How accurate are the AI outputs?
Style and structure: very high. Factual claims: always verify — AI can hallucinate, especially with niche data.
Last updated: 2026-05-04
Reliability by category
| Output type | Reliability | Notes |
|---|---|---|
| Style / tone / structure | 95%+ | The AI excels at this. |
| Grammar / spelling | 99%+ | Proofreading is built in. |
| General concepts (well-known topics) | 90%+ | Verify if you'll publish without editing. |
| Specific facts (dates, names, numbers, prices) | 60-80% | **Always verify.** |
| Live data (news, stock prices, current events) | Variable | Use Live Trends tool for X/Twitter; otherwise assume training-data cutoff. |
| Code | 75-90% | Test before deploying. |
| Math / calculations | 70-85% | Use Calculator tools for anything precise. |
| Citations / source URLs | Low | AI often invents URLs. Always click-verify. |
How we mitigate hallucination
- Tool design: factual tools (Statistics Generator, Citation Finder) cross-check against real sources where possible.
- Disclaimers: all factual outputs include a verification reminder.
- Premium models reduce error rate by 30–50%. Pro users on GPT-4o or Claude Opus see fewer hallucinations than Free on qwen2.5-3b.
- Brave Search integration: some Pro tools augment outputs with live web search to ground facts.
Best practice
Treat AI as a fast first draft. For anything you'll publish, sign, or stake credibility on:
- Verify all named entities (people, companies, products).
- Verify all numbers.
- Spot-check 2–3 claims you don't already know.
- Run a final edit for voice match.
Was this helpful?
Still stuck? Contact support
