AI-assisted return claims are reshaping ecommerce fraud controls

At a glance

What this is: Riskified’s report shows gen AI is now routinely used to draft return claims, making polished language a weaker legitimacy signal.

Why it matters: Fraud, identity verification, and ecommerce risk teams need to move from claim-text screening toward behavioural, transaction, and account-history signals that better separate honest customers from abuse.

By the numbers:

• 50% of consumers have used gen AI tools to help draft return or refund claims.
• 85% of consumers surveyed see at least some form of strategic or deceptive return behavior as normal.
• 51% of global consumers have encountered return-related content on social platforms.

👉 Read Riskified’s 2026 report on AI-assisted return claims and fraud control

Context

AI-assisted return claims are changing ecommerce fraud screening because the language of a claim is no longer a dependable indicator of intent. In practice, merchants now have to judge whether a return request is honest, exaggerated, or machine-assisted, which pushes fraud operations closer to identity verification and behavioural risk scoring than simple text review.

That matters because return abuse sits at the boundary between fraud, trust and safety, and identity governance. When consumers can use AI to generate plausible narratives, merchants need controls that assess historical behaviour, account reputation, and transaction context instead of treating polished wording as proof of legitimacy.

Key questions

Q: What breaks when return claims can be written by AI?

A: What breaks first is the assumption that polished wording or emotional detail correlates with honesty. AI can make weak claims sound credible, so merchants need to rely more on behavioural identity, account history, and transaction context. The practical failure is over-trusting the message and under-weighting the requester’s history.

Q: Why do AI-assisted return claims complicate fraud detection?

A: They complicate fraud detection because they improve the quality and consistency of deceptive claims without changing the underlying pattern of abuse. That reduces the value of manual reading and keyword-based checks. Teams should expect stronger narratives, manipulated images, and repeatable scripts to blend into normal customer traffic.

Q: How do merchants know if return controls are actually working?

A: They should look for lower abuse rates without a matching rise in false positives, customer complaints, or abandonment from legitimate shoppers. Useful signals include repeat-offender suppression, stable approval rates for low-risk customers, and reduced refund leakage on high-risk items. If friction rises everywhere, the controls are too blunt.

Q: Who is accountable for AI-assisted return fraud decisions?

A: Fraud, ecommerce, and trust and safety leaders share accountability because the problem spans policy design, customer experience, and risk analytics. If personal data is used for profiling or automated decisioning, privacy and governance teams should also review the controls. Accountability should sit with the owner of the return decisioning workflow.

Technical breakdown

AI-assisted return claims weaken text-based fraud screening

Generative AI makes it easy to produce fluent, emotionally persuasive return narratives that look consistent even when the underlying claim is weak. That breaks older screening approaches that implicitly trusted wording quality, grammar, or tone as weak indicators of legitimacy. The risk is not only synthetic text, but also manipulated images, coordinated scripts, and template-based deception scaled through consumer-facing AI tools. In ecommerce, the signal has shifted from how the claim reads to how the account behaves over time.

Practical implication: fraud teams should de-emphasise claim language and build decisioning around account history, order patterns, and prior return behaviour.

Return abuse is now a behavioural risk problem

The report’s strongest point is that deceptive return behaviour has become normalised for a large share of consumers. That means merchants are not only detecting bad claims, they are operating in an environment where social proof, peer learning, and AI assistance lower the friction for abuse. This makes traditional binary rules too blunt. The more effective model is layered risk scoring that combines customer lifetime value, return frequency, dispute history, item type, and channel context.

Practical implication: use segmented controls that apply friction only where behavioural evidence justifies it.

Identity intelligence is becoming the better fraud signal

Identity intelligence in this context does not mean only verified identity documents. It means the broader set of signals that show whether a requester is acting like a known honest customer or a repeat abuser. That includes account age, payment consistency, device continuity, shipping patterns, and prior interactions. For merchants, this is a trust decision under uncertainty, and AI simply makes that uncertainty more visible. The governance challenge is to apply stronger controls without punishing legitimate shoppers.

Practical implication: integrate behavioural and account-level identity signals into return workflows before claims reach manual review.

Threat narrative

Attacker objective: The objective is to obtain refunds, store credit, or replacements for returns that would otherwise be challenged or denied.

1. Entry begins when a consumer uses gen AI to draft a return or refund claim that sounds credible and policy-aligned.
2. Escalation follows when the same tools are used to create manipulated photos, stronger narratives, or coordinated claim patterns that bypass simple review.
3. Impact is fraudulent refund leakage, operational waste, and rising friction for honest customers as merchants tighten controls.

NHI Mgmt Group analysis

AI-assisted return fraud creates an identity verification gap: merchants are no longer validating only the item claim, they are validating the credibility of the requester under machine-assisted persuasion. That changes the problem from simple disputes to trust assessment, where behavioural history matters more than polished language. In practice, fraud and identity teams should treat return flows as a governed trust journey, not a text review queue.

Return abuse has become normalised faster than merchant controls have adapted: when 85% of consumers see some deceptive behaviour as normal, policy design has to assume partial adversarial pressure. That does not justify blanket suspicion, but it does mean controls must distinguish between first-time customers and repeat abusers. The right governance model is selective friction, not universal hardening.

Behavioral identity is now the most useful named concept for returns: the article shows that order history, prior returns, and account context outperform claim text as a legitimacy signal. That is a practical example of behavioral identity governance, where the merchant evaluates what the customer has done over time rather than what the customer can say in one interaction. For fraud and trust teams, this is the control shift that matters.

Strategic return behaviour should be treated as a trust and safety problem with fraud outcomes: the social media layer means abusive tactics spread through shared playbooks, not isolated intent. That makes the issue harder to solve with static policy language alone. Teams need governance that combines fraud analytics, customer experience thresholds, and clear escalation paths so legitimate shoppers are not pushed into the same controls as repeat bad actors.

AI-generated claims expose the limits of manual review at scale: the more fluent and image-rich the claim becomes, the more review teams are pushed into subjective judgment. Subjectivity is expensive, inconsistent, and easy to overwhelm. Practitioners should respond by strengthening pre-dispute controls and evidence-based decisioning rather than relying on reviewer intuition.

What this signals

Behavioral identity governance is becoming a practical fraud requirement: as AI makes claim text easy to synthesise, merchants will need stronger identity context around each return request. That means linking fraud controls to account reputation, payment continuity, and prior dispute behaviour instead of relying on narrative quality. For teams building policy, the next step is to treat return flows as a governed trust decision and not just an operations workflow.

The operational question is not whether AI-assisted claims will exist, but how much friction merchants can add before honest customers feel punished. That pushes practitioners toward segmented policy design, transparent thresholds, and better use of evidence. The merchants that succeed will be the ones that can distinguish abuse from legitimate dissatisfaction quickly enough to protect margin without damaging trust.

For practitioners

• Shift return screening to behavioural identity signals Weight account age, prior return rate, payment consistency, device continuity, and shipping history more heavily than claim wording or tone.
• Apply selective friction by risk tier Use tiered policies so high-risk claims trigger rejection or verification, medium-risk claims get added disclosures, and low-risk claims are accepted with monitoring.
• Tighten evidence requirements for high-value returns Require stronger proof for expensive items, unusual damage claims, or customers with a pattern of prior abuse, while leaving ordinary shoppers on the fast path.
• Separate honest-customer experience from abuse controls Design return policies that preserve convenience for low-risk customers and apply additional checks only where the behavioural evidence supports it.

Key takeaways

• AI now helps consumers write return claims that look credible, which weakens text-based fraud screening.
• The scale matters because the problem is normalising, with most consumers accepting at least some deceptive return behaviour.
• The most effective response is behavioural identity scoring, selective friction, and stronger evidence requirements for high-risk cases.

Key terms

• Behavioral Identity: Behavioral identity is the profile formed by a user’s observed patterns over time, such as purchase history, return frequency, device continuity, and payment consistency. It is often more reliable than the wording of a single interaction because it captures repeated behaviour instead of one persuasive request.
• Return Fraud: Return fraud is the abuse of merchant return policies to obtain refunds, replacements, or store credit without a legitimate basis. It can include exaggerated damage claims, item substitution, wardrobing, and AI-assisted narratives that make a weak claim appear credible to review teams.
• Selective Friction: Selective friction is a control strategy that adds verification or delay only where risk signals justify it. In returns, it preserves a smooth experience for low-risk customers while using tighter checks, evidence demands, or manual review for claims with stronger abuse indicators.

What's in the full report

Riskified's full analysis covers the operational detail this post intentionally leaves for the source:

• The survey methodology across 2,091 consumers in seven markets and the senior retail leader interview set.
• The merchant decision patterns behind narrowing return reasons, shortening windows, and shifting toward store credit.
• The forum-reported examples of manipulated damage photos and step-based return screening policies.
• The regional differences in consumer exposure to return-related social content and how that shapes abuse risk.

👉 The full Riskified report covers consumer behaviour, merchant policy changes, and regional return abuse patterns.

Deepen your knowledge

NHI Mgmt Group covers identity security, NHI governance, and agentic AI through independent research, practitioner guides, and the NHI Foundation Level course, the industry's only accredited NHI security programme. Explore the course if your programme needs a stronger control model for identity, access, and machine-led behaviour.