# What is an edge case in AI testing? > An edge case is an unusual input at the limits of what a system expects: an empty field, an odd question, a missing record. Test them before customers do. Source: https://learn.ynteractive.com/content/glossary/edge-case · Updated 2026-10-21 · Free from Stepthrough (https://learn.ynteractive.com) [AI basics](https://learn.ynteractive.com/content/glossary#topic-ai-basics) · [AI glossary](https://learn.ynteractive.com/content/glossary) Quick answer An edge case is an unusual input or situation at the limits of what a system was built for: a question the FAQ doesn't cover, an empty field, a record in an odd state. Most tests use typical inputs, so edge cases are where AI tools and workflows most often fail, and where you should test on purpose. Last updated October 21, 2026 [Take the free module →](https://learn.ynteractive.com/courses/ai-apps-worth-knowing/36)Module 36 of the free [AI Apps Worth Knowing](https://learn.ynteractive.com/courses/ai-apps-worth-knowing) course, about 15 minutes. No account needed. Updated October 21, 2026 · By [Robert Breen](https://learn.ynteractive.com/content/about) ## Why it matters for a small business AI tools are tuned on common cases and do well on them. Problems hide at the edges: the customer who asks something off-script, the invoice with zero line items, the order that was cancelled after it shipped. On those inputs a chatbot may guess, and a workflow may produce something that looks normal but is wrong. A good test set always includes a few edge cases next to the ordinary ones. List the strange things that have really happened, plus questions the AI should not be able to answer. The goal isn't that the AI knows everything, but that it fails safely: it says it doesn't know, hands off to a person or stops with a clear message. ## In a real lesson: Chatbase: Answer Customer FAQs with an AI Agent In [Chatbase: Answer Customer FAQs with an AI Agent](https://learn.ynteractive.com/content/chatbase-ai-agent-answer-customer-faqs), Dana's agent is trained on Juniper Lane's 22-question FAQ. After a normal test question about Saturday hours, you ask one the FAQ doesn't cover: "Do you work with nonprofits?" Thanks to Guardrail 4, the agent doesn't guess. It says it isn't sure and gives the email and phone number for a person. That is an edge case tested on purpose. Later, the chat logs show that answer with a low confidence score of 0.369, and because Dana does take nonprofits, you use Revise answer to add a Q&A entry. Asked again, the agent answers correctly. In the [Kosuke lesson](https://learn.ynteractive.com/content/kosuke-automated-qa-test-ai-changes), a different edge shows up: a test is blocked because the preview's database has no processing order to move to Shipped. [Try this lesson free](https://learn.ynteractive.com/modules/chatbase-faq-support-agent) or [read the step-by-step guide](https://learn.ynteractive.com/content/chatbase-ai-agent-answer-customer-faqs). ## Common confusions ### Edge case vs bug An edge case is an unusual input. A bug is the wrong behavior it can expose. Many edge cases are handled fine; testing tells you which aren't. ### Edge case vs red teaming Red teaming deliberately attacks a system to make it misbehave. Edge case testing checks honest but unusual inputs. ## Tips - Write down every strange real case you meet and add it to your test set. - Always test one question the AI should not be able to answer. - Check what happens with empty, missing or duplicate data, not only with clean records. ## Related terms [Chatbot fallback response](https://learn.ynteractive.com/content/glossary/chatbot-fallback-response) · [Golden dataset](https://learn.ynteractive.com/content/glossary/golden-dataset) · [Test data](https://learn.ynteractive.com/content/glossary/test-data) · [AI red teaming](https://learn.ynteractive.com/content/glossary/red-teaming) · [Guardrails](https://learn.ynteractive.com/content/glossary/guardrails) · [Prompt testing](https://learn.ynteractive.com/content/glossary/prompt-testing) ## More AI basics terms [Embeddings](https://learn.ynteractive.com/content/glossary/embeddings) · [Explainable AI (XAI)](https://learn.ynteractive.com/content/glossary/explainable-ai) · [False positive and false negative](https://learn.ynteractive.com/content/glossary/false-positive) · [Fine-tuning](https://learn.ynteractive.com/content/glossary/fine-tuning) ## Where you use it: free lessons - [Chatbase: Answer Customer FAQs with an AI Agent](https://learn.ynteractive.com/content/chatbase-ai-agent-answer-customer-faqs) (Chatbase, 15 min) - [Kosuke: Let the AI Test Its Own Changes with Automated QA](https://learn.ynteractive.com/content/kosuke-automated-qa-test-ai-changes) (Kosuke, 12 min) - [Truquence: Catch an n8n Run That Says Success but Gets It Wrong](https://learn.ynteractive.com/content/truquence-n8n-workflow-success-but-wrong-output) (Truquence, 11 min) ## Frequently asked questions **How many edge cases should I test?** A handful per task is a good start: one missing-data case, one off-topic question and one or two odd situations from your real work. **What should an AI do on an edge case?** Fail safely: say it doesn't know, ask for the missing detail or hand off to a person, rather than guess. **Do tools flag edge cases for me?** Some do. Truquence returns UNKNOWN for workflows with untaken branches or binary items, so a person looks instead of the run passing quietly. [All AI glossary terms, A to Z](https://learn.ynteractive.com/content/glossary) · [Free prompt templates](https://learn.ynteractive.com/content/prompts)