AI Brand Monitoring: How To Track What LLMs Say About You

AI brand monitoring is the ongoing practice of tracking what ChatGPT, Google AI Overviews and AI Mode, Gemini, Perplexity, Claude and Copilot say about your business: whether they mention you, how accurately they describe you, and which sources they cite. You do it by running a fixed set of prompts on a schedule, measuring how often you appear across repeated runs, logging errors, and fixing the pages and third-party sources behind them.

This guide is for e-commerce and Shopify teams who want a monitoring program they can actually keep running. You'll get the five things worth tracking, what research says about how unstable AI answers are (and why that changes how you measure), a 7-step setup, a log template, a manual vs tool comparison, the first-party data you already have, and a symptom, cause and fix grid for what you find.

What is AI brand monitoring?

Traditional brand monitoring watches what people say: reviews, social posts, press. Rank tracking watches where your pages sit on a results page. Neither tells you what an AI assistant says when a shopper asks "what's the best protein powder for beginners" or "is [your brand] legit." AI brand monitoring fills that gap by checking the answers themselves.

That matters because an AI answer often replaces the click. In a Pew Research Center study of 900 U.S. adults' Google browsing in March 2025, users clicked a traditional result on 8% of visits when an AI summary appeared, compared with 15% when none did, and they clicked a link inside the summary on just 1% of visits. If the summary names three brands and yours is not one of them, many shoppers never see your store at all. If it names you with last year's price or an old return window, that shapes the decision before they arrive.

How it differs from related work

• Social listening tracks what people post about you. AI brand monitoring tracks what models say about you.
• SEO rank tracking tracks positions for keywords. AI answers have no stable positions, as the research below shows, so you track presence rates instead.
• A one-off audit, like our 10-minute ChatGPT brand check, gives you a snapshot. Monitoring is the recurring version, with a fixed prompt set so results compare month to month.

What to track: the five core metrics

Most teams start by asking "are we mentioned?" That is only the first of five questions. Each one points to a different fix, so log them separately.

Cited sources are the metric teams skip most and need most. You cannot edit an AI answer, but you can edit, or ask others to edit, the pages it draws from. Knowing that a two-year-old comparison post or a stale marketplace listing keeps getting cited tells you exactly where to spend effort.

Why AI answers change every time you ask

Here is the finding that should shape every monitoring program. In January 2026, SparkToro published research run with Gumshoe in which about 600 volunteers ran 12 brand recommendation prompts through ChatGPT, Claude and Google's AI Overviews (or AI Mode when no Overview appeared) a combined 2,961 times in November and December 2025. Prompts included chef's knives, headphones and e-commerce marketing consultants.

The study found less than a 1 in 100 chance that ChatGPT or Google's AI would return the same list of brands twice, and roughly 1 in 1,000 odds of the same list in the same order. Claude was only slightly more consistent. In other words, "we rank third in ChatGPT" is not a meaningful statement. A different run, a minute later, will likely give a different answer.

What did hold up was presence. Some brands appeared in nearly every response even though their position moved around:

The authors' conclusion, which we agree with, is that visibility percentage across many prompts, each run multiple times, is a reasonable metric, and that any tool claiming to report a fixed "ranking position" in AI answers deserves skepticism. They suggested running each prompt many times, usually at least 60 to 100, and noted they had not settled how many runs is statistically enough. The study is small and was co-run by a company that sells tracking software, so treat it as strong directional evidence rather than settled science.

How to set up AI brand monitoring step by step

This setup works for a small team with a spreadsheet and scales up when you add a tool.

1. Write a fixed prompt set. Start with 20 to 40 prompts in four groups: category questions ("best [product] for [use]"), problem questions ("how do I [problem your product solves]"), brand questions ("is [brand] worth it", "[brand] reviews") and comparison questions ("[brand] vs [competitor]"). Add a few support questions shoppers really ask, such as "does [brand] ship to Canada." Pull wording from site search, support tickets and Search Console queries so prompts sound like customers, not marketers.
2. Pick your platforms. ChatGPT, Google AI Overviews and AI Mode, and Perplexity are a sensible core for most e-commerce brands. Add Gemini, Claude and Copilot if your customers use them.
3. Pick three to five competitors. Track the same prompts for them. Presence only means something next to the brands you lose sales to.
4. Control your test conditions. Use ChatGPT's Temporary Chat or logged-out sessions where you can, and record date, platform, model if shown, location and device for every run, so a change in your numbers is not just a change in setup.
5. Run each prompt more than once. For manual checks, three to five runs per prompt per platform gives you a rough presence rate. For a reliable rate on the prompts that matter most, you need many more runs, which is where tools earn their keep.
6. Log everything in one place. Use the template in the next section. Every error row gets an owner and a re-check date.
7. Set a cadence and a trigger list. Monthly is enough for most brands. Run an extra check after a product launch, price change, rebrand, policy change or PR issue.

What a monitoring log should look like

A shared spreadsheet is enough to start. One row per prompt, per platform, per run date, with these columns:

Add a sentiment column and a free-text "how we were described" column if you have time. Then roll the rows up into a monthly summary: presence rate by prompt group, by platform, against each competitor, plus a count of open accuracy errors. That summary is what leadership should see, not the raw answers.

Manual tracking vs AI brand monitoring tools

Many teams do both: a person reads answers for accuracy and framing, and a tool handles the repeated runs needed for presence rates. Before you buy a tool, ask each vendor four questions:

• Does it collect answers from the consumer apps, the APIs, or both?
• How many times does it run each prompt, and does it report presence rate rather than a "rank"?
• Does it capture the cited URLs, not just the brand mentions?
• Can you export the raw answers so you can check its sentiment and error labels?

What makes AI tools mention a brand

Monitoring tells you where you stand. To move the numbers, it helps to know what tends to go with being mentioned. Ahrefs studied 75,000 brands against AI Overview mentions in its Brand Radar data (published May 2025) and found that off-site brand signals had the strongest relationship with how often a brand was mentioned.

Branded web mentions, meaning how often your brand name appears across the web, had the highest correlation at 0.664. Branded anchors and branded search volume came next. Classic link metrics such as Domain Rating and backlink counts were weaker. Ahrefs is clear that these are correlations, mostly moderate to weak, not proof of cause. But they line up with what monitoring logs tend to show: the brands that keep getting named are the ones that show up in roundups, reviews, forums and press, not only on their own sites. Our guide to unlinked brand mentions covers how to find and grow them.

First-party data to add to your monitoring

Prompt testing shows what AI says. Your own analytics show what it does for revenue. Add these to the monthly summary:

• Google Search Console. Google's AI features documentation says AI Overviews and AI Mode traffic is reported in the Performance report under the Web search type. There is no separate AI filter, so watch for queries with rising impressions and falling clicks.
• Bing Webmaster Tools. Microsoft has been rolling out an AI Performance report that shows how often your pages are cited in Copilot and Bing's AI answers. Check whether it is live in your account.
• GA4. Build a segment for referrals from chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com and claude.ai, and compare their conversion rate and revenue with other channels. OpenAI says ChatGPT adds utm_source=chatgpt.com to many referral links, which makes those visits easier to spot. Some AI visits still land in Direct.
• Shopify orders. If you run a post-purchase "how did you hear about us" survey, add ChatGPT and other AI assistants as options. It catches the shoppers who saw your name in an answer and typed your URL later.

How to fix what monitoring uncovers

You can't edit an AI answer directly. You can change the sources it learns from and retrieves. Use the log to match each problem to its likely cause:

A few rules for the fixes

• Check crawler access first. OpenAI's crawler documentation says sites that opt out of OAI-SearchBot will not be shown in ChatGPT search answers. Blocking GPTBot, by contrast, only signals your content should not be used for model training. Perplexity and Anthropic document their own search crawlers too.
• Fix your own pages before chasing others. Google says there are no extra requirements to appear in AI Overviews or AI Mode beyond being indexed and eligible to show with a snippet, so clear, current, crawlable pages are the foundation.
• Expect two speeds of change. Answers built from live search can update once corrected pages are recrawled. Answers built from a model's training data only change when a new model ships, so those lag.
• Re-check on the log's date. Close an error only when it stops appearing across several runs, not after one clean answer.

Turning monitoring into a monthly habit

The brands that get value from AI brand monitoring keep it boring: the same prompts, the same platforms, the same log, every month, with an owner who routes every error to a fix. Start with 20 prompts and a spreadsheet this week. After two or three months you'll know which prompts matter, which sources keep getting cited and whether a tool is worth paying for.

If you're building reporting for clients or leadership, read our guide to LLM brand visibility next. If you'd rather have QCK monitor and improve how AI describes your store, see our AI search optimization service or get in touch.