Every AI surface, tracked.
Ten engines. Forty-seven languages. Three regions. One workspace. Coverage that grows the day a new model ships.
A reference, not a marketing list.
Each engine, the cadence we capture it on, the regions where we have panel coverage, and the languages it supports today.
ChatGPT
Response text, citations, ranking position of cited sources, and sentiment toward your brand. Both free and Plus accounts are sampled separately.
MethodologyClaude
Full conversation captures including tool use and artifact mentions. We track how Claude frames your category and which competitors it surfaces first.
MethodologyPerplexity
Citation order matters most here. We log which sources appear in the top three, their dwell weight, and whether your domain is among the linked references.
MethodologyGemini
Captured across Search-integrated Gemini and the standalone app. We separate the two because answer composition differs by surface.
MethodologyGrok
Sentiment shifts faster on Grok than anywhere else. We sample three times per day for prompts flagged as high-volatility by our model.
MethodologyCopilot
Microsoft Copilot in Bing and inside Windows. Edge surface and standalone surface are tracked as separate environments because the prompts behave differently.
MethodologyMeta AI
Captures from WhatsApp, Instagram, and Facebook surfaces. Response style varies meaningfully across the three — we record which app produced each answer.
MethodologyDeepSeek
Lower query volume in our panel, so we batch weekly. Useful for any brand whose buyers skew technical or whose category has strong APAC presence.
MethodologyAI Overviews
Google AI Overviews. We log triggered prompts, the source set, position within the overview, and whether the overview was shown at all for a given query.
MethodologyAI Mode
Google AI Mode is still rolling out. Coverage is US-only and limited to opted-in panelists. Treat numbers as directional until we lift to general availability.
MethodologyHow the capture actually works.
Three deliberate choices that make our numbers different from the AI tracking tools that came before.
Browser capture, not API.
Real responses from opt-in user panels — the same answers a buyer would see. The API returns different content, often with safety rails the consumer surface does not have.
Why browser panels beat APIsDaily refresh.
Every prompt is re-asked at twenty-four-hour intervals. Sentiment drift, citation churn, and ranking changes are detected within a day, not a week.
Refresh cadence methodologyReal consumer prompts.
Our 1.8B prompt dataset comes from opt-in browser panels of actual searches, not synthetic generation. Buyer language, not researcher language.
How we source promptsForty-seven languages. Three regions. Every cell tracked.
- English
- Spanish
- French
- German
- Italian
- Portuguese
- Dutch
- Polish
- Russian
- Ukrainian
- Czech
- Romanian
- Hungarian
- Swedish
- Norwegian
- Danish
- Finnish
- Greek
- Turkish
- Hebrew
- Arabic
- Farsi
- Hindi
- Bengali
- Urdu
- Tamil
- Telugu
- Marathi
- Punjabi
- Indonesian
- Malay
- Thai
- Vietnamese
- Tagalog
- Mandarin
- Cantonese
- Japanese
- Korean
- Swahili
- Zulu
- Afrikaans
- Catalan
- Basque
- Galician
- Croatian
- Bulgarian
- Slovak
| Engine | US | EU | APAC |
|---|---|---|---|
| ChatGPT | |||
| Claude | |||
| Perplexity | |||
| Gemini | |||
| Grok | — | ||
| Copilot | |||
| Meta AI | — | ||
| DeepSeek | — | ||
| AI Overviews | |||
| AI Mode | — | — |
A 24-hour cycle. Every prompt, every engine, every day.
Polling is staggered across the day so traffic never bursts and panels stay fresh. The visual below is a real-time loop.
What we capture, and what we still do not.
An honest line between today and the roadmap. We would rather you choose Clovion knowing the gaps than discover them after onboarding.
- Voice-mode capture for ChatGPT and Gemini
- Yandex and Naver for full APAC coverage
- Mistral Le Chat as a tracked engine
- In-product surfaces (Shopify Sidekick, Notion AI)
- Real-time alerting on sentiment inversions
- Voice queries (audio only) — text transcripts only today
- Enterprise-deployed Copilot in private tenants
- Region-locked engines outside US, EU, APAC
- In-app assistants without public response surfaces
- Image-generation prompts (text answers only)
Two more engines than the nearest tool.
The gap matters because the missing engines are the ones where your buyers are already asking questions.
Coverage, refresh, and the things people ask.
If your engine, language, or region isn't covered, tell us. The roadmap is shaped by who is asking.
We typically have beta coverage within two weeks of a major engine going public, and lift to daily tracking once the response surface stabilizes. AI Mode was beta within nine days of its preview release.
APIs return different content than the consumer-facing chat. Safety rails differ, system prompts differ, and citation behavior often differs. We capture from real browser sessions because that is what your buyers see.
Our 1.8B-prompt dataset is sourced from opt-in browser panels — real searches from real users who have consented to share anonymized query data. We then cluster these against your category to find the prompts that matter to your brand.
It means we are capturing responses but the data is directional, not statistically stable. Either the engine is still rolling out, panel size is small, or response formats are still changing week to week.
Yes. Every report can be filtered to US, EU, or APAC. Cross-region comparison is a default view in the dashboard — useful for brands localizing positioning across markets.
Forty-seven languages today. Coverage varies by engine — ChatGPT and Gemini lead at full coverage, DeepSeek is the lightest at fourteen. The full breakdown is in the table above.
Track every surface from day one.
Ten engines, three regions, forty-seven languages — running before your first standup.