How to Monitor Brand Mentions in ChatGPT

Social listening tools tell you when someone posts about your brand. Google Alerts tells you when a page mentions you. Neither tells you what ChatGPT says when a buyer asks it to recommend a company in your category.

That conversation is private, and it happens many times a day. You can't see other people's chats, but you can do the next best thing: ask the same questions your buyers ask, consistently, and track the answers over time. This guide shows how.

Step 1: Build a prompt set that mirrors real buyers

Your monitoring is only as good as your questions. Mix four types:

Prompt type	Example	What it tells you
Category discovery	"What are the best accounting tools for freelancers?"	Whether you make the shortlist at all
Use-case / persona	"Which CRM is easiest for a three-person real estate team?"	Whether you're matched to your ideal customer
Comparison	"[Your brand] vs [Competitor]: which is better for small teams?"	How you're positioned head to head
Brand direct	"What is [your brand] and how much does it cost?"	Whether facts about you are accurate and current
Reputation	"Is [your brand] reliable? What do customers complain about?"	Sentiment and recurring objections

Tips for a good set:

Start with 20 to 50 prompts and keep them fixed, so results stay comparable week to week.
Write them the way customers talk, not the way you'd write a keyword list.
Include location or industry qualifiers if they matter to how you sell ("in Toronto", "for dental clinics").
Add a few prompts where you currently don't appear. Those are your growth targets.

Step 2: Run each prompt more than once

AI answers vary. The same prompt can return a different list of brands a minute later. A single check is an anecdote; several checks give you a rate.

Run each prompt several times per cycle (three is a sensible minimum) and record each result separately.
Use fresh sessions with no memory or custom instructions, so your own history doesn't shape the answer.
Note whether web search was used in the answer. Answers with live search behave differently from answers drawn only from training.
Keep the same cadence: weekly suits most brands; daily makes sense only during launches or a reputation issue.

Step 3: Record the right fields

For every answer, capture the same fields. This turns chat transcripts into data you can analyse.

Field	How to record it
Mentioned?	Yes / No
Position	Order in which your brand appears (1st, 2nd…) or "not listed"
Competitors named	Every other brand in the answer
Description accuracy	Correct / partly wrong / wrong, with a note of what's wrong
Sentiment	Positive / neutral / negative framing
Cited URLs	Every source link, and whether any are yours
Search used?	Whether the answer showed web sources
Date and model	So changes can be traced to model updates

Step 4: Turn records into metrics

Once you have a few weeks of records, five numbers summarize your position:

Metric	Formula	Why it matters
Mention rate	Answers mentioning you ÷ total answers	Your overall presence
Average position	Mean position when mentioned	Being 1st vs 6th changes click-through and perception
Share of voice	Your mentions ÷ all brand mentions across the set	How you compare to the whole competitive field
Accuracy rate	Accurate descriptions ÷ answers describing you	Catches outdated prices, features or positioning
Owned citation rate	Answers citing your URLs ÷ answers with citations	Whether your pages are trusted as sources

Break each metric down by prompt type. Many brands find they score well on "brand direct" prompts and poorly on "category discovery", which is the gap that actually costs them customers.

Step 5: Act on what you find

Monitoring only pays off when it changes what you do. Typical patterns and responses:

What you see	Likely cause	What to do
Not mentioned in category questions	Weak off-site reputation in that category	Earn reviews, comparison-list inclusions and industry coverage
Mentioned, but described wrongly	Outdated pages or third-party listings	Update your pages and correct listings and old press
A competitor always listed first	Stronger presence in cited sources	Study the pages cited for them; fill the gaps on yours
Third-party pages cited, never yours	Your pages lack a direct, quotable answer, or are hard to crawl	Add clear answer sections; check crawler access and rendering
Negative sentiment	Recurring complaints in reviews or forums	Fix the underlying issue; respond publicly; ask happy customers for reviews
Sudden drop across many prompts	Model update or shift in which sources are retrieved	Check whether cited sources changed before changing your strategy

Manual tracking vs a monitoring tool

You can start with a spreadsheet. It's a great way to learn your landscape. But it gets heavy quickly: 30 prompts × 3 runs × 4 engines is 360 answers a week to read and log.

	Manual (spreadsheet)	Monitoring tool
Cost	Your time	Subscription
Scale	A few dozen prompts, one engine	Hundreds of prompts, several engines
Consistency	Varies with who's checking	Same method every run
History & trends	You build the charts	Built in
Alerts	None	Notifies you when something changes
Best for	Getting started, small budgets	Ongoing tracking, agencies, several brands

Whichever route you choose, keep the method fixed. Changing prompts or cadence every month makes trends meaningless.

Common mistakes

Checking once and panicking. One bad answer is noise. Look at rates over weeks.
Only asking about yourself. "What is [brand]?" will nearly always mention you. Category questions are where visibility is won.
Ignoring the sources. The cited URLs show you exactly which pages shape the answer.
Testing while logged in with history. Personal memory and instructions skew results.
Tracking ChatGPT alone. Buyers also use Gemini, Google AI Mode, Perplexity and Claude, and each draws on different sources.

Key takeaways

Ask like a buyer	20–50 fixed prompts across discovery, use-case, comparison, brand and reputation
Repeat runs	Answers vary, so measure rates, not single replies
Five metrics	Mention rate, position, share of voice, accuracy, owned citations
Read the sources	Cited URLs show which pages shape what ChatGPT says
Close the loop	Every finding should lead to a page update, a correction or an outreach task