Zero Fabricated Answers.
158 Questions Tested.
We ran 158 questions against the live Impact AI bot—exact matches, reworded questions, typos, pricing questions, and out-of-scope traps. Here's what happened: 100% accuracy on known questions, zero fabricated facts, and a 1.5-second median response time.
Fabricated answers
Accuracy on known questions
Median response time
Audit Results by Category
158 questions across five categories. Each bar shows the percentage of correct responses. The "out-of-scope" category measures how often the bot correctly refused to answer a question it had no verified answer for.
What "fabricated" means
A fabricated answer is one where the bot invented a fact—a price, a name, a policy, or a credential—that was not in the verified Q&A data. The audit found zero.
The 10 "flags" were grader artifacts
The automated grader flagged 10 responses as potential fabrications. Manual review showed every one was the bot giving more verified detail than the short reference answer, or offering the real website after declining. Zero real fabrications.
80% needed no AI model
Four out of five questions were answered by the deterministic matching engine—a database lookup against your verified Q&A pairs. The AI model only fired for the hardest 20%, and even then it was grounded in your facts.
Response Time vs. General-Purpose AI
Median response time per question. Impact AI resolves 80% of queries with a database lookup—no AI model needed. General-purpose chatbots generate every answer from scratch.
Why Impact AI is faster
- 80% of queries hit a database lookup—no model call at all
- Hybrid scoring (semantic + lexical) runs in milliseconds
- The AI model only fires for the hardest 20%, grounded in your facts
Why ChatGPT is slower
- Generates every answer from scratch—no lookup
- Large model needs to process the full prompt each time
- No access to your verified business data unless you build it
Why reasoning models are slowest
- Explicit step-by-step reasoning before every answer
- 24+ seconds makes it unusable for live customer chat
- Burns 20–33+ Watt-hours per query—200x more energy
Calculate Your ROI
Drag the sliders to match your business. The calculator uses the audit's 80% containment rate—the share of questions Impact AI answers without a human.
Questions handled without a human
4,000/mo
Support hours saved
333 hrs/mo
Money saved (after plan cost)
$99,900/yr
ROI: 28,607% (287x return)
Impact AI plan cost
$29/mo (Professional)
Calculations use the audited 80% containment rate. Actual results depend on your training data quality. Enterprise plan ($99/mo) supports 10,000+ interactions/month.
The Environmental Cost
Every AI query burns electricity and water for cooling. Deterministic answers are a database lookup. Here's what that means at scale.
At scale: 1,000 customer deployments on Impact AI save roughly 11 million litres of water per year compared to self-hosted GPUs—and avoid the energy equivalent of powering hundreds of homes. The 80% deterministic containment rate means 80% of queries have the carbon footprint of a database read, not a neural network inference.
Ready to put this to work?
Start free. Build your bot from your own FAQs. See the same zero-fabrication, sub-2-second performance on your data.
No credit card required for Starter. Professional $29/mo, Enterprise $99/mo.