Gains:
- Ability to design the advisory bot responsibly so that it only connects to verified corporate information and provides facts by citing the source.
- Ability to direct medical, legal and sensitive questions to experts and reliable sources, avoiding advice, and transfer them to humans when necessary
- Protecting the privacy of chat data and being transparent to the user about the use of AI
The place where the library meets people is the information desk. A student searches for sources for his homework, a citizen asks how to access an official document, a researcher wants to learn how to use a database. This is called reference service: it is the job of understanding the user's information needs and delivering it to the right source and the right method. In this unit, you will learn how to use artificial intelligence-based chatbots (software that helps by corresponding with the user) and AI-powered counseling responsibly.
One definition: a reference interview is a dialogue in which the librarian asks questions to clarify the user's true need. Because most of the time the user cannot fully express what he is looking for; Someone who says "something about history" may actually be looking for a specific aspect of a specific event. This human, empathetic, and judgmental work is at the heart of librarianship, and AI cannot take it over entirely.
Step by step: AI-powered consultation
1. Clarify the need. Whether it's a human or a bot, good consultation first understands the need. The AI can suggest clarifying questions to ask the user; but it is human's job to read the user's context and emotion.
2. Refer to the source, don't make up the answer. The library bot's job is to direct the user to real resources; It is not about generating answers from general knowledge. The bot should say "look at this database, this collection for this topic"; Must give factual claims by citing the source.
3. Set the limit. The bot should not provide medical, legal or financial advice; instead, it should refer you to reliable sources and, when necessary, an expert. Serious harm would result if the AI gave the wrong medical or legal “answer” to a user.
4. Delegate to humans. When the bot can't figure it out or the topic is sensitive, it should direct the user to a real librarian. A good system knows the bot's limits.
5. Be transparent. The user must know that he is talking to a bot. The library must not mislead the user with a hidden bot impersonating a human.
Tip: Set up an advisory bot as a resource citing system based on your collection and FAQ (frequently asked questions) information. Having the bot as an assistant that "finds you the right resource" rather than "knowing everything" preserves both trust and accuracy.
Chatbot limits and risks
A library chatbot looks appealing: works 24/7, helps multiple users at once, answers simple questions quickly. But the risks are real. The bot may hallucinate and tell the user a non-existent resource, an incorrect boot time, or a faulty procedure. The bot may mishandle the user's sensitive question (a health concern, a personal crisis). And if the bot records what the user types, there is a privacy risk.
Therefore, the bot is positioned not as a replacement for the human advisor, but as a layer that eases his burden. Simple, repetitive questions are received by the bot; Humans handle complex, sensitive and judgment-requiring questions.
Caution: The questions a user types into the chatbot may be sensitive (health, legal, personal situations). The record of this correspondence should be as protected for privacy as the loan records. Providing chat data to general AI tools for training violates user trust.
three mini cases
Case 1 — The boat lightened the load. A university library built a bot based on collection and FAQ information that answers common questions such as "opening hours, loan period, database access". The bot handled approximately 40% of the questions coming to the information desk; librarians were able to devote more time to complex research questions. The bot always linked to the official page in its answers.
Case 2 — Wrong time corrected. A bot said "the library is closed on Sundays" from its global memory; whereas library markets were open. A user would be wasted because of this. The team reconfigured the bot to only connect to the organization's current, verified information; Disabled the response from global memory.
Case 3 — Transfer to human occurred. A user asked the bot about a personal health concern. Instead of providing medical advice, the bot directed users to reliable health information sources and offered the option to hand over to the human, saying "you can talk to a librarian if you want." Sensitive topic handled correctly.
Accessibility and inclusivity
Good user service serves everyone; this is a direct extension of the library's equal access value. AI tools carry both opportunity and risk here. Opportunity side: AI can translate complex text into plain language, generate responses in different languages, make a long document readable for a visually impaired user, or facilitate interaction with a voice assistant. This brings in users who were previously excluded from the service. Risk side: AI tools often work better in certain languages and with certain user profiles; In a low-resource language or at a different literacy level, response quality may decline, reinforcing inequality. The librarian should observe in which user groups the bot or AI tool works well and in which it works poorly, and engage human support where it is weak. The output of a simplification or translation, like any other factual response, should be reviewed to ensure that it does not distort meaning; False simplification can be misleading, especially in sensitive texts such as health, law or government procedure.
Caution: When simplifying or translating a text, the AI may unintentionally change the meaning or omit an important warning. Output produced for accessibility purposes should also be audited for accuracy; While saying "it should reach everyone", care should be taken not to spread false information.
Four copyable templates
1) Counseling interview questions:
Suggest 3-4 polite, open-ended questions to ask to clarify the following vague user question (scope, level, purpose, type).Do not assume user intent. Question: [here]
2) Bot response directing to the source:
Your role: library information bot. Below is the user question and our list of resources/guides. Based solely on this list, write a short, polite response that directs the user to the appropriate resource. When making a factual claim, state the source; If it is not on the list, say "let me direct you to a librarian". Question: [here] / List: [here]
3) Limit and turnover response:
Does the following user question contain medical/legal/financial advice or more sensitive personal topics? If so, draft a response that politely directs the user to the trusted source and an expert/librarian, WITHOUT giving advice. Question: [here]
4) FAQ answer fact check:
Is every factual information (time, duration, procedure, source) in the bot response below based on the official FAQ/information text I provided? Mark each claim as "present / not present in the source". Answer: [here] / FAQ: [here]
Weak prompt / Strong prompt
Weak prompt:
Answer this user's question.
The bot generates answers from its general memory; may give false time, false source or faulty procedure; It does not respect borders and periods.
Powerful prompt:
Your role: library information bot. Rely solely on the current institutional information and resource list I have provided below; Do not provide information other than this. If medical/legal advice is requested, refer to a specialist. Keep the answer short and polite, state the source for the fact, and if there is no information, say "let me direct you to the librarian." User question: [here] / Organization information: [here]
Powerful prompt; It connects the bot to verified enterprise information, sets limits, and defines the transfer path.
Advisory question type table
Question type
Is the boot suitable?
Approach
Opening time, duration, procedure
Yes
Response from current institution information
Simple source routing
Yes
Link to collection/contact
complex research
partially
Clarify the need, transfer it to people
Medical/legal/sensitive
no
Direct to an expert and reliable source
Common mistakes
- Leaving the bot to respond from global memory. Risk of misinformation and fabricated sources.
- Giving a factual answer without citing sources. The user cannot verify.
- Providing medical/legal advice. Causes serious harm; guidance should be given.
- Not establishing the path of transfer to humans. Complex and sensitive questions remain unresolved.
- Leaving chat data unprotected. There would be a violation of privacy.
In summary
In user services, AI and chatbots take over simple, repetitive questions, allowing the librarian to focus on complex, sensitive, and judgment-requiring tasks. But the bot should be built as a layer that eases the burden, not in place of the human: it should rely only on verified corporate information, provide facts as sources, avoid medical/legal advice and refer to a specialist, delegate sensitive questions to the human, and protect the privacy of chat data. Transparency is essential: the user must know that they are talking to a bot.
Application task
Prepare a short "list of current information + resources" for your institution (hours, loan rules, a few guides). Generate answers to three different user questions with the "Bot response to source" template and check whether each fact is based on the source with "FAQ response fact check". Finally, create the correct prompt text for a sensitive question with the “Limit and turnover response” template.
checklist
- [ ] I only connected the bot to verified corporate information.
- [ ] I cited sources in factual answers.
- [ ] I have set up routing and handoffs for medical/legal/sensitive questions.
- [ ] I transparently stated that the user was talking to a bot.
- [ ] I have protected the privacy of chat data.