AI has become an indispensable “personal assistant” for many business owners. Unfortunately, this assistant often fabricates information. Moreover, research suggests that as AI use becomes habitual, users begin to over-trust the technology. The probability of mistakes escalates, leading to potentially disastrous business consequences.
Here are the most recent stories about SMEs’ overreliance on AI:
- Canadian small businesses are losing money after relying on general-purpose AI for tax, bookkeeping, and financial advice, according to a survey of accountants and bookkeepers commissioned by Dext.
- A Canadian realtor had to apologize after AI hallucinations in listing photos went far beyond acceptable retouching.
- An Australian travel company owned by a married couple faced backlash after AI-generated blogpromoted nonexistent hot springs, triggering what the owner called “the online hate and damage to our business reputation has been absolutely soul-destroying.”
These five strategies will help you drastically reduce the AI Hallucinations rate and the number of critical mistakes.
1. Develop AI Policy
The very first step in preventing AI hallucinations that every organization using AI should take is to document acceptable AI use in a clearly defined AI policy.
At minimum, the AI policy should define which tools are allowed, who can use them, what information is prohibited from being uploaded, and what use cases are permitted.
The AI policy should also specify when human-in-the-loop review of AI outputs is required and define clear accountabilities. It must establish disclosure standards outlining when and how AI use should be communicated.
The AI policy should be communicated both internally and externally, and compliance should be regularly audited to ensure adherence.
For an example of an AI Policy, please visit my website at https://www.nataliabrattan.com/ai-policy/
2. Define Acceptable Vendor’s AI Use
Acceptable AI use is becoming a common clause in contracts to protect businesses from AI hallucinations and errors in vendor deliverables.
For example, if you hire a designer who relies on AI Use to generate visuals, what happens if AI hallucinations produce content that unintentionally copies another brand’s design? Similarly, if an advisor relies on AI Use to draft tax summaries, what if AI hallucinations fabricate information?
Your contracts should clarify allowed and prohibited use of AI, how information is handled and what level of human oversight exists for AI use.
Clear contractual language aligns vendor AI Use with your own AI Policy.
3. Select Tools for AI Use Input Control
Large language models, like ChatGPT and Google Gemini are best suited for tasks where facts do not require verification. Examples of such tasks are ideation, grammar improvements, competitive analysis or title improvements.
When accuracy is critical, organizations may prefer Retrieval-Augmented Generation (RAG) for their AI Use. A RAG model allows AI to generate answers only from verified sources uploaded by the user, rather than drawing from its broad training data.
Studies show that RAG can reduce AI hallucinations by 70–80 percent.
A popular RAG tool is Google NotebookLM, which allows users to anchor AI use to up to 300 sources. Microsoft Copilot Studio is another option, which is particularly well suited for Microsoft-centric environments.
4. Use Grounding Prompting Strategies for AI Use
There are a several prompting strategies which help to minimise AI Hallucinations, which are also called Grounding strategies.
First, reduce complexity of your prompts. Break your requests into small, specific tasks.
Second, add constraints that force traceability. Use instructions like “Cite the source for every claim,” “Do not use outside knowledge,” and “Say ‘unknown’ if the answer is not explicitly supported by the provided sources.” If AI response cannot be traced to the source, ignore it.
Alos, use Chain-of-Thought prompting to improve reasoning quality by requiring an explicit step-by-step explanation before the final output. A practical template is: “Think step by step, then provide a short final answer.” For instance: “Compare the two vendor options in the report. Think step by step, then give a 5-bullet recommendation.”
5. Align AI Use Tasks with Error Tolerance
AI hallucinations often occur when AI tools are pushed beyond their limits. For example, asking AI to interpret tax regulations, generate legal clauses, or calculate financial projections without verification may push its limits. In contrast, using AI to draft social media captions or brainstorm webinar titles is typically within those limits.
In practical terms, organizations should move away from ad-hoc prompting and instead test a small set of clearly defined, low-risk business tasks across different AI tools.
Select the AI tools that perform best within your acceptable risk tolerance. Once validated, formalize the workflow with documented prompts and instructions, approved data sources, and regular checkpoints. Allow the AI to execute the standardized process, and periodically audit outputs to detect performance drift.
Key Takeaways:
1. AI Makes Mistakes
Every AI tool clearly states that errors can occur. For small businesses, these mistakes can damage reputation, mislead customers, create compliance breaches, and result in financial losses.
2. Establish Clear AI Rules
Protect your business by defining formal rules for AI use through a documented AI policy. Add provisions for acceptable AI use to every contract with your vendors.
3. Create Strong Controls
You can significantly lower both the likelihood and impact of AI hallucinations by selecting appropriate tools, limiting tasks to safe use cases, and using structured, grounded prompting techniques.
Your role in staying updated is integral to our shared mission of fostering a community of innovators. CanadianSME Magazine is a valuable treasure trove of entrepreneurial knowledge. Click here to subscribe to our monthly editions for updates on Canadian businesses. Follow our handle, @canadian_sme, on X to stay updated on all business trends and developments. Your support is crucial to our mission.
Disclaimer: This article is based on publicly available information intended only for informational purposes. CanadianSME Small Business Magazine does not endorse or guarantee any products or services mentioned. Readers are advised to conduct their research and due diligence before making business decisions.
