How to Check If Your Brand Appears in ChatGPT, Perplexity, and Gemini
AI answers change between runs, between sessions, and between platforms -- yet every guide tells you to ask ChatGPT once and count. A single check is a coin flip. This is the sampling method: a defined prompt set across ChatGPT, Perplexity, and Gemini, run in a fresh session N times, scored on mention rate, share of voice, and citation sources.
The fastest answer to "how do I check if AI mentions my brand": run a defined prompt set in a fresh, logged-out session across ChatGPT, Perplexity, and Gemini at least five times per prompt, then log mention rate, share of voice, and the URLs each platform cites. Anything less is guessing.
That isn't the answer most "check your AI visibility" articles give. The dominant template is: ask ChatGPT five questions, count whether your brand shows up, declare yourself visible or not. With ChatGPT at 900 million weekly active users and Google AI Overviews now appearing on a substantial share of queries (BrightEdge's one-year AIO data), it's tempting to want a fast answer. A single check produces a false reading -- AI answers change between runs, between sessions, between platforms, between locations, and between whether the model is answering from memory or searching live.
A single AI check is a coin flip. Measure the distribution, not the answer.
When a brand runs its first check across five AI models, the result usually swings between runs -- sometimes the brand appears, sometimes a namesake takes the spot, sometimes the answer drifts into an adjacent category entirely. That isn't measurement noise to suppress; that's the actual signal. This post is the sampling method that turns that variance into a stable reading: why one check lies, the three metrics that actually matter, the manual procedure done right, per-platform differences, and when to stop checking by hand.
Why a single check lies
Five failure modes turn a single-shot check into a false reading. Each one is independent -- fix one and the other four still distort your measurement.
Non-determinism
ChatGPT, Perplexity, and Gemini sample tokens probabilistically. The same prompt produces different outputs run to run, even with identical phrasing and no other context changes. Your brand might appear in three of five runs, never in the fourth, and twice with the wrong category attached. Mention rate is the mathematically honest summary of that distribution; a single check is one draw from it.
Personalization and memory
Your logged-in ChatGPT account carries chat history, saved memory, and (where enabled) location context that customers don't have. The version of "what is the best CRM for SaaS" that ChatGPT answers for you is not the version it answers for a customer in a different city with no prior interactions with your domain. Always check in a fresh, logged-out session: incognito window, cleared cookies, clean slate.
Location bias
Even in a logged-out session, the platform still infers location from your IP. Gemini in particular reweights toward local entities and Google-owned regional properties; ChatGPT and Perplexity shift answers for region-relevant queries ("best CRM in Australia" returns a different brand set than "best CRM in the US"). If your audience spans multiple regions, the brand check has to run from each one. A US-IP check tells you nothing about how the brand surfaces in Germany or Singapore.
The two-ChatGPTs problem
ChatGPT in default chat mode answers from parametric memory -- the patterns the model learned during training. ChatGPT with browsing on or in search-grounded mode pulls live results. The two return materially different answers, especially for recent brands or fast-moving categories. Test both. The mechanics live in our how to get cited by ChatGPT spoke.
Cross-platform divergence
A brand can be highly visible in ChatGPT and entirely absent from Gemini -- different platforms read different sources. ChatGPT leans on Wikipedia plus curated training data; Perplexity leans heavily on Reddit, comparison content, and editorial publications; Gemini is anchored on Google's Knowledge Graph and Google-owned properties like YouTube. A single-platform check tells you nothing about the other four.
The three metrics that actually measure visibility
Most brand-monitoring guides default to a single number -- "we appear in X% of answers" -- and stop there. That's not visibility; it's a partial reading. Three metrics together give the full picture, and one of them is the one most guides skip entirely.
The three metrics that actually measure AI visibility
Mention rate, share of voice, and citation sources are the primary metrics. Sentiment is supporting -- worth tracking, not worth optimizing for in isolation.
| Metric | What it measures | How to compute | Notes |
|---|---|---|---|
| Mention rate | How often the platform mentions your brand at all in response to a defined prompt set. | (# of runs where the brand appears) / (total runs across the prompt set). | Sensitive to non-determinism -- always compute over N>=5 runs per prompt, not a single check. |
| Share of voice | Your brand's mention frequency relative to named competitors on the same prompts. | (your mentions) / (your mentions + sum of competitor mentions) across the prompt set. | The number that actually maps to category position. Track over time, not as a one-shot reading. |
| Citation sources | Which URLs the platform cited when answering -- your own domain, third-party reviews, Wikipedia, Reddit, etc. | Log every cited URL. Group by domain. Compute per-platform source-mix percentages. | The metric most guides skip. Tells you whether you need on-site work, off-site work, or both. |
| Sentiment (supporting) | Whether mentions are positive, neutral, or negative. | Classify each mention (manual or LLM-assisted). Track the share that's negative or critical. | Supporting, not primary. Mention rate and source mix change the strategy; sentiment changes the prose, not the playbook. |
The metric most guides skip is the third one: citation sources. The Hashmeta 20,000-page study showed that mention and citation behave as different signals -- pages can be mentioned without being cited, and cited from third-party sources without ever surfacing your own URL. Princeton's GEO paper corroborates this from the optimization side: Citation Addition was a top-3 lever out of nine tested. Logging which domain each platform actually cited per run tells you whether you need on-site work (your domain isn't the source) or off-site work (your domain is the source but the answer is shallow).
The manual method, done right
Five steps. Each one is mechanical; the discipline is doing all five before drawing any conclusion.
1. Build a defined prompt set
Mix branded and unbranded prompts. Branded prompts ask about you by name; unbranded prompts ask about your category, problem, or use case. Both matter. Aim for 10-20 prompts. Example seed prompts for a hypothetical CRM brand:
Branded: - "What is [Brand]?" - "Is [Brand] good for [use case]?" - "[Brand] vs [top competitor]" Unbranded: - "Best CRM for early-stage SaaS" - "Open-source alternatives to [category leader]" - "How do I track AI citations for my brand?"
2. Use a fresh, logged-out session
Open an incognito or private browser window. Don't log into your work account. Don't use a session that has any prior history with your brand. If location matters for your category, change the browser's location signal or use a VPN -- Gemini and ChatGPT both shift answers by region.
3. Run each prompt at least five times
N=5 is the floor. N=10 is better. Between runs, start a new chat or clear the session so the previous answer doesn't influence the next. Record the full answer text and every cited URL, not just whether your brand appeared. Citation sources are the metric you can't reconstruct after the fact.
4. Repeat across platforms
ChatGPT (default + search-grounded), Perplexity, Gemini. Add Claude and Grok if those surfaces matter to your audience. Each platform is an independent measurement; do not average across platforms -- your visibility on ChatGPT is a different number from your visibility on Gemini.
5. Compute the three metrics
Per platform: mention rate (% of runs that mention you), share of voice (your mentions divided by yours + named competitors), and citation source mix (count distinct domains, group by relationship -- your own, competitor, third-party review, Reddit, Wikipedia). Build a row per platform and track the same three numbers weekly to see whether work you ship moves them. A workable logging schema:
| Platform | Prompt | Run# | Mentioned | Competitors mentioned | Cited URLs |
|---|---|---|---|---|---|
| ChatGPT | "Best CRM for SaaS" | 1 | Y | Vendor A, Vendor B | wikipedia.org, reddit.com/r/saas, your-domain.com |
| ChatGPT | "Best CRM for SaaS" | 2 | N | Vendor A, Vendor C | techcrunch.com, hubspot.com, vendor-a.com |
| ... | ... | ... | ... | ... | ... |
A Google Sheet with one row per run is enough to start. Pivot on platform + prompt to get the three metrics; pivot on cited URL to get source mix. Resist the urge to aggregate across platforms -- a 60% mention rate on ChatGPT and 10% on Gemini is two different numbers, not one average.
Per-platform differences worth knowing before you run
Each platform draws from a different mix of sources and disambiguates entities differently. A high mention rate on one tells you nothing about another. The table below summarises what to look at on each surface.
Per-platform source bias: what to look at on each surface
Each platform pulls from a different mix of sources. A single result on ChatGPT tells you almost nothing about your visibility on Perplexity or Gemini -- they are independent measurement surfaces.
| Platform | Primary sources | What to check | Watch out for |
|---|---|---|---|
| ChatGPT | Wikipedia, curated training data, plus live web search when search-grounded. | Run both the parametric mode (default chat) and the search-grounded mode (browsing on). Answers diverge. | Personalization and memory in your logged-in account. Always check in a fresh / logged-out session. |
| Perplexity | Editorial publications, Wikipedia, Reddit, comparison content, podcast transcripts. | High citation density per answer -- log every cited source URL, not just whether you're mentioned. | Reddit and community-thread bias. Your competitor may be winning because a Reddit thread mentions them, not their site. |
| Gemini | Google's index, Knowledge Graph entities, YouTube, Google-owned properties. | Whether AI Overviews triggers for your target queries, and whether your brand appears in the panel. | Entity-anchored answers default to the Knowledge Graph match. If your namesake has a stronger entity, you may be invisible regardless of content quality. |
Editorial synthesis based on per-platform crawler architecture and observed citation patterns across the five tracked AI platforms.
The structural pattern is consistent: Wikipedia is the most universally weighted source across all five platforms; Knowledge Graph entities are Gemini-specific; Reddit is Perplexity's heaviest community-source weight. Our cross-platform hub walks each surface in more depth.
When to stop checking manually
The honest math on the manual method: 20 prompts × 5 runs × 5 platforms × 2 locations × weekly cadence equals 2,000 runs per cycle. That's roughly one full-time week per cycle, with no time left to analyse the results. The manual method is for getting started, validating your prompt set, and building intuition for what good looks like -- not for ongoing measurement.
Once you've run the manual method once across the platforms that matter, you have two questions answered: which prompts give signal, and which platforms matter for your category. The next question -- "does this number change when we ship work?" -- is where automation pays back. Tooling runs the same prompt set on a schedule, logs every answer and every citation source, and surfaces trend lines without burning a week per cycle. If your manual cadence exceeds 200 runs per week, you're past the scale wall.
The honest summary
Manual spot-checks lie. They lie because of non-determinism, because of personalization, because parametric and search-grounded modes give different answers, and because what's visible on ChatGPT may be invisible on Gemini. The fix is sampling discipline: a defined prompt set, a fresh session, N runs, multiple platforms, and the three metrics -- mention rate, share of voice, and citation sources -- tracked together.
For most teams, the manual method is a one-time validation pass. After that, the scale wall arrives quickly and tooling takes over. The work the tooling can't do is choosing the prompt set, defining the competitors that count, and deciding which platforms matter. That stays human.
Run the sampling method without burning a week per cycle
Our free AI Visibility Checker runs a defined prompt set across ChatGPT, Perplexity, and Gemini and returns visibility and share-of-voice in minutes -- no logged-in skew, no single-check coin flips. For background on the broader category, see our GEO primer.
Check your AI visibility freeWritten by
Nisha Kumari
Nisha Kumari is Co-Founder at Ranqo, where she leads growth strategy and client acquisition. With a background in digital marketing and financial management, she specializes in SEO, Generative Engine Optimization, and helping brands build visibility across AI platforms.
Share this article
Related articles
How to Get Cited by ChatGPT: The Two-ChatGPTs Playbook
There are two ChatGPTs. The parametric model answers from a frozen training corpus; the search-grounded model fetches via Bing. 87% of SearchGPT citations match Bing's top results (Seer, Feb 2025), which means most 'ChatGPT SEO' advice is actually Bing SEO advice. This deep dive covers the two-mode duality, the three-bot crawler architecture (GPTBot vs ChatGPT-User vs OAI-SearchBot), citation conservatism (7.92 vs Perplexity's 21.87), the OpenAI licensing layer, training-cutoff timing, and the 12-question ChatGPT citation checklist.
The Anti-GEO Playbook: 15 Mistakes That Make AI Skip Your Brand
Most GEO advice tells you what to do. This playbook tells you what to stop doing. 15 verified, data-backed mistakes that make AI platforms skip your brand -- ranked by severity, with the exact citation impact of each one. Every statistic is sourced from a published study.
How to Get Cited by Google AI Overviews & Gemini: Three Surfaces, Three Citation Games
Google has three AI surfaces, not one. AI Overviews appeared on 11%+ of Google queries one year after launch (BrightEdge May 2025), AI Mode rolled out in the US on May 20, 2025, and the Gemini app sits at 15.1% of AI chatbot market share (First Page Sage April 2026). Each rewards different work -- and AI Overviews is the only surface in the entire AI ecosystem where traditional SEO directly carries over. The third spoke of the platform-cluster strategy.