Your Google Analytics shows zero traffic from ChatGPT. That's not because ChatGPT isn't sending people — it's because ChatGPT traffic is invisible to standard analytics. Here's how to find out what ChatGPT is actually saying about your brand, and how to track it consistently over time.
1. Why ChatGPT traffic is invisible to Google Analytics
When someone uses ChatGPT to research a product and then clicks through to your site, the traffic appears as direct or “(not provided)” in your analytics — not as a referral from chat.openai.com. ChatGPT does not pass referrer headers in the standard way that search engines do. The result is a growing blind spot: AI-influenced visits are happening, but your analytics dashboard cannot see the source.
This is fundamentally different from the traditional SEO problem. With Google, you can at least see impressions and clicks in Search Console. With ChatGPT, there is no equivalent. If ChatGPT mentions your competitor in response to a category question and 200 people click through to their site today, your competitor's analytics will show direct traffic — and neither of you will know the real source.
The core problem
Google results and ChatGPT citations overlap by only about 8%. That means 92% of the brands ChatGPT recommends are not based on Google ranking. Your SEO rank tells you almost nothing about your ChatGPT visibility — they are effectively separate lists.
2. What “monitoring brand mentions in ChatGPT” actually means
Monitoring your brand in ChatGPT means systematically asking ChatGPT the questions your potential buyers are asking — and recording whether your brand appears, in what position, with what framing, and citing which sources. It is not a one-time check. ChatGPT responses change as the underlying model is updated, as new training data is incorporated, and as the web pages it pulls from are updated.
The goal is a consistent, repeatable baseline — not a snapshot. A single query run in isolation tells you almost nothing. A weekly measurement against the same set of prompts, compared to the same set of competitors, over 90 days tells you whether your brand is gaining or losing ground in the AI layer.
3. The manual method: step by step
Manual monitoring is slower but requires no tools. It is a good starting point for understanding your baseline before committing to an automated approach.
Build a prompt list
Write 10–20 questions that mirror real buyer intent in your category. Include category discovery prompts ("best tools for X"), comparison prompts ("X vs Y"), problem-aware prompts ("how do I fix X?"), and branded prompts ("what is [Your Brand]?"). Avoid prompts that already include your brand name — those only measure what ChatGPT thinks of you among people who already know you. The high-value signal comes from prompts where your brand has to earn its place.
Run each prompt in a fresh ChatGPT session
Always start a new conversation for each prompt — do not run multiple prompts in the same thread. ChatGPT incorporates prior context within a conversation, which skews results. Use the same GPT model (e.g. GPT-4o) for every run and record the version. If possible, run in a private browser window to avoid personalisation effects.
Record the results in a spreadsheet
For each prompt, capture: (1) whether your brand was mentioned, (2) what position it appeared in the response (1st, 2nd, etc.), (3) which competitors were mentioned and their positions, (4) any URLs or sources cited, and (5) the exact sentiment — was the mention positive, neutral, or qualified with a caveat. Copy the full response text if you want to track phrasing changes over time.
Repeat weekly against the same prompt list
Consistency matters more than frequency. A weekly cadence on the same 15–20 prompts gives you a time-series view. One-off checks give you noise. The pattern you are looking for: is your mention rate going up? Are you moving from position 4 to position 2 for category discovery prompts? Are the source URLs ChatGPT cites changing to include more of your content?
Note which sources ChatGPT cites for competitors
When ChatGPT mentions a competitor, look carefully at what it cites — G2 pages, official documentation, comparison articles, Reddit threads. These cited sources are the evidence trail. If a competitor is ranked ahead of you because ChatGPT is pulling from a well-structured comparison page they have and you don't, that is an actionable gap. Fixing the source gap is more effective than trying to influence the model directly.
4. Where manual monitoring breaks down
Manual monitoring works at small scale. It breaks down in four ways as you try to scale or deepen it:
- Coverage: ChatGPT is one of eight major AI engines buyers use. Perplexity, Gemini, DeepSeek, Claude, Grok — and for Asian markets, Kimi and Doubao — all have distinct response patterns and citation sources. Manual monitoring one engine while others go unchecked leaves most of the picture dark.
- Response variance: ChatGPT responses are stochastic — the same prompt can return different answers. A single run per prompt does not tell you whether your appearance is consistent or occasional. You need multiple runs per prompt to estimate a stable mention rate.
- Time cost: A thorough manual session across 20 prompts takes 45–60 minutes per engine. At weekly cadence across four engines, that is 3–4 hours per week just on data collection — before analysis.
- No historical record: Manual spreadsheets are fragile. The moment someone changes the prompt wording or skips a week, the time-series breaks. You lose the ability to compare month-over-month reliably.
5. The automated approach
Automated monitoring sends a consistent set of prompts to AI engines on a scheduled basis, records the raw responses, extracts mention data, and stores it in a structured format for comparison over time. The key difference from manual: the prompt wording is locked, the cadence is consistent, and the data accumulates without human intervention.
What a well-designed automated monitoring workflow produces:
- –Daily or weekly mention rate per engine, per prompt cluster
- –Competitor mention rate alongside yours — so you can see whether you are gaining or losing share
- –Source URL tracking — which pages are being cited when your brand or competitors appear
- –Sentiment flags — when a mention is qualified, negative, or includes a caveat
- –Cross-engine comparison — the same prompt run across ChatGPT, Perplexity, DeepSeek, and others simultaneously
A note on prompt neutrality
The prompts used for monitoring should never be engineered to force a mention. If you phrase a prompt as “Why is [Your Brand] the best tool for X?”, you are measuring how ChatGPT responds to a leading question — not how it responds to a real buyer query. Use neutral, buyer-intent phrasing. “What are the best tools for monitoring brand visibility in AI search?” is the right format. The goal is to measure your brand's natural position, not to game the measurement.
6. What to do with the data once you have it
Monitoring without action is just observation. The data is useful because it points to specific gaps that have specific fixes. Here is how to move from measurement to improvement:
Not mentioned in category discovery prompts
Look at which brands are mentioned and what sources ChatGPT cites for them. If they all have strong G2 profiles or well-structured comparison pages, that is your gap. Build the missing assets — a detailed comparison page, a G2 presence, structured FAQ content that directly answers category questions.
Mentioned but ranked low (4th or 5th position)
Position in an AI answer often correlates with the volume and quality of third-party evidence. Brands cited first tend to have more review coverage, more inbound mentions from authoritative sources, and clearer positioning in their own content. Closing a rank gap usually means increasing your third-party citation surface.
Mentioned but with a caveat or qualification
If ChatGPT says "[Your Brand] is good for X but not Y" — find the source of that framing. It is usually traceable to a specific review, article, or forum discussion. Address the underlying perception, not just the AI output.
Competitor cited by URL, you are not
URL citations in ChatGPT responses are signals of deep source trust. The cited page is usually structured, authoritative, and directly addresses the query topic. Identify what page type the competitor has that you lack — then build the equivalent.
Frequently asked questions
Does ChatGPT update its responses as new content is published?
Yes, but not instantly. ChatGPT's knowledge is updated periodically, and the browsing-enabled version (ChatGPT with web search) can pull from live sources. For monitoring purposes, treat responses as reflecting a mix of training data and recent web content — not a real-time index. Changes you make to your content may take weeks to weeks to register in ChatGPT responses.
Do I need to monitor every AI engine, or is ChatGPT enough?
ChatGPT has the largest user base but is not the only engine that matters. Perplexity is heavily used for research queries and cites sources explicitly. Gemini is deeply integrated with Google search. DeepSeek is gaining significant traction globally. For brands targeting Asian markets, Kimi and Doubao have distinct recommendation patterns based on different source ecosystems. Monitoring only ChatGPT gives you a partial picture.
How many prompts do I need to monitor?
Start with 15–20 prompts covering three categories: category discovery (what are the best tools for X), comparison (X vs Y), and problem-aware (how do I solve X). This covers the main AI query patterns for most B2B and B2C categories. You can expand from there once you have a baseline.
Can I monitor brand mentions in ChatGPT without paying for tools?
Yes — the manual method described above requires only a ChatGPT account and a spreadsheet. The tradeoff is time and consistency. Automated tools are worth evaluating once you have validated that AI monitoring is important for your category and you need weekly data across multiple engines.
What is the difference between a brand mention and a citation in ChatGPT?
A mention means ChatGPT named your brand in its response. A citation means ChatGPT included a URL to your site as a source. Citations are higher-value signals — they indicate that ChatGPT has enough trust in a specific page to surface it as evidence for a claim. Most brands are mentioned before they are cited. Being cited requires stronger source authority than simply being named.