How do I test what AI says about my business without my own account skewing the results?
TL;DR: Use a logged-out or fresh session, phrase the question the way a real customer would (not "tell me about [my business]"), and repeat it across ChatGPT, Claude, Gemini, and Perplexity. Testing from your own logged-in, personalized account can return skewed results, since some AI assistants factor in account history and location settings.
The claim
Self-testing bias is a real methodology problem — the fix is neutral, repeated, customer-phrased prompts, not asking about yourself directly.
Testing method comparison
| Testing method | Bias risk | Fix |
|---|---|---|
| Logged into your personal AI account | High — history/location can skew results | Use a logged-out or fresh session where possible |
| Asking "tell me about [my business]" | High — this isn't how customers ask | Ask the way a customer would: "best X near me" |
| One-time test, one engine | Low reliability | Repeat weekly, across all 4 engines |
| Testing without specifying location | Moderate | Specify the customer's likely location explicitly in the prompt |
Step-by-step
- Use a fresh or logged-out session when possible.
- Phrase the question exactly as a customer would, not as a business owner checking on themselves.
- Specify a location the way a real customer would ("in [city]" or "near [neighborhood]").
- Run the same phrasing across ChatGPT, Claude, Gemini, and Perplexity.
- Repeat on a schedule so one unusual result doesn't skew your read on the trend.
FAQ
Does my location or IP address affect results? It can, especially for engines with location-aware retrieval — always specify location explicitly in the prompt rather than relying on it being inferred correctly.
Is there a way to fully eliminate bias? Not perfectly, but neutral phrasing, fresh sessions, and repeated testing across engines gets you close enough to spot real trends.
Should I trust one test result? No — treat any single test as a data point, not a verdict; the trend across weeks and engines is what's meaningful.
By Pinal Dave Last updated: 2026-07-24