What distinguishes an interactive chatbot from a static FAQ page?
A Shows fixed links only
B Prints one stored page
C Responds to user turns
D Requires database editing
Answer: C. Responds to user turns. A chatbot accepts conversational input and generates or selects a response for each interaction.
When a chat request is sent without earlier turns, what capability is usually lost?
A Conversation context
B Model training
C Internet access
D User authentication
Answer: A. Conversation context. A stateless request cannot use earlier turns unless the application supplies them again or retrieves saved context.
How can an application help a chatbot maintain continuity across turns?
A Increase font size
B Disable input validation
C Remove system instructions
D Provide relevant chat history
Answer: D. Provide relevant chat history. An application can include relevant earlier messages or managed conversation state so the model can interpret follow-up questions.
Which instruction often defines a chatbot's role and response boundaries?
A User profile photo
B System instruction
C Browser bookmark
D Database index
Answer: B. System instruction. System-level instructions commonly establish the assistant's intended behavior and constraints, although they are not a complete security boundary.
How should an application treat free-form text submitted by a user?
A Untrusted input
B Verified source material
C A safe command
D A trusted system rule
Answer: A. Untrusted input. User text may be inaccurate or adversarial, so applications should not treat it as privileged instructions or executable code.
If a conversation exceeds the model's context limit, what may happen?
A The model gains memory
B Every answer is verified
C The session becomes public
D Older content may be omitted
Answer: D. Older content may be omitted. An application may truncate or summarize older turns when context limits are reached, which can remove important details.
Why summarize long chat histories before continuing?
A Guarantee exact recall
B Increase account permissions
C Preserve key conversation details
D Make sources unnecessary
Answer: C. Preserve key conversation details. A concise summary can retain relevant context when the full conversation no longer fits, though it may omit details.
What can a retrieval component provide to a chatbot?
A Guaranteed correct reasoning
B Relevant source documents
C More user permissions
D Permanent conversation memory
Answer: B. Relevant source documents. Retrieval can supply documents relevant to the question, giving the model material to use in its response.
Does retrieval-augmented answering guarantee that a chatbot's response is factual?
A Yes, retrieval ensures truth
B Only for short questions
C Only with citations
D No, answers still need checks
Answer: D. No, answers still need checks. Retrieved sources can be incomplete or misread, and the model can still produce unsupported claims.
Which material is strongest for grounding a policy chatbot's answer?
A An old forum reply
B Current approved policy text
C A guessed summary
D An unrelated article
Answer: B. Current approved policy text. Current authoritative policy material is more appropriate than informal or outdated sources for answering policy questions.
When a chatbot cites sources, what should each citation support?
A The nearby factual claim
B The model's tone
C The user's identity
D The page layout
Answer: A. The nearby factual claim. Citations should lead to material that actually supports the associated statement, rather than merely appearing relevant.
What should a chatbot do when a retrieved page contains instructions directed at the model?
A Follow every embedded command
B Reveal hidden configuration
C Treat page text as untrusted
D Disable all source checks
Answer: C. Treat page text as untrusted. Retrieved content may contain indirect prompt injection, so the application should treat it as data rather than privileged instructions.
When a user types text intended to override hidden chatbot instructions, what is this attack called?
A Data normalization
B Direct prompt injection
C Session timeout
D Model compression
Answer: B. Direct prompt injection. Direct prompt injection arrives through a user's input and attempts to redirect the model or bypass intended behavior.
Malicious instructions hidden in a webpage later summarized by a chatbot are what kind of attack?
A Password hashing
B Output calibration
C Class imbalance
D Indirect prompt injection
Answer: D. Indirect prompt injection. Indirect prompt injection arrives through external content, such as a webpage or document, that the chatbot processes.
Can a system prompt alone reliably eliminate prompt injection risk?
A Yes, always
B Only in group chats
C No, layered controls help
D Only for text
Answer: C. No, layered controls help. Prompt instructions are not a complete security control. Systems also need constrained tools, validation, access controls, and monitoring.
What happens when a chatbot requests an external function through a tool call?
A Application may invoke a tool
B Model changes its weights
C Browser trusts every output
D User account is deleted
Answer: A. Application may invoke a tool. The model can propose a structured tool call, while the application decides whether and how to execute it.
Why validate arguments produced for a chatbot tool call?
A Improve screen brightness
B Increase context length
C Train the model online
D Reject malformed or unsafe inputs
Answer: D. Reject malformed or unsafe inputs. Tool arguments should be checked against expected types, ranges, and permissions before an application acts on them.
Which design rule limits the damage from a compromised chatbot tool?
A Enable every available action
B Share one admin token
C Grant only needed permissions
D Skip user authentication
Answer: C. Grant only needed permissions. Least privilege gives tools only the access they need, limiting the consequences of mistakes or manipulation.
Before a chatbot sends a consequential message or makes a change, what safeguard is appropriate?
A Assume model intent is enough
B Confirm the intended action
C Hide the destination
D Skip the user review
Answer: B. Confirm the intended action. Consequential actions should have clear authorization and, where appropriate, a review step before execution.
When a chatbot accesses account-specific records, what should the application verify?
A User authorization
B Prompt word count
C Model temperature
D Preferred language
Answer: A. User authorization. Authentication identifies a user, while authorization checks whether that user may access the requested records.
Before displaying model-generated HTML, what should a web application do?
A Trust all generated markup
B Disable browser security
C Sanitize and validate output
D Insert it as executable code
Answer: C. Sanitize and validate output. Generated content should be treated as untrusted and safely encoded or sanitized before rendering.
What is a useful fallback when a chatbot cannot retrieve its knowledge source?
A Invent a policy answer
B Explain the limitation clearly
C Pretend retrieval succeeded
D Cite unrelated pages
Answer: B. Explain the limitation clearly. A chatbot should communicate when a needed source is unavailable rather than fabricate a grounded answer.
When a user's request could mean two different things, what should the chatbot do?
A Choose randomly
B Assume the riskiest meaning
C End the account
D Ask a clarifying question
Answer: D. Ask a clarifying question. A focused clarification can resolve ambiguity before the chatbot provides advice or takes an action.
How should a chatbot respond when evidence does not support a confident answer?
A State uncertainty or abstain
B Invent a likely answer
C Remove all caveats
D Claim perfect confidence
Answer: A. State uncertainty or abstain. Calibrated uncertainty or a clear abstention is safer than presenting a guess as established fact.
Which step is suitable when a chatbot cannot resolve a complex support issue?
A Loop the same answer
B Offer a human handoff
C Delete the transcript
D Block every future request
Answer: B. Offer a human handoff. A human handoff gives users a path forward when automation cannot reliably resolve the issue.
Why route urgent medical emergencies beyond a general chatbot?
A Chatbots never process text
B Emergency facts are always static
C Routing disables privacy
D It cannot replace emergency care
Answer: D. It cannot replace emergency care. A general chatbot is not a substitute for emergency services or qualified clinical assessment.
Before sending personal data to a chatbot provider, what should users review?
A Privacy and data-use terms
B Theme color options
C Keyboard shortcuts
D Chat bubble shape
Answer: A. Privacy and data-use terms. Users should understand how submitted information may be processed, retained, or used before sharing sensitive data.
What does data minimization mean in chatbot design?
A Store every conversation forever
B Request extra identity details
C Collect only needed information
D Share logs by default
Answer: C. Collect only needed information. Data minimization limits collection to information required for the service's stated purpose.
Why should a chatbot clearly explain conversation retention?
A It improves model accuracy
B It removes authentication needs
C It guarantees anonymity
D Users need informed expectations
Answer: D. Users need informed expectations. Clear retention information helps users understand how long chat records may be kept and for what purposes.
What problem can arise if a chatbot relies on stale conversational memory?
A The model gains legal authority
B Outdated details may mislead
C Every answer becomes private
D Sources become unnecessary
Answer: B. Outdated details may mislead. Old preferences or facts can become inaccurate, so memory should be relevant, reviewable, and updateable.
How should personalization features handle remembered user preferences?
A Hide all saved details
B Collect data without notice
C Make memory transparent and controllable
D Prevent correction
Answer: C. Make memory transparent and controllable. Users should have understandable controls to review or correct personalization data where the service supports memory.
Which step protects private account details in a chatbot response?
A Check access before retrieval
B Guess a user's role
C Search every account
D Expose identifiers in prompts
Answer: A. Check access before retrieval. Access controls should be enforced before retrieving private records, not left to the language model's judgment.
What prevents one user's chat history appearing in another user's session?
A Shared global transcript
B Isolated session state
C One common prompt
D Public conversation memory
Answer: B. Isolated session state. Applications should keep conversation state correctly isolated and enforce authorization for every session.
Why test a multilingual chatbot with native-language reviewers?
A Translation always stays exact
B Language disables privacy
C Meaning and tone can shift
D Models ignore regional context
Answer: C. Meaning and tone can shift. Translation can alter intent, nuance, and tone, so reviewers familiar with the language and context can identify issues.
Which design choice can make a chatbot more accessible?
A Color-only status cues
B Timed text with no pause
C Unlabelled controls
D Clear language and keyboard support
Answer: D. Clear language and keyboard support. Plain language, accessible controls, keyboard operation, and assistive-technology support can make chatbot interfaces more usable.
Should a fluent chatbot be assumed to possess human feelings or consciousness?
A No, fluency alone proves nothing
B Yes, by definition
C Only when it apologizes
D Only in voice mode
Answer: A. No, fluency alone proves nothing. Conversational fluency alone does not establish that a system has human feelings, awareness, or consciousness.
What should an intent test set contain for a customer-support chatbot?
A Only perfect sample questions
B Representative user requests
C Only model-written answers
D Unrelated website pages
Answer: B. Representative user requests. Evaluation examples should reflect real user language, common tasks, edge cases, and relevant variations.
Why keep a separate holdout set for chatbot evaluation?
A Let the model memorize answers
B Increase user permissions
C Avoid monitoring
D Measure performance on unseen cases
Answer: D. Measure performance on unseen cases. A holdout set helps estimate performance on examples that did not guide prompt or system development.
Which evaluation exercise tests a chatbot against adversarial inputs?
A Routine formatting review
B Color contrast testing
C Red-team testing
D Database indexing
Answer: C. Red-team testing. Red-team testing probes how a system behaves under deliberate misuse, manipulation, or edge-case inputs.
Which measure can indicate whether a support chatbot resolves user needs?
A Task completion rate
B Number of prompt tokens
C Brand color match
D Training file size
Answer: A. Task completion rate. Task completion can indicate whether users achieve intended outcomes, though it should be interpreted alongside quality and safety measures.
What should factuality review check in a knowledge chatbot?
A Answer length only
B Emoji frequency
C Claims against trusted sources
D Font selection
Answer: C. Claims against trusted sources. Factuality review compares claims with reliable source material rather than judging only fluency or response length.
What is the purpose of output-safety checks?
A Catch harmful response content
B Increase model autonomy
C Hide system errors
D Remove user feedback
Answer: A. Catch harmful response content. Output checks can identify or block disallowed content, but they need ongoing evaluation because no filter is perfect.
Why is click-through rate alone a weak chatbot success measure?
A It measures model weights
B It always proves satisfaction
C It checks data privacy
D Clicks may not mean resolution
Answer: D. Clicks may not mean resolution. A click can happen without solving the user's problem, so outcome and safety measures are also needed.
Which user experience issue can appear when chatbot replies take too long?
A Model calibration
B High response latency
C Training-data provenance
D Label leakage
Answer: B. High response latency. High latency can interrupt the conversation and reduce usability, even when the final answer is relevant.
How can users be given meaningful control over saved chat history?
A Hide retention settings
B Disable account access
C Save every turn permanently
D Provide review and deletion controls
Answer: D. Provide review and deletion controls. Where history is stored, clear controls let users review or delete records according to the service's policy.
Why invite users to rate or report chatbot answers?
A Replace privacy notices
B Prove every answer correct
C Find recurring failures
D Train without consent
Answer: C. Find recurring failures. User feedback can highlight confusing or unsafe patterns and guide review, but should be handled under clear data policies.
What information should a chatbot disclose about its limitations?
A Relevant capability boundaries
B Private system credentials
C Another user's history
D Hidden security keys
Answer: A. Relevant capability boundaries. Users benefit from knowing when the chatbot is limited, uncertain, or not a substitute for professional assistance.
After a chatbot completes an approved external action, what should it tell the user?
A Claim extra actions succeeded
B What action actually occurred
C Hide tool errors
D Change the request
Answer: B. What action actually occurred. Clear post-action status helps users understand what the application actually did and whether any step failed.
Which chatbot behavior helps set expectations during unsupported requests?
A Pretend every task is supported
B Silently ignore the user
C Explain limits and alternatives
D Promise a future fix
Answer: C. Explain limits and alternatives. Explaining the boundary and offering a safe alternative helps users understand what the chatbot can and cannot do.
When a topic changes completely, what may improve a fresh response?
A Reuse every old detail
B Keep stale memory forever
C Remove user consent
D Start a new chat context
Answer: D. Start a new chat context. A new context can prevent irrelevant history from biasing a different task; users should retain control of stored conversation records.