Our Methodology
How We Collect Data
Ataiva uses official APIs from OpenAI, Anthropic, Google, and Perplexity. Each prompt in your tracking set is sent to each AI engine, and the full response is stored. We parse every response for brand mentions, sentiment, citations, and competitor references.
API vs Real User Experience
API responses bypass personalization, memory, and browsing plugins. Our data shows what AI engines say in a neutral context. Real users may see different results based on their history and location.
This is a deliberate trade-off: we sacrifice personalization for reproducibility. Every data point we show you can be independently verified.
Why API Testing Is More Accurate Than Browser Testing
Many AI visibility tools test in incognito browser mode. Research has shown this is fundamentally flawed: incognito mode on ChatGPT has no web search capability, while 100% of real users (both free and paid) have full web search enabled.
Ataiva uses official APIs with web search capabilities matching what real logged-in users experience. This means our citation data reflects actual AI behavior, not a stripped-down guest mode that no real user ever sees.
| Method | Web Search | Matches Real Users | Used By |
|---|---|---|---|
| Incognito browser | ❌ Disabled | ❌ No (0% of users) | Most competitors |
| Logged-in browser | ✅ Enabled | ✅ Yes, but adds personal bias | Few tools (expensive) |
| Official API (Ataiva) | ✅ Enabled | ✅ Yes, neutral context | Ataiva |
Source: Discovered Labs research, 2025
How We Handle Variance
AI responses are non-deterministic — the same prompt can produce different answers each time. We run prompts multiple times and report the frequency of mentions, not a single snapshot.
When we say your mention rate is 73%, it means AI mentioned you in 73 out of 100 queries. This frequency-based approach smooths out the inherent variance in AI outputs.
Scoring Algorithm
Your composite visibility score is calculated from four weighted components:
- Mention Rate (40%) — How often AI mentions your brand across all prompts and engines
- Sentiment (25%) — The tone of mentions: positive, neutral, negative, or mixed
- Citation Coverage (20%) — How often AI cites your domain as a source
- Engine Breadth (15%) — Whether you appear across multiple AI engines, not just one
Update Frequency
- Free plan: Weekly scans
- Pro plan: Daily scans
- Business plan: Real-time monitoring
What We Don't Do
- We don't scrape browser sessions
- We don't use undocumented APIs
- We don't fake data
- We don't guarantee that improving your score will increase traffic — the AI referral channel is too new for proven attribution
Why This Still Matters
Even with variance, consistent monitoring reveals meaningful patterns. A brand mentioned 70% of the time is objectively more visible than one mentioned 20% of the time. Trends over weeks and months show whether your content strategy is working — or whether competitors are pulling ahead.
AI-driven discovery is growing fast. The brands that track and optimize for it now will have a compounding advantage.
Related: AI Search vs Google — The Shift