How to Beginner to Intermediate · 3 min read

How to reduce AI hallucinations

Quick answer

To reduce AI hallucinations, use precise prompt engineering with clear context and constraints, select models known for factual accuracy like claude-3-5-sonnet-20241022, and implement output verification with external knowledge sources or retrieval-augmented generation. Combining these approaches minimizes fabricated or incorrect AI responses.

PREREQUISITES

Python 3.8+
OpenAI API key (free tier works)
pip install openai>=1.0

Setup

Install the OpenAI Python SDK and set your API key as an environment variable to securely authenticate requests.

bash

pip install openai>=1.0

Step by step

This example demonstrates reducing hallucinations by providing clear instructions and context to the gpt-4o model, and verifying output with a simple check.

python

import os
from openai import OpenAI

client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])

prompt = (
    "You are a helpful assistant. Answer factually and cite sources if possible. "
    "If you don't know, say 'I don't know'."
    "\nQuestion: What is the capital of France?"
)

response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": prompt}]
)

answer = response.choices[0].message.content
print("AI answer:", answer)

# Simple verification example
if "Paris" in answer:
    print("Verified: Correct capital detected.")
else:
    print("Warning: Possible hallucination detected.")

output

AI answer: The capital of France is Paris.
Verified: Correct capital detected.

Common variations

You can reduce hallucinations further by:

Using retrieval-augmented generation (RAG) to provide up-to-date documents as context.
Choosing models with stronger factual grounding like claude-3-5-sonnet-20241022.
Applying temperature=0 or low values to reduce randomness.
Implementing multi-step prompting to verify or cross-check answers.

python

import os
import anthropic

client = anthropic.Anthropic(api_key=os.environ["ANTHROPIC_API_KEY"])

system_prompt = "You are a helpful assistant that only answers with verified facts."
user_prompt = "What is the capital of France?"

message = client.messages.create(
    model="claude-3-5-sonnet-20241022",
    max_tokens=100,
    system=system_prompt,
    messages=[{"role": "user", "content": user_prompt}]
)

print("Claude answer:", message.content[0].text)

output

Claude answer: The capital of France is Paris.

Troubleshooting

If you see inconsistent or fabricated answers, try:

Lowering the temperature parameter to 0 or 0.2.
Adding explicit instructions to refuse to answer if unsure.
Using external knowledge bases or APIs to fact-check outputs.
Splitting complex queries into smaller, verifiable steps.

✅

Key Takeaways

Use clear, explicit prompts instructing the model to avoid guessing and cite sources.
Select models with strong factual accuracy and tune parameters like temperature to reduce randomness.
Incorporate external verification or retrieval-augmented generation to ground answers in real data.

Verified 2026-04 · gpt-4o, claude-3-5-sonnet-20241022

Verify ↗