How to Monitor Your Brand’s Mentions in ChatGPT

To monitor your brand’s mentions in ChatGPT, run a fixed set of realistic customer questions on a repeatable schedule and record the answers, brand and competitor mentions, citations, and run conditions. Keep those observations separate from referral visits: mentions show how the brand appeared in sampled answers, while analytics sessions with utm_source=chatgpt.com show attributable visits from ChatGPT links.

A prompt log and a spreadsheet are enough to begin. If the sample grows, monitoring software can help organize recurring runs. Neither a small prompt sample nor a single visibility score represents every ChatGPT conversation.

1. Define what you are measuring

“Visibility in ChatGPT” can refer to several different observations. Choose the measures that answer your business question and define them before collecting results.

Measure	What to record	What it tells you
Brand mention	Whether the answer names your brand, product, or an agreed variant	Whether the brand appeared in this sampled answer
Recommendation or framing	Whether it recommends, compares, criticizes, or neutrally describes the brand	How the answer presented the brand in context
Citation	The cited URL and the page it points to	Which source the answer surfaced alongside its claims
Referral visit	Analytics visit attributed to `utm_source=chatgpt.com`	Visits attributed to ChatGPT links, not total mentions

Set a counting rule. For example, decide whether a linked citation without a written brand mention counts as visibility, and whether a mention of a parent company counts for a product. Keep the rule stable between reporting periods. OpenAI says ChatGPT adds utm_source=chatgpt.com to referral URLs, which site owners can use to identify attributed visits in analytics. That referral measure does not count answers that mention a brand without sending a visit. OpenAI publisher FAQ.

2. Build a representative prompt set

Start with questions in customers’ own words. Sales calls, support tickets, reviews, search queries, and community questions can help you find the wording and needs worth monitoring. Preserve the exact prompt text in your log.

Include several kinds of intent so the sample is not only branded searches:

Category discovery: “What are the best [category] options for [use case]?”
Comparison: “How does [brand] compare with [competitor] for [need]?”
Fit: “Is [brand/product] a good fit for [audience or constraint]?”
Drawbacks: “What are the drawbacks of [brand/product]?”
Learning: “Which sources do you recommend for learning about [category]?”

These are prompt templates, not claims about how often people ask these questions or predictions that ChatGPT will answer consistently. Keep a compact core set for trend comparisons; add a separate exploratory set when you want to discover new wording or topics. Do not silently rewrite the core prompts between checks, because that changes what the sample measures.

3. Run comparable checks and preserve evidence

Choose a cadence that fits your reporting needs, such as a recurring weekly or monthly review. Keep the interval consistent for the core prompt set.
Run each prompt in the ChatGPT surface you intend to observe. Record whether web search was requested or appeared in the interaction. Search is not necessarily active in every conversation: it may appear when a user asks for it or when ChatGPT decides it would help.
Save the exact prompt and a reviewable record of the answer. Record date, language, region if relevant, account or other context that could affect the run, and any search state you observed.
Apply your fixed counting rules. Record brand and competitor mentions, the wording or framing, cited URLs, and any sentiment label if your team uses one.
Keep the raw answer or capture with your labels. A summary rate is easier to discuss when you can inspect the observations behind it.

OpenAI describes ChatGPT search as conditional and notes that source access can be affected by technical issues, paywalls, or robots.txt settings. It also cautions that answers and citations can be incomplete, outdated, or incorrect. Treat each observation as a sample, and open cited pages before relying on them for factual claims. OpenAI transparency information, accuracy guidance, and ChatGPT Search help.

A spreadsheet schema that is easy to audit

Use one row per prompt run. Suggested columns:

Column	Example or purpose
Run date	Date and time, with timezone
Prompt ID	Stable identifier such as `category-01`
Exact prompt	Literal prompt text submitted
Surface and conditions	ChatGPT surface, language, region, account context if relevant
Search observed?	Yes, no, or unclear; note what you saw
Brand mentioned?	Yes/no using your agreed definition
Competitors mentioned	Names appearing in the answer
Framing	Recommendation, comparison, neutral mention, drawback, or your defined labels
Cited URLs	Exact links shown with the answer
Evidence	Answer text, reviewable capture, or an internal evidence reference
Notes	Changes, errors, or unusual run conditions

For a simple sample mention rate, divide the number of sampled answers meeting your mention definition by the number of valid answers reviewed, then report the numerator, denominator, prompt set, and date range beside the percentage. For example, “6 of 20 answers in the 1–30 September core prompt sample mentioned the product.” Do not present that sample as a universal ChatGPT benchmark or as the percentage of all ChatGPT users who saw the brand.

4. Check crawler access and referral analytics separately

If your goal includes having public pages considered for ChatGPT search summaries and snippets, review your site’s robots.txt and security controls to see whether they block OAI-SearchBot. OpenAI recommends avoiding that block when you want your pages considered. Crawler access supports discovery; it does not guarantee a mention or citation. OpenAI also says that a relevant link and page title may still surface in some cases when a page is disallowed. See the publisher FAQ.

Separately, configure an analytics report or filter for utm_source=chatgpt.com to inspect attributable referral sessions. Keep this report separate from the prompt log: crawler access is a discovery diagnostic, prompt checks record sampled answers, and referral analytics record visits. None is a complete count of brand mentions.

5. Choose manual tracking or monitoring software

Manual sampling is a reasonable starting point when the prompt set is small and a person can review each answer. It offers direct control over prompt wording and makes it straightforward to preserve context and raw evidence. Its workload grows with the number of prompts and cadence, and manually observed results still represent a sample.

Monitoring software may help when you need recurring runs across many prompts or want organized mention, citation, sentiment, or comparison reporting. OtterlyAI’s own materials describe daily prompt monitoring, brand coverage, sentiment, and citation tracking. These are vendor-described functions, not an independent finding about performance or a universal standard. OtterlyAI feature information and its prompt-monitoring help.

Before choosing a service, compare:

Whether you can keep your own exact customer prompts stable.
Whether it covers ChatGPT only or additional AI search surfaces.
Whether runs preserve region, language, and other context that matters to you.
Whether you can inspect saved answers and cited URLs behind summary metrics.
Whether cadence and prompt limits fit your monitoring volume.
Subscription price, retention, data-use terms, exports, and cancellation terms at purchase time.

No independent side-by-side product test is available in this guide, so do not treat a vendor’s coverage score as proof that it measures every ChatGPT answer or predicts traffic.

6. Capture cited pages as reviewable evidence

A citation can help you investigate which page appeared with an answer, but the page should be reviewed directly before you rely on the answer’s claim. A screenshot can preserve what a cited page looked like at a particular point in your review. It is supporting evidence for that page view, not a record of every user’s ChatGPT response or proof that the citation was complete or correct.

For a do-it-yourself capture, open the cited URL in a browser, wait for its content to load, capture the page or relevant element, and save the image with the run date and prompt ID. If the page depends on consent banners, delayed content, or a specific viewport, record those conditions so later captures can be interpreted consistently.

Or skip the browser setup

ScreenshotNeo is a website screenshot API and MCP server from Yorker Media. For a cited page you want to archive, a single request can return an image. See the ScreenshotNeo API documentation for request options.

curl -G "https://api.screenshotneo.com/v1/shot" -d access_key=YOUR_API_KEY --data-urlencode url=https://openai.com -o citation.webp

import requests

r = requests.get(
    "https://api.screenshotneo.com/v1/shot",
    params={"access_key": "YOUR_API_KEY", "url": "https://openai.com"},
    timeout=90,
)
r.raise_for_status()
with open("citation.webp", "wb") as f:
    f.write(r.content)

const q = new URLSearchParams({
  access_key: 'YOUR_API_KEY',
  url: 'https://openai.com'
});
const res = await fetch(`https://api.screenshotneo.com/v1/shot?${q}`);
if (!res.ok) throw new Error(`Screenshot request failed: ${res.status}`);
const bytes = new Uint8Array(await res.arrayBuffer());
await import('node:fs/promises').then(fs => fs.writeFile('citation.webp', bytes));

Replace the example URL with the cited page you are reviewing. Keep the API key private; do not put it in public client-side code. ScreenshotNeo removes cookie and consent banners, newsletter popups, and chat widgets before capture, and each of those steps can be turned off. Bot checks, blank pages, timeouts, failed loads, and cache hits are not billed; response headers report the page verdict and billing status. Its MCP server provides take_screenshot, get_page_info, and capture_pdf for AI agents. The free plan includes 1,000 screenshots a month with no card; paid plans start at $5 for 3,000 shots. That can make dated cited-page evidence easier to collect, but it does not automate or measure ChatGPT mentions.

Performance, reliability, and cost

Keep the core sample manageable. Start with a prompt set you can review consistently. Add prompts when you have a clear question, and separate exploratory runs from the stable trend sample.
Preserve raw observations. Save answers, prompt wording, timestamps, and conditions so a changed score can be investigated rather than merely reported.
Expect variation. Search may not be active in every interaction, and answers or sources may vary. Record what happened instead of assuming every run is equivalent.
Report sample scope. Include prompt count, date range, language or region where relevant, and the counting rule beside any rate. No authoritative general benchmark for ChatGPT brand mentions is established by the sources used here.
Budget for the workflow you need. Manual work costs reviewer time; software adds subscription cost and may impose prompt limits or data terms. Compare current pricing and terms directly before purchasing. ScreenshotNeo’s capture pricing is separate from monitoring software and should be considered only if you need page screenshots.

Troubleshooting

Problem	Likely cause	What to do
The brand is absent in one answer	A single answer is one sample; the prompt, search state, or source access may differ.	Keep the observation, rerun only according to your established cadence, and assess the series rather than treating one result as a trend.
Results changed after editing prompts	The sample no longer measures the same wording.	Restore the original prompt for the core series; assign edited or new wording a new prompt ID.
A citation is missing or its page cannot be accessed	Search may not have been active, or the source may be unavailable because of technical access, a paywall, or crawler controls.	Record the missing citation or access issue as observed. Check the page and relevant site controls; do not infer a mention from an inaccessible source.
Your pages are not appearing in search summaries	`OAI-SearchBot` may be blocked, or eligibility alone may not lead to a citation.	Review robots and security rules if discovery is your goal. Allowing the crawler does not guarantee a result.
Analytics shows few visits while prompt checks show mentions	A mention does not necessarily include a link or lead to a click.	Keep mention observations and `utm_source=chatgpt.com` referral sessions as separate metrics.
A monitoring score cannot be audited	The tool or process may not retain exact prompts, answer evidence, or citations.	Prefer a workflow that lets you review underlying runs; retain raw evidence and document the metric definition.
A screenshot is blank or incomplete	The page may still be loading, require interaction, or fail to render in the capture conditions.	Check the URL and page availability, then use an appropriate wait or interaction option and record the capture conditions. A screenshot cannot establish why ChatGPT did not cite a page.

Frequently asked questions

Does a ChatGPT mention mean the brand was cited?

No. Record a written mention and a linked citation as distinct observations; an answer can have one without the other.

Does allowing OAI-SearchBot guarantee ChatGPT will mention my site?

No. It supports consideration of public pages for search summaries and snippets, but does not guarantee discovery, a mention, or a citation.

Can referral analytics tell me how often ChatGPT mentioned my brand?

No. The utm_source=chatgpt.com parameter helps identify attributed visits, while prompt monitoring records sampled answers.

How many prompts should I monitor?

There is no universal number. Choose a set broad enough to represent your important customer intents and small enough to review consistently, then report its size and scope.

Should I trust every cited source?

No. OpenAI notes that answers and citations can be incomplete, outdated, or incorrect. Open the source when accuracy matters.