Hallucination
A hallucination is a language model stating information that does not exist or is wrong as if it were true, in a confident voice. Invented sources, wrong figures and non-existent product features are typical. Giving the model verified sources, asking it to cite them and passing critical outputs through human approval reduce the risk.
- In Turkish
- Halüsinasyon
- Read more · Service
- AI customer service assistant
Why it matters
A hallucination means the AI writes a wrong answer as fluently as a right one, so it is hard to notice. A wrong price sent to a customer, a campaign that does not exist or an invented source can cause reputation and legal problems. Critical outputs must always be checked.
Example
Say a firm’s chat assistant told a customer “we give a 2-year warranty on this product”, when the warranty was 1 year. The customer complained. The assistant was told to answer only from the warranty document, and warranty questions were put under human approval.
Related terms
- Human-in-the-loopHuman-in-the-loop means leaving the decision or final check at a particular step of an automated process to a person.
- RAGRAG is a method in which a language model, before answering, retrieves relevant passages from a business’s own documents and takes them as context.
- AI agentAn AI agent is a system built on a language model that picks its own steps towards a goal and uses tools to take them.
- APIAn API is the interface a piece of software offers to other software, defining which requests it answers and in what form.
- Error workflowAn error workflow is a separate workflow that runs when another one fails.
- IntegrationAn integration is a connection that lets two or more pieces of software pass data to each other without manual transfer: a site and a CRM, a shop and accounting, a form and an email tool.
The definitions are general information. If you think one is incomplete or wrong, write to us and we will reply within one working day. Get in touch
From our journal
From what we have written on Automation and AI; they show how the term turns up in the work.
- How to write prompts for work: 5 parts, an example, a checkThe five parts from role and context to example, a before and after on meeting notes, how to check the answer, what never goes into a public tool, and a shared prompt library for the team.
- What is RAG? AI that answers from your documentsMaking a language model search your own documents before it writes an answer: the parts of the system, where it goes wrong, testing with a fixed question set and personal data points.
- e-Fatura in North Cyprus: who must switch, how to applyWhat the Income and Tax Office’s pages and the Regulation it publishes say: paper invoices and e-Fatura stand side by side, who the obligation covers, the steps of the application, and what a business selling online has to connect.
- How to clean customer data before a CRM importTaking a copy, one row per person and a fixed set of columns, one format, duplicates, empty values, source and permission, a twenty-row test and the rules that keep the list clean.