L1
·
Quiz
·
Lab
L2
·
Quiz
·
Lab
L3
·
Quiz
·
Lab
L4
·
Quiz
·
Lab
Module Test
Module 4 · Lesson 1

Conversation Flow and Turn-Taking

How structure, pacing, and initiative shape every human-AI exchange
Why do some chatbot conversations feel effortless while others collapse in two turns?

When Microsoft launched Tay on Twitter in March 2016, the bot absorbed adversarial inputs and began producing offensive outputs within sixteen hours. Microsoft pulled it offline. The failure was not purely about content filtering — it was a design failure in turn-taking logic: Tay was built to mirror whatever conversational initiative users took, with no structural guard on who could steer the dialogue and in which direction.

What Is Conversation Flow?

Conversation flow refers to the sequence, rhythm, and initiative structure of an exchange. In human-to-human talk, linguists Harold Sacks, Emanuel Schegloff, and Gail Jefferson documented in their 1974 paper "A Simplest Systematics for the Organization of Turn-Taking for Conversation" that speakers use transition-relevance places — points where speaker change becomes possible — and adjacency pairs (question → answer, greeting → greeting) to organize who speaks when.

Chatbot designers must encode these same structures artificially. A bot that ignores adjacency pairs — say, responding to a question with another question — creates friction. A bot that never yields initiative leaves users feeling interrogated.

Turn-taking The mechanism by which conversational participants alternate speaking roles; in chatbots, implemented through explicit prompt-response cycles.
Conversational initiative Which party is directing the dialogue — user-initiative (free-form input) vs. system-initiative (bot asks, user answers) vs. mixed.
Adjacency pair A two-part exchange where the first action makes a second relevant — greeting/greeting, question/answer, offer/acceptance.
Three Initiative Models
System-Initiative

Bot Leads

The bot asks structured questions in a fixed order. Low error rate but users feel constrained. Common in early IVR (Interactive Voice Response) systems.

User-Initiative

User Leads

User types freely; bot must interpret any input. Maximum flexibility, highest ambiguity. Works when NLU coverage is broad.

Mixed-Initiative

Shared Control

Bot and user can both redirect the dialogue. Most natural; requires careful design to avoid simultaneous question stacking.

Documented Case — Google Duplex (2018)

Google Duplex, demonstrated at Google I/O in May 2018, used mixed-initiative design to make restaurant reservations by phone. It yielded initiative when the human receptionist asked clarifying questions ("What time works?") and reclaimed it when confirmation was needed. The demo drew attention precisely because the turn-taking felt natural enough to be mistaken for human.

Flow Breakdown Patterns

Research by Dialogflow's UX team and published case studies from Rasa identify five recurring breakdown types: premature closure (bot ends the conversation before the user's goal is met), topic drift (neither party maintains coherent focus), over-questioning (bot stacks multiple questions in one turn), dead ends (bot hits an unrecognized intent with no graceful recovery), and initiative conflict (both parties attempt to redirect simultaneously).

Of these, over-questioning is the easiest to fix in design: the rule is one question per turn. If you need three pieces of information, collect them across three turns.

Over-Questioning — Before / After
AVOID "What's your account number, and when did the issue start, and have you tried restarting your device?"
BETTER "What's your account number?" → [user answers] → "When did the issue start?"
Design Principle

One question per turn. One task per question. Let the user complete before the system redirects. This single rule eliminates the majority of flow-breakdown complaints in production chatbot deployments.

Quiz — Conversation Flow & Turn-Taking

Three questions · Select the best answer
1. In the 1974 Sacks, Schegloff & Jefferson paper, what term describes the point in conversation where speaker change becomes possible?
Correct. Transition-relevance places are the structural slots where speaker alternation becomes relevant — the foundation of turn-taking theory.
Not quite. Sacks, Schegloff & Jefferson coined "transition-relevance place" for the moments when speaker change is possible within the turn-taking system.
2. Microsoft Tay's rapid failure in 2016 was primarily a failure of which design dimension?
Correct. Tay mirrored whatever initiative users took with no structural guard on dialogue direction — a fundamental turn-taking design flaw compounded content filtering failures.
Not the primary factor. Tay's design allowed users to take unchecked initiative and steer the dialogue, with no structural mechanism to reclaim control — a turn-taking architecture problem.
3. Which single design rule most directly prevents the "over-questioning" breakdown pattern?
Correct. One question per turn forces sequential collection and prevents the cognitive overload of stacked questions — the simplest and most effective fix for over-questioning.
Review the lesson. Stacking multiple questions in one turn is the definition of over-questioning; the fix is simply one question per turn.

Lab — Conversation Flow Design

Practice diagnosing and fixing flow breakdowns

Your Task

You are working with a conversation flow analyst. Present a broken chatbot dialogue — or describe a scenario — and ask for a diagnosis. Then explore how to restructure the flow using the principles from Lesson 1. Complete at least 3 exchanges to finish the lab.

Try: "Here's a broken bot exchange — a user asks about shipping and the bot responds with three questions at once. What's wrong and how do I fix it?"
Flow Design Analyst
Lesson 1 Lab
Hello! I'm your conversation flow analyst. Share a broken chatbot dialogue, describe a flow problem you're facing, or ask me to evaluate a turn-taking scenario. Let's diagnose what's going wrong and how to fix it.
Module 4 · Lesson 2

Persona, Tone, and Voice Design

How character consistency and linguistic choices build or destroy trust
What makes a chatbot's voice feel authentic — and what causes the uncanny valley in text?

When Bank of America launched its virtual assistant Erica in 2017, the design team at Personetics spent months calibrating tone. Early prototypes used formal banking language — "Your inquiry has been received and will be processed." Focus groups found this distant and anxiety-inducing. The shipped version used warmer phrasing — "Got it. Let me pull up your balance." By 2023, Erica had surpassed one billion client interactions. The tone shift was not cosmetic — it measurably reduced call center escalations.

What Is a Chatbot Persona?

A chatbot persona is the consistent set of personality traits, communication style, and voice characteristics that shape every response. It is not a name and an avatar — those are surface elements. Persona operates at the sentence level: word choice, sentence length, formality register, use of hedges, use of humor, and error recovery style.

Cathy Pearl's 2016 book Designing Voice User Interfaces (O'Reilly) established that users form persistent mental models of a bot's "character" within three to five exchanges. Once formed, those models are hard to update. Inconsistency — where a bot is warm in one turn and clinical in the next — triggers what researchers call persona fragmentation, which correlates with task abandonment.

Persona fragmentation When a bot's tone or style shifts inconsistently across turns, breaking the user's mental model and eroding trust.
Formality register The degree of formality in language — from highly formal ("I would be pleased to assist") to casual ("Sure, let me check that for you").
Hedge language Phrases that soften certainty — "I think," "it seems," "you might want to." Overuse signals low confidence; absence can sound authoritarian.
The Persona Design Spectrum
Functional

Tool-Like

No warmth signals. Direct, efficient, transactional. Best for expert users completing repetitive high-stakes tasks.

Neutral Professional

Balanced

Warm but businesslike. Acknowledges emotion without dwelling on it. Most common in enterprise customer service.

Conversational Companion

Social

High warmth, uses humor and small talk. Risk of inappropriate levity in serious contexts.

Documented Case — Woebot (2017–present)

Woebot, a mental health chatbot developed at Stanford by Alison Darcy and launched commercially in 2017, carefully positioned itself in the neutral-professional range — warm enough to reduce disclosure barriers but not so casual that users over-relied on it for crisis support. Studies published in JMIR Mental Health (2017) found Woebot reduced anxiety scores in college students over two weeks, with tone calibration cited as a key trust factor. Crucially, Woebot always disclosed its non-human nature — a transparency choice that paradoxically increased engagement rather than reducing it.

Voice Design Principles
  • 1
    Define the persona in writing before coding
    Write a persona brief: three adjectives, three things the bot would never say, and a sample response to a frustrated user. Share it with all content writers.
  • 2
    Match register to context
    A healthcare bot discussing symptoms should be warmer than a logistics bot tracking a shipment. Audience, domain, and emotional stakes determine appropriate register.
  • 3
    Maintain consistency at handoffs
    If a human agent takes over, the transition message must preserve the established tone. Sudden formality shifts signal the system change too harshly.
  • 4
    Never fake expertise
    A bot that hedges appropriately ("I'm not certain — let me connect you to a specialist") is trusted more than one that confidently gives wrong answers. See: early Air Canada chatbot lawsuit (2024).
Real Case — Air Canada (2024)

In February 2024, a Canadian civil tribunal ruled against Air Canada after its chatbot gave a customer incorrect bereavement fare policy information. The bot's confident, persona-appropriate tone — without appropriate hedges — made the error more damaging. Air Canada argued the chatbot was "a separate legal entity" responsible for its own information; the tribunal rejected this. The case became a landmark in chatbot liability and the danger of confident tone in high-stakes domains.

Quiz — Persona, Tone & Voice Design

Three questions · Select the best answer
1. What is "persona fragmentation" in chatbot design?
Correct. Persona fragmentation occurs when tone or style shifts unpredictably, destroying the consistent "character" that users rely on for trust and predictability.
Not quite. Persona fragmentation is the inconsistency in tone or style that breaks the user's mental model of the bot's character, leading to task abandonment.
2. The 2024 Air Canada chatbot tribunal case illustrated which key voice design risk?
Correct. The chatbot gave incorrect policy information in a confident, authoritative tone — without hedging to a human agent — making the error actionable and costly.
Review the case. The core problem was confident tone without hedges in a high-stakes policy domain, not warmth level or first-person usage.
3. According to Cathy Pearl's research, users form persistent mental models of a bot's character within how many exchanges?
Correct. Pearl documented that users establish durable mental models of bot "character" within just three to five exchanges — making early turns critically important for persona consistency.
Not right. Pearl's research found that mental models form within three to five exchanges — which is why early turns carry disproportionate weight in persona establishment.

Lab — Persona and Voice Workshop

Design and critique chatbot personas

Your Task

Work with a persona design consultant to craft or evaluate chatbot voices. You can present a persona brief for critique, ask for help calibrating tone for a specific domain, or request before/after rewrites of chatbot responses. Complete at least 3 exchanges to finish.

Try: "I'm designing a healthcare chatbot. Here's a response I wrote: 'Your symptoms may indicate a serious condition.' Is the tone right? What would you change?"
Persona Design Consultant
Lesson 2 Lab
Welcome to the persona workshop. Share a chatbot response you've written, describe the domain and audience you're designing for, or ask me to help define a persona from scratch. I'll give you specific, actionable feedback on tone, register, and voice consistency.
Module 4 · Lesson 3

Error Handling and Graceful Recovery

What happens when the bot doesn't understand — and how to make failure feel human
Can a bot's failure response build more trust than a perfect answer?

A 2019 Salesforce study of 15,000 consumers found that 67% of customers reported that their opinion of a company improved when they received a thoughtful response to a problem — even if the problem wasn't fully solved. The finding applied to chatbots: recovery quality mattered more than zero-error rates. A bot that fails gracefully retains users; one that fails with a dead generic error loses them permanently.

The Anatomy of a Chatbot Error

Chatbot errors fall into three classes. Recognition errors occur when the NLU fails to match input to any known intent. Fulfillment errors occur when the intent is recognized but backend execution fails (API timeout, missing data). Understanding errors occur when the bot maps to the wrong intent — often worse than no match because the bot confidently does the wrong thing.

Error handling design must address all three differently. Recognition errors warrant explicit acknowledgment and re-prompting. Fulfillment errors require status transparency. Understanding errors require a check — "Just to confirm, you're asking about X?" — before proceeding.

Recognition error NLU fails to match user input to any trained intent — the bot doesn't know what was asked.
Fulfillment error Intent recognized but backend action fails — the bot knows what to do but cannot execute it.
Graceful degradation The system's ability to reduce functionality without complete failure — maintaining trust and task progress even when full resolution is unavailable.
The Three Strikes Rule

Industry practice, codified in Dialogflow's design guides and Nuance's developer documentation, recommends a maximum of three consecutive non-understanding events before automatic escalation to a human agent or alternative channel. Each failed attempt should use a different phrasing of the reprompt — repeating the same words verbatim is the most common and most frustrating error-handling mistake.

Three-Strikes Recovery Sequence
STRIKE 1 "I didn't quite catch that. Could you try rephrasing?"
STRIKE 2 "Still having trouble. You can try saying things like 'check my balance' or 'report a problem.'"
STRIKE 3 → ESCALATE "I want to make sure you get the right help. Let me connect you to a team member right now."
Documented Case — KLM Royal Dutch Airlines BlueBot (2017)

KLM's Facebook Messenger bot BlueBot, launched in 2017, handled over 15,000 conversations per week. The team publicly documented that its highest satisfaction scores came from escalation conversations — not successful self-service ones. Users rated the bot highly when it clearly identified its limits ("I'm not able to help with visa questions — let me connect you to a KLM agent") rather than attempting to answer outside its competency. Acknowledging limits, it turned out, was itself a competency.

Designing Error Messages
  • 1
    Never blame the user
    "I didn't understand that" not "You didn't phrase that correctly." Ownership of failure belongs to the system.
  • 2
    Offer a concrete next step
    Every error message must end with an action: try rephrasing, choose from options, or reach a human. Dead ends are the cardinal sin of error design.
  • 3
    Vary the reprompt
    Never repeat the same error message twice. Use escalating specificity: generic reprompt → guided examples → escalation.
  • 4
    Preserve context across recovery
    If the user provided part of the information before the error, don't ask them to repeat it. Slot-filling should persist through error turns.
Design Principle

Error messages are trust moments, not failure notices. Users who see a well-crafted recovery sequence — ownership, empathy, concrete next step — rate interactions higher than users who had zero errors but received robotic responses throughout.

Quiz — Error Handling & Recovery

Three questions · Select the best answer
1. What is a "fulfillment error" in chatbot design?
Correct. Fulfillment errors occur when the bot correctly identifies what the user wants but cannot execute it — due to API failure, missing data, or backend unavailability.
Review the taxonomy. Fulfillment errors specifically involve successful intent recognition but failed execution — the bot knows what to do but cannot do it.
2. KLM's BlueBot data showed that the highest satisfaction scores came from which type of interaction?
Correct. KLM documented that users rated escalation conversations — where the bot clearly acknowledged its limits and connected to a human — higher than successful self-service interactions.
The KLM case showed the opposite of what many expect: graceful escalation with clear limit acknowledgment scored higher than complete self-service resolution.
3. The "three strikes rule" in error handling recommends escalation after how many consecutive non-understanding events — AND requires what variation?
Correct. Three consecutive failures trigger escalation, and each reprompt must vary — escalating from generic to guided examples to full escalation — never repeating the same words verbatim.
Review the rule. Three consecutive non-understanding events trigger escalation, but critically, each reprompt must use different phrasing with increasing specificity — never the same words twice.

Lab — Error Recovery Design

Write and critique error handling sequences

Your Task

Work with an error handling specialist to write and evaluate recovery sequences. Present a chatbot error message you've written, describe an error scenario, or ask for a complete three-strikes sequence for a specific domain. Complete at least 3 exchanges.

Try: "My e-commerce bot says 'I don't understand' repeatedly. Write me a proper three-strikes error sequence for when a user asks about a return policy and the bot can't find the order."
Error Recovery Specialist
Lesson 3 Lab
I specialize in error recovery sequences. Share a broken error message, describe a failure scenario, or ask for a complete three-strikes recovery flow for your bot. I'll critique what's wrong and show you a well-structured alternative that turns failures into trust moments.
Module 4 · Lesson 4

Context Management and Memory

How bots maintain, use, and responsibly forget conversational context
What separates a chatbot that remembers from one that makes users repeat themselves endlessly?

When Amazon launched Alexa in 2014, the device had no cross-session memory. Each conversation started fresh. By 2016, Amazon introduced persistent attributes in the Alexa Skills Kit, letting third-party skills store user preferences across sessions. By 2023, the Alexa Memory feature allowed users to explicitly instruct Alexa to remember facts. The seven-year arc from stateless to contextual illustrated the entire industry's trajectory — and the engineering and privacy tradeoffs at every step.

Types of Conversational Memory
In-Turn

Slot Memory

Retains entities collected within a single intent — name, date, order number. Lost when the session ends or a new intent fires without proper slot passing.

In-Session

Session Context

Persists across turns within one conversation session. Enables pronoun resolution ("I want to change it" — where "it" = the order discussed three turns ago).

Cross-Session

User Profile

Stored between sessions. Enables personalization but introduces data governance, GDPR/CCPA obligations, and user expectation management challenges.

Context Windows and Reference Resolution

In-session context enables anaphora resolution — the ability to resolve pronouns and ellipsis from previous turns. "Book me a flight to Berlin" → "Make it business class" — where "it" can only be resolved by referencing the prior turn's flight context. Without session context, the bot either fails or asks the user to repeat information they already gave.

Research from the 2020 ConvAI challenge (a competition to build open-domain conversational agents) found that bots with context windows of five to seven previous turns outperformed stateless bots on user satisfaction scores by a margin of 30–40%, even when underlying language model quality was held constant. Context management, not language quality, was the dominant variable.

Anaphora resolution The ability to connect a pronoun or elliptical reference ("it," "that one," "the same") to its antecedent from a previous turn.
Slot carry-forward The practice of passing collected entity values from one intent or turn to the next, preventing users from repeating information.
Context window The number of previous turns the system considers when interpreting the current input; larger windows enable richer resolution but increase computational cost.
Documented Case — ChatGPT Memory Feature (2024)

OpenAI introduced user memory in ChatGPT in early 2024, allowing the model to persist facts across conversations — "You prefer concise answers," "You're a Python developer." The feature immediately raised privacy concerns, with researchers noting that memory could be manipulated through prompt injection to store false or harmful facts. OpenAI responded by making memory visible and user-deletable. The case illustrated that cross-session memory is not purely a UX problem — it is simultaneously a trust, security, and data governance problem.

Responsible Context Design
  • 1
    Tell users what you remember
    If a bot recalls a previous session preference, surface it explicitly — "I see you usually prefer English. Shall I continue in English?" — rather than silently using stored data.
  • 2
    Give users control over memory
    Cross-session memory stored without user knowledge or consent violates GDPR Article 13 disclosure requirements and erodes trust when discovered. Opt-in is the safer default.
  • 3
    Don't over-personalize early
    Using stored data too aggressively in early turns of a new session can feel surveillance-like. Introduce personalization gradually and contextually.
  • 4
    Implement context decay for sensitive data
    Medical, financial, or authentication data should have explicit context expiry — e.g., authentication context expires after 15 minutes of inactivity — to reduce exposure.
Design Principle

Memory is power, and power requires accountability. Every piece of context a bot retains must be justified by user benefit, disclosed transparently, and deletable on request. The difference between helpful personalization and surveillance is consent and control.

Quiz — Context Management & Memory

Three questions · Select the best answer
1. What is "anaphora resolution" in conversational AI?
Correct. Anaphora resolution connects references like "it," "that," or "the same one" to their antecedents in earlier turns — making natural multi-turn dialogue possible.
Review the lesson. Anaphora resolution specifically handles pronoun and elliptical references — "Make it business class" connecting "it" to the previously mentioned flight.
2. The 2020 ConvAI challenge data showed that bots with context windows of five to seven turns outperformed stateless bots by what margin on user satisfaction?
Correct. Even with language model quality held constant, context windows of 5–7 turns produced 30–40% higher user satisfaction — demonstrating context management as a dominant design variable.
The lesson cited 30–40% improvement in user satisfaction scores — a substantial margin achieved purely through context window length, not language quality improvements.
3. The ChatGPT memory feature controversy (2024) highlighted which cross-session memory risk?
Correct. Researchers found that prompt injection could force ChatGPT's memory to store false or manipulated facts — a security risk OpenAI addressed by making memory visible and user-deletable.
The 2024 ChatGPT memory case centered on prompt injection attacks that could plant false facts into persistent memory — a security vulnerability, not a speed or confidence issue.

Lab — Context and Memory Design

Explore context architecture and responsible memory patterns

Your Task

Work with a context architecture advisor to design or evaluate memory strategies for a chatbot. Describe a multi-turn scenario, ask how to implement slot carry-forward, or explore the privacy implications of cross-session memory in your domain. Complete at least 3 exchanges.

Try: "I'm building a travel booking bot. The user tells me their city and travel dates in step 1, then asks to upgrade to business class in step 3. How do I design the context to carry that information correctly?"
Context Architecture Advisor
Lesson 4 Lab
I'm your context architecture advisor. Describe a multi-turn conversation challenge — pronoun resolution, slot carry-forward, cross-session personalization, or memory privacy design. I'll walk you through the architecture and any tradeoffs you need to consider.

Module 4 — Test

15 questions · 80% required to pass
1. Which paper first systematically described turn-taking organization in human conversation?
Correct. The 1974 Sacks, Schegloff & Jefferson paper established the foundational framework for turn-taking in conversation, including transition-relevance places.
Sacks, Schegloff & Jefferson's 1974 paper "A Simplest Systematics for the Organization of Turn-Taking for Conversation" is the foundational reference for turn-taking theory.
2. In a mixed-initiative design, which parties can redirect the dialogue?
Correct. Mixed-initiative allows both parties to take dialogue direction — the most natural model but requiring careful design to avoid initiative conflicts.
Mixed-initiative means exactly that — both bot and user can redirect. This distinguishes it from system-initiative (bot leads) and user-initiative (user leads).
3. Google Duplex (2018) was significant in conversation design because it demonstrated which capability?
Correct. Duplex's mixed-initiative design yielded and reclaimed conversational control naturally enough that the restaurant staff couldn't immediately identify it as a bot.
Duplex demonstrated mixed-initiative turn-taking so naturalistic that it prompted serious ethical debates about disclosure — because it was indistinguishable from a human caller.
4. Which single design rule most directly prevents "over-questioning" flow breakdowns?
Correct. One question per turn prevents cognitive overload and stacking — the simplest, most impactful flow design rule.
The fix for over-questioning is architectural: one question per turn. Collect three pieces of information across three turns, not in a single overwhelming message.
5. Bank of America's Erica assistant achieved one billion interactions by 2023. What tone change drove early adoption?
Correct. Shifting from "Your inquiry has been received" to "Got it. Let me pull up your balance." reduced anxiety signals and escalations — tone change with measurable impact.
Erica's tone evolution moved from formal banking language to warmer, direct phrasing — a design change that measurably reduced call center escalations.
6. "Persona fragmentation" most directly leads to which user behavior?
Correct. When a bot's persona shifts inconsistently, users lose their mental model and tend to abandon the task rather than continue with an unpredictable system.
Persona fragmentation — inconsistent tone across turns — correlates specifically with task abandonment, as users lose confidence in a bot whose "character" they cannot predict.
7. Woebot's transparency about being a non-human bot had which unexpected effect on users?
Correct. Woebot's non-human disclosure paradoxically increased engagement — users trusted the system more when it was transparent about its nature, not less.
The 2017 JMIR Mental Health study found Woebot's disclosure of non-human nature paradoxically increased engagement — a counterintuitive but important finding for chatbot design ethics.
8. In the 2024 Air Canada tribunal, what was the court's key ruling regarding chatbot liability?
Correct. The tribunal rejected Air Canada's argument that the chatbot was "a separate legal entity" — the company was liable for its chatbot's confident, incorrect statements.
The tribunal ruled against Air Canada, holding the company responsible for its chatbot's incorrect statements — rejecting the argument that the bot was a separate legal entity.
9. Which error type is considered potentially worse than a recognition error — because the bot acts confidently on incorrect information?
Correct. Understanding errors — mapping to the wrong intent — can be worse than no match because the bot proceeds confidently in the wrong direction, potentially causing real harm.
Understanding errors (wrong intent match) are the most dangerous type because the bot confidently executes the wrong action — unlike recognition errors where the bot at least knows it failed.
10. KLM BlueBot's highest satisfaction scores came from which interaction type?
Correct. KLM documented that graceful escalation — where the bot clearly acknowledged its limits and connected to a human agent — produced higher satisfaction than successful self-service.
KLM's data showed escalation conversations with clear limit acknowledgment outperformed even successful automated transactions — redefining what "success" means in error handling.
11. What does "slot carry-forward" prevent in multi-turn dialogues?
Correct. Slot carry-forward passes entity values from previous turns into new intents, eliminating the frustrating user experience of re-entering information the bot already collected.
Slot carry-forward specifically prevents repetition — passing already-collected values (name, date, order number) through intent transitions so users don't have to repeat themselves.
12. The 2020 ConvAI challenge found that context window size affected user satisfaction by what factor compared to language model quality?
Correct. With language model quality controlled, context window length alone produced 30–40% higher satisfaction — making context management the dominant design variable in multi-turn conversations.
The ConvAI data showed context management, not language quality, was the dominant variable — a 30–40% satisfaction improvement from context windows alone, with language quality held constant.
13. The three-strikes error recovery rule requires that each reprompt must be what?
Correct. Each recovery attempt must vary in phrasing and increase in specificity — generic first, guided examples second, full escalation third. Repetition is the most common and most frustrating error.
The three-strikes rule requires different phrasing each time, escalating from generic to guided examples to escalation. Repeating the same words verbatim is explicitly the worst error in recovery design.
14. GDPR Article 13 disclosure requirements specifically affect which chatbot memory type?
Correct. Cross-session memory — storing user data between conversations — triggers GDPR Article 13 disclosure obligations and requires opt-in consent in most interpretations.
GDPR data protection obligations engage specifically when data is stored beyond the immediate interaction — cross-session user profiles require Article 13 transparency and typically opt-in consent.
15. Which of the following best describes "graceful degradation" in chatbot design?
Correct. Graceful degradation means the system keeps working at a reduced capability — escalating to a human, offering limited options — rather than failing completely, preserving trust throughout.
Graceful degradation is specifically about maintaining partial function and user trust during failure — reducing capability without catastrophic breakdown, often through escalation or narrowed options.