copilot and Grok. Clicks from AI Overviews and AI Mode count as Organic Search instead (Google). Our GA4 guide covers the setup.

Find the sources to win

The “other sources” column is the most useful part of the log. Sort it by how often each domain is cited across all runs. The list usually splits into:

• Your own pages. Note which ones get cited and which never do. The ones that never do are candidates to rewrite with the answer in the first lines.
• Third-party pages that list your rivals but not you. Comparison posts, directories, review sites and community threads. These are pages to ask for a correction, pitch or answer in.
• Docs and reference pages. If a rival’s docs are cited for a how-to question, a clear doc page of your own may be missing.

How to get cited in AI answers covers what to change on the page. The work behind it is on citations.

Tools vs a spreadsheet

A spreadsheet is enough to start. With 25 prompts, 3 engines and 3 runs each, that is 225 rows per run day. That is slow but doable by hand. It also forces you to read the answers, which is where most of the insight is.

Paid AI visibility trackers automate the runs and the charts, which helps once the prompt set is stable and you want more runs per prompt. Before you pay for one, ask how it runs prompts, how many times per prompt, which locations and models it uses, and whether it shows the raw answers. SparkToro’s advice is to pick providers that publish their methodology and show the math (SparkToro).

Whichever you use, keep the prompt set yours. If you change tools, you want to carry the same prompts across.

FAQ

What is AI visibility?

How often AI answer engines such as ChatGPT, Perplexity and Google’s AI Overviews and AI Mode name or cite your brand when people ask questions in your category. It is usually measured as share of answers across a fixed set of prompts.

How many prompts should I track?

Start with 20 to 30 real buyer questions grouped into a few topics. That is enough to show a trend without making the weekly run too slow. Freeze the wording so changes in the answers are not changes in the questions.

How often should I run the prompts?

Weekly or monthly, on a fixed day, with several runs per prompt per engine. Answers vary between runs, so a steady schedule and repeat runs matter more than running often.

Can I see AI Overviews data in Search Console?

Yes. Search Console has generative AI performance reports that show impressions in AI Overviews and AI Mode by page, country, device and date. Google says they reached all websites on 31 August 2026. The data also stays in the main Performance report under Web.

Why does my brand show up in one run and not the next?

AI tools generate a new answer each time, rewrite questions into different searches and can use location and memory. SparkToro found less than a 1 in 100 chance of the same list of brands twice. Track the share across many runs, not one result.

Does llms.txt improve AI visibility?

Not in any way you can count on. llms.txt is an optional proposal. Google’s guide says one “will neither harm nor help your site’s visibility or rankings in Google Search”, and no major AI company has said its crawlers read the one on your site. It doesn’t allow or block anything. Allowing the search crawlers in robots.txt matters more.

Want help setting this up?

Tracking AI answers sits next to the rest of your measurement on analytics, and the work to raise the number is on AI search. How we price the work is on the pricing page.

If you want us to build the prompt set and the log with you, book a free 15-minute call.

Book your free Growth Check