Digital Merchandising and Customer Experience Design

Introduction

Digital merchandising is the deliberate organization, presentation, recommendation, and promotion of products in a digital storefront. It performs a role analogous to physical merchandising, but with important differences: digital stores can personalize displays, observe behavior at scale, test alternatives rapidly, and make products discoverable through search, recommendations, social channels, and external platforms. These capabilities also create risks. A storefront can become confusing, manipulative, overly personalized, inaccessible, or optimized for short-term conversion at the expense of trust and long-term value.

Customer experience design provides the broader discipline within which digital merchandising operates. It considers the complete relationship between a customer and a business across discovery, evaluation, purchase, fulfillment, use, service, and repurchase. The central managerial challenge is therefore not simply to make a page attractive. It is to reduce the effort and uncertainty involved in making a suitable decision while creating value for both customer and enterprise.

This chapter presents an integrated approach to user experience, customer journey mapping, information architecture, site search, recommendations, trust signals, persuasive design, experimentation, and ethical governance. The emphasis is application: students should be able to diagnose a storefront, prioritize improvements, design interventions, and evaluate results without using deceptive practices.

Learning Objectives

By the end of this chapter, readers should be able to:

1. Explain how digital merchandising contributes to customer value and commercial performance.
2. Map customer journeys and identify moments of friction, uncertainty, and opportunity.
3. Design an information architecture and search experience that support discovery.
4. Evaluate recommendation systems, trust signals, and persuasive design critically.
5. Select appropriate experience and business metrics and interpret them responsibly.
6. Design a defensible A/B experiment and distinguish causal evidence from observational association.
7. Apply ethical, accessibility, privacy, and consumer-protection principles to storefront decisions.

1. From Product Display to Experience System

A digital storefront is a socio-technical system. Its visible interface is only one layer. Beneath it are product data, inventory, pricing rules, search indexes, recommendation models, payment services, analytics, customer support, and fulfillment operations. A visually polished interface cannot compensate for inaccurate stock information, late delivery, confusing returns, or poor product quality.

Digital merchandising has four connected purposes:

• Discovery: helping customers find relevant products.
• Evaluation: making differences, benefits, risks, and suitability understandable.
• Decision support: reducing uncertainty and cognitive effort.
• Commercial coordination: presenting products in ways consistent with inventory, margin, brand strategy, and customer value.

A useful principle is relevance before persuasion. Merchandising should first help a customer identify what fits their needs; promotional techniques should not obscure material information or pressure customers into unsuitable purchases.

The distinction between usability and experience is important. Usability concerns whether a user can accomplish a task effectively, efficiently, and satisfactorily. Experience includes usability but also encompasses emotion, trust, identity, perceived fairness, anticipation, and memory. A checkout may be usable yet produce a poor experience if fees appear unexpectedly. Conversely, an entertaining interface may be memorable but commercially weak if customers cannot compare products.

A service-system view

The customer’s experience is co-produced. The retailer controls some touchpoints, such as product pages and checkout, but suppliers, carriers, payment providers, review communities, marketplaces, and customer circumstances also shape the outcome. Consequently, experience metrics should be connected to operational realities. A “promise delivered” indicator is more meaningful than a banner claiming fast shipping if the fulfillment system cannot support it.

2. Customer Journey Mapping and Experience Diagnosis

A customer journey map is a structured representation of a target customer’s progression through stages, touchpoints, tasks, questions, emotions, and outcomes. It is not merely a decorative timeline. Its value lies in exposing dependencies between channels and identifying where the organization’s assumptions differ from customer reality.

A robust map typically includes:

1. Persona or segment: the customer group and relevant context.
2. Goal: what the customer is trying to accomplish.
3. Stages: such as need recognition, discovery, evaluation, purchase, receipt, use, support, and repurchase.
4. Touchpoints: search engines, advertisements, category pages, product pages, reviews, checkout, email, delivery, and service.
5. Customer questions: for example, “Will it fit?” or “Can I return it?”
6. Evidence and emotion: what the customer sees, feels, and believes.
7. Pain points: friction, uncertainty, delay, inconsistency, or lack of control.
8. Business processes: data, systems, policies, and teams behind the touchpoint.
9. Opportunities: interventions that could improve customer and business outcomes.

Journey mapping should be evidence-based. Interviews and diary studies reveal goals and language; usability tests reveal task failures; analytics reveal behavioral patterns; service logs reveal recurring problems. No single source is sufficient. For example, a high exit rate on a product page may indicate poor content, an unattractive price, comparison shopping, or a technical error.

Moments of truth

A moment of truth is a point at which a customer forms or revises a significant judgment. Examples include seeing the first relevant search results, discovering the total price, reading a critical review, receiving an order, or requesting a return. These moments deserve priority because they often affect trust disproportionately. A journey map should distinguish high-volume problems from high-consequence problems; a low-frequency payment failure may be more urgent than a minor navigation issue.

Figure: Customer experience emerges from customer-facing touchpoints and the organizational capabilities that support them.

3. Information Architecture and Findability

Information architecture (IA) organizes content, products, labels, navigation, and relationships so that users can understand where they are and what actions are possible. In commerce, IA must support both browsing and targeted retrieval. A customer may begin with a broad category (“running shoes”) or a precise query (“women’s waterproof trail shoes size 7”).

Core IA decisions

Taxonomy defines categories and attributes. Categories should reflect customer mental models rather than only internal departments. Attributes such as size, material, compatibility, color, use case, and price enable filtering and comparison.

Labeling determines the words used for navigation, filters, and calls to action. Labels should be familiar, specific, and consistent. Internal terminology may be efficient for staff but meaningless to customers.

Hierarchy determines parent-child relationships and depth. Excessive depth increases navigation effort, while overly broad categories create scanning burdens. A category may be organized by product type, user need, occasion, or audience; the appropriate structure should be tested with representative customers.

Metadata makes products retrievable and comparable. Product data governance is therefore a merchandising capability, not an administrative afterthought. Missing or inconsistent attributes degrade filters, search, recommendations, accessibility, and analytics simultaneously.

Useful research methods include card sorting, tree testing, first-click testing, search-log analysis, and moderated usability testing. Card sorting explores how users group concepts; tree testing evaluates whether users can locate items in a proposed hierarchy. These methods produce evidence about findability but should be combined with commercial and operational constraints.

Designing for mobile and accessibility

Mobile users often operate under time, attention, and connectivity constraints. Navigation should expose priority tasks, preserve context, and avoid forcing users to repeatedly reopen filters. Accessibility requires more than compliance testing. Semantic headings, keyboard operation, visible focus, sufficient contrast, meaningful alternative text, clear error messages, scalable text, and compatibility with assistive technologies improve experience for all users. Product images should supplement, not replace, textual information.

4. Search as a Decision-Support System

On-site search is not merely a box connected to a database. It is an interaction in which the system interprets imperfect language and returns an ordered set of choices. Search quality has at least four dimensions:

• Retrieval: whether relevant products are found.
• Ranking: whether the most useful results appear first.
• Understanding: whether queries, filters, and corrections are clear.
• Commercial integrity: whether ranking reflects legitimate relevance rather than undisclosed manipulation.

Search logs reveal demand language, missing products, vocabulary gaps, and failed queries. Important capabilities include autocomplete, spelling correction, synonyms, tolerance for singular/plural differences, faceted filtering, zero-result recovery, and transparent sorting controls. However, automatic correction can be harmful when it silently changes a meaningful query. A system should explain corrections and allow the original query to be restored.

Ranking may incorporate textual relevance, availability, popularity, quality, price, margin, or personalization. These objectives can conflict. A high-margin product is not necessarily the best result for a customer. If commercial factors affect ranking, governance should define acceptable boundaries and, where material, communicate them appropriately.

Search performance can be monitored through measures such as search exit rate, reformulation rate, zero-result rate, product-detail engagement, and conversion after search. These are diagnostic rather than definitive. A low exit rate may indicate success or may reflect customers abandoning the site without leaving the page.

5. Recommendations and Personalization

Recommendation systems propose products based on signals such as current context, stated preferences, browsing behavior, purchase history, product similarity, and aggregate behavior. Common approaches include:

• Content-based recommendation: recommends items with attributes similar to an item or profile.
• Collaborative filtering: uses patterns across users or items.
• Contextual recommendation: incorporates device, location, season, or task context.
• Hybrid recommendation: combines several methods to manage their weaknesses.

Recommendations can reduce search costs and expose complementary products, substitutes, or replenishment options. They can also create filter bubbles, reinforce popularity, expose sensitive inferences, and recommend unsuitable or unavailable products. New products and new customers create a “cold-start” problem because little behavioral data exists. Diversity and novelty should therefore be balanced with predicted relevance.

A recommendation should answer a customer-relevant question: “What else might help me complete this task?” Labels such as “Often bought together,” “Similar items,” and “Because you viewed” communicate different rationales. The rationale should be truthful. A product should not be labeled “frequently bought together” if the statement is unsupported.

Personalization must be proportionate. The fact that a system can infer a preference does not mean it should display that inference. Sensitive characteristics and vulnerable circumstances require heightened safeguards. Customers should retain meaningful control over recommendations and should not be denied essential information because they decline personalization.

6. Trust Signals and Persuasive Design

Trust is a judgment about competence, integrity, benevolence, and predictability. In commerce, trust signals include:

• clear total pricing, including delivery and mandatory charges;
• accurate availability and delivery estimates;
• identifiable business and contact information;
• authentic reviews with understandable moderation practices;
• secure payment indicators used accurately;
• transparent returns, warranties, and cancellation terms;
• product specifications, dimensions, compatibility, and limitations;
• accessible customer service;
• explanations of recommendations and sponsored placement.

Trust signals must be operationally supported. A security badge cannot compensate for a misleading return policy. Reviews should not be presented as representative if they are selectively displayed without disclosure. Incentivized reviews should be identified, and businesses should not suppress legitimate criticism merely because it is unfavorable.

Persuasive design is the use of interface and communication choices to influence behavior. Ethical persuasion makes relevant value visible and supports informed choice. Deceptive or manipulative design—often called a dark pattern—uses obstruction, hidden costs, false urgency, disguised advertising, forced continuity, confirmshaming, or confusing opt-outs.

A practical ethical test asks:

1. Does the design disclose information material to the decision?
2. Can a reasonable customer understand the consequence of the action?
3. Is refusal or cancellation as understandable as acceptance?
4. Would the business defend the design if shown to customers and regulators?
5. Does the design create avoidable harm for vulnerable users?

Urgency messages should be based on real constraints. “Only two left” is appropriate only when it accurately describes the relevant stock and does not reset deceptively. Countdown timers should represent a genuine deadline, not an endlessly repeating pressure mechanism.

7. Measurement: From Activity to Outcomes

Experience measurement should connect customer behavior, perception, and business outcomes. Common measures include:

where CTR is click-through rate, C is the number of clicks, and I is the number of impressions. Both counts must refer to the same defined population and period. CTR measures response to an impression; it does not establish purchase intent or profitability.

where O is the number of defined orders and S is the number of eligible sessions or visitors. The denominator must be specified because session-based and customer-based conversion rates answer different questions.

where AOV is average order value, R is attributed revenue, and O is the number of orders. Revenue definitions should state whether taxes, discounts, shipping, refunds, and cancellations are included.

Other useful measures include search success, task completion, error rate, time to product discovery, return rate, repeat purchase, customer complaints, and accessibility defects. No metric should be optimized in isolation. Increasing conversion by promoting unsuitable products may increase returns and reduce lifetime value.

A useful decision framework separates:

• Outcome metrics: revenue, contribution margin, retention, repeat purchase, and customer satisfaction.
• Behavioral metrics: search reformulation, filter use, add-to-cart, checkout completion, and returns.
• Diagnostic metrics: page latency, broken links, missing attributes, and error messages.
• Guardrail metrics: complaints, cancellations, refunds, accessibility failures, privacy incidents, and unequal performance across segments.

Worked example: interpreting an intervention

An online store receives 20,000 eligible sessions. The existing experience produces 600 orders; a redesigned experience produces 660 orders. Existing conversion is 600 ÷ 20,000 = 3.0%, while redesigned conversion is 660 ÷ 20,000 = 3.3%. The absolute increase is 0.3 percentage points, and the relative increase is (3.3% − 3.0%) ÷ 3.0% = 10%.

This result alone does not prove that the redesign caused the increase. Traffic mix, seasonality, campaign exposure, technical changes, and random variation must be considered. Nor does it establish business success if average order value falls or returns rise. A graduate-level evaluation therefore examines the experiment design, uncertainty, segment effects, and guardrail metrics.

8. Experimentation and Causal Reasoning

A/B testing randomly assigns eligible traffic to a control and a treatment, then compares predefined outcomes. Randomization helps balance observed and unobserved characteristics in expectation, allowing a causal interpretation when implementation is sound.

Before launch, define the hypothesis, primary metric, population, treatment, duration, exclusion rules, and guardrails. Avoid changing the primary metric after observing results. Ensure that users are assigned consistently and that the treatment is technically delivered. Tests should run long enough to cover relevant purchasing cycles and should not be stopped merely because an early result looks favorable.

For a binary outcome, an estimated conversion difference is:

where p_T is treatment conversion and p_C is control conversion. The estimate is meaningful only with an appropriate design and uncertainty assessment. Sample size depends on baseline conversion, minimum practically important effect, desired power, significance threshold, allocation, and variance. Statistical significance is not equivalent to managerial importance; a tiny effect may be detectable but not worth implementation cost.

Experimentation also raises ethical questions. Do not knowingly expose customers to unsafe payment flows, materially misleading information, discriminatory treatment, or avoidable service failure. Tests involving prices, eligibility, or essential terms require particular scrutiny. Holdout groups may be inappropriate where the intervention is a necessary correction.

9. An Integrated Design Process

A disciplined improvement cycle can be organized as follows:

1. Frame the customer problem. Identify the task, segment, context, and desired outcome.
2. Collect evidence. Combine interviews, usability sessions, analytics, search logs, service data, and operational information.
3. Map the journey. Locate moments of uncertainty and failure across channels.
4. Diagnose the system. Trace visible problems to content, data, technology, policy, or process causes.
5. Generate and prioritize opportunities. Consider customer impact, business value, effort, risk, and reversibility.
6. Prototype. Test content, layout, navigation, search behavior, and recommendations before full implementation.
7. Validate accessibility and integrity. Check keyboard use, assistive technology, privacy, disclosures, pricing, and cancellation.
8. Experiment or pilot. Use controlled testing where feasible and qualitative evaluation where numbers are insufficient.
9. Monitor outcomes and guardrails. Review performance by device, channel, and relevant customer groups.
10. Institutionalize learning. Record decisions, results, limitations, and follow-up actions.

Prioritization should not be a contest for the highest conversion estimate. A modest improvement that reduces returns, complaints, or support costs may create more durable value than a high-pressure promotion.

Illustrative Case: Northstar Outdoor Equipment

Background

Northstar Outdoor Equipment is a fictional mid-sized online retailer selling hiking and camping products. Traffic is growing, but customers report difficulty choosing among similar products. Analytics show strong category-page visits but substantial exits from product pages. Customer service receives repeated questions about waterproof ratings, pack capacity, sizing, and returns.

Problem

Management initially proposes adding more “Buy now” buttons and a personalized product carousel. The UX team argues that the deeper problem is uncertainty, not insufficient persuasion. Product information is inconsistent, filters use internal terminology, and delivery estimates are shown only late in checkout.

Analysis

The team conducts 12 customer interviews, a tree test of the category structure, a review of 90 days of search logs, and a usability study with eight target users. Findings include:

• customers search by activity and conditions, while navigation is organized mainly by supplier category;
• “hydrostatic head” is displayed without an explanation of what the rating means;
• pack dimensions and torso fit are missing for some products;
• search queries for “rain jacket” return fashion jackets alongside technical shells;
• the return policy is accurate but difficult to find;
• recommendations often show accessories that are incompatible with the selected product.

Approach and solution

Northstar creates a customer-centered taxonomy organized by activity, product type, and conditions. It standardizes product attributes and establishes ownership for data quality. Product pages receive a comparison table, plain-language explanations, fit guidance, delivery estimates, and a visible returns summary. Search adds synonyms, activity filters, and zero-result recovery. Recommendations are restricted to compatibility-checked accessories and clearly labeled as “compatible additions.”

The team tests two changes separately: improved product information and revised search filters. The primary outcome is completed purchase, with return rate, customer questions, and product-page task success as guardrails. The experiment plan is reviewed by legal, accessibility, merchandising, and operations representatives.

Outcome

In this illustrative case, the information intervention increases completed purchases modestly, but its most valuable effect is a reduction in pre-purchase support contacts and returns for sizing-related reasons. Search improvements increase the proportion of sessions reaching a relevant product page. Northstar decides not to implement a personalized carousel until recommendation quality can be audited for compatibility and diversity.

Learning points

1. The visible conversion problem originated partly in product data and service policy.
2. Clear information can be persuasive without applying pressure.
3. Recommendations require domain constraints, not only behavioral similarity.
4. Guardrail metrics reveal value that a short-term conversion metric may miss.
5. Cross-functional governance is essential because experience crosses organizational boundaries.

10. Limitations, Governance, and Contemporary Developments

Digital experience work faces several limitations. Behavioral data records what people did, not always why they did it. Averages can conceal differences by device, disability, language, geography, or customer expertise. Personalization can improve relevance while reducing comparability and making evaluation more difficult. Experiments can identify local effects that do not generalize to new seasons or markets.

Privacy governance requires data minimization, purpose limitation, appropriate consent where required, access controls, retention limits, and transparent explanations. Organizations should avoid collecting behavioral data merely because it is technically available. Automated recommendations and ranking systems should be monitored for systematic exclusion, proxy discrimination, feedback loops, and unsafe outcomes. Human review is particularly important when systems affect access to credit, essential goods, employment-related equipment, or vulnerable populations.

Generative AI can assist with product-description drafts, query expansion, support responses, attribute extraction, and prototype generation. It can also hallucinate specifications, reproduce bias, expose confidential data, or create inconsistent claims. AI-generated merchandising content therefore requires source verification, human accountability, version control, and clear escalation paths. A product page should never rely on generated text for safety-critical or legally material facts without verification.

Contemporary practice increasingly emphasizes composable commerce, headless interfaces, real-time inventory, conversational search, visual search, and omnichannel identity. These technologies are valuable only when they solve a customer problem reliably. Complexity, latency, vendor dependence, and governance costs must be included in the business case.

Best-Practice Checklist

• Start with customer tasks and evidence rather than interface fashion.
• Maintain a governed product-information model.
• Use labels customers understand and test the information architecture.
• Design search for imperfect language and transparent recovery.
• Explain recommendation rationales and enforce compatibility constraints.
• Present total costs, delivery, returns, and limitations before commitment.
• Treat accessibility as a design requirement, not a final inspection.
• Define primary, diagnostic, and guardrail metrics before experimentation.
• Interpret conversion alongside margin, returns, complaints, and retention.
• Prohibit deceptive urgency, hidden costs, obstructive cancellation, and disguised advertising.
• Minimize personal data and provide meaningful control.
• Document assumptions, tests, decisions, and unintended effects.

Chapter Summary

Digital merchandising is a coordinated system for discovery, evaluation, decision support, and commercial performance. Customer experience design extends beyond individual pages to the complete journey and the organizational capabilities that support it. Journey maps help identify moments of truth, while information architecture, metadata, and search make products findable and understandable. Recommendations can reduce effort, but they require truthful rationales, compatibility controls, diversity, privacy safeguards, and attention to cold-start and feedback-loop problems.

Trust depends on accurate information, transparent terms, reliable operations, authentic reviews, and accessible service. Persuasive design is legitimate when it supports informed choice; it becomes deceptive when it hides consequences or undermines refusal. Measurement should combine outcome, behavioral, diagnostic, and guardrail metrics. A/B testing can support causal inference when randomization, implementation, predefined measures, and ethical constraints are sound, but statistical significance does not substitute for managerial judgment. Finally, AI and personalization should be governed as socio-technical capabilities, with human accountability, data minimization, accessibility, and consumer protection at the center.

Key Terms

A/B testing: Randomized comparison of a control and treatment experience.

Accessibility: Design that enables people with diverse abilities to use a product or service.

Average order value (AOV): Revenue divided by the defined number of orders.

Card sorting: Research method in which participants group and label concepts.

Conversion rate: Defined orders divided by eligible sessions or visitors.

Customer journey map: Representation of stages, touchpoints, needs, emotions, and organizational processes.

Dark pattern: Interface practice that manipulates or obstructs users in ways that undermine informed choice.

Faceted navigation: Filtering system based on product attributes.

Findability: The ease with which users can locate relevant information or products.

Information architecture: Organization and labeling of content, products, and navigation.

Metadata: Structured information describing products and content.

Moment of truth: Touchpoint at which a customer forms a significant judgment.

Personalization: Adaptation of content or experience using customer or contextual information.

Product taxonomy: Structured classification of products and their relationships.

Recommendation system: System that proposes products or content based on signals and rules or models.

Search reformulation: Modification of a query after an unsatisfactory search result.

Trust signal: Evidence that supports judgments of competence, integrity, security, or reliability.

Usability: The effectiveness, efficiency, and satisfaction with which users achieve goals.

Review Questions

5 Short Answer Questions

1. What are the four principal purposes of digital merchandising?
2. Why should a customer journey map include organizational processes as well as customer touchpoints?
3. Distinguish information architecture from product metadata.
4. Identify two risks associated with personalization.
5. Why are guardrail metrics important in experimentation?

5 Long Answer Questions

1. Explain how usability, trust, and customer experience interact in an e-commerce storefront.
2. Compare content-based, collaborative, contextual, and hybrid recommendation approaches.
3. Discuss how search ranking can balance customer relevance with commercial objectives without undermining integrity.
4. Explain the ethical difference between legitimate persuasive design and a dark pattern, using examples.
5. Describe a complete evidence-based process for improving a digital storefront from diagnosis through evaluation.

3 Analytical / Application Questions

1. A retailer has a high add-to-cart rate but a low checkout completion rate. Design a journey-mapping and research plan to identify likely causes, and specify the metrics and guardrails you would use.
2. A fashion store wants to display “Only 1 left—buy now” messages generated from inventory data. Analyze the conditions under which this could be truthful and ethical, and propose governance controls.
3. Design an A/B experiment for improving product discovery through search. State the hypothesis, treatment, control, primary metric, guardrail metrics, key assumptions, and threats to validity.

Multiple Choice Questions with Answers

Question 1

Which statement best describes the primary role of a customer journey map?

A. To replace usability testing with a visual diagram

B. To represent customer stages, touchpoints, needs, and supporting processes

C. To calculate the profitability of each product category

D. To determine the exact algorithm used by a search engine

Correct Answer: B. To represent customer stages, touchpoints, needs, and supporting processes

Question 2

Which information-architecture decision most directly improves product comparison?

A. Adding more promotional banners

B. Hiding product attributes until checkout

C. Standardizing relevant attributes such as size, material, and compatibility

D. Using internal department names as category labels

Correct Answer: C. Standardizing relevant attributes such as size, material, and compatibility

Question 3

A store records 400 clicks from 10,000 impressions. What is its CTR?

A. 0.4%

B. 4%

C. 25%

D. 40%

Correct Answer: B. 4%

Question 4

Which recommendation label most clearly communicates a behavioral rationale?

A. Guaranteed best choice

B. Because you viewed this item

C. Officially superior product

D. Everyone must buy this

Correct Answer: B. Because you viewed this item

Question 5

Which is an example of a guardrail metric?

A. Number of colors in the design system

B. Return rate after a conversion experiment

C. Number of employees in the merchandising team

D. Length of the product title only

Correct Answer: B. Return rate after a conversion experiment

Question 6

Why is a zero-result recovery feature valuable in site search?

A. It prevents all customers from using filters

B. It helps customers continue when their exact query has no match

C. It guarantees that every product will be sold

D. It removes the need for accurate product data

Correct Answer: B. It helps customers continue when their exact query has no match

Question 7

Which practice is most consistent with ethical persuasive design?

A. Showing mandatory fees only after payment

B. Making cancellation harder than subscription enrollment

C. Displaying accurate delivery and return information before commitment

D. Using a false countdown timer to create urgency

Correct Answer: C. Displaying accurate delivery and return information before commitment

Question 8

In the worked example, conversion rises from 3.0% to 3.3%. What is the relative increase?

A. 0.3%

B. 3%

C. 10%

D. 30%

Correct Answer: C. 10%

Question 9

What is the principal causal advantage of random assignment in an A/B test?

A. It guarantees a profitable result

B. It helps balance customer characteristics between groups in expectation

C. It eliminates the need for a primary metric

D. It ensures every customer sees both experiences

Correct Answer: B. It helps balance customer characteristics between groups in expectation

Question 10

Which is an important governance safeguard for AI-generated product descriptions?

A. Allowing generated specifications to publish without review

B. Removing source references to make text shorter

C. Verifying material claims and assigning human accountability

D. Using generated text only for customers who complain

Correct Answer: C. Verifying material claims and assigning human accountability