PM or coder looking for your next role? Build an ATS-friendly resume designed to get interview calls.

AI Product Requirements · September 4, 2026 · 13 min read

AI Feature PRD Template: Evals, Failure Modes, Guardrails and Acceptance Criteria

Use this copy-ready template to define what an AI feature should do, how its variable behavior will be evaluated, what can go wrong, which controls must intervene, and what evidence is required before release.

An AI feature PRD needs more than a user story and a happy path. It should define:

  1. the customer problem and intended outcome;
  2. the model’s role and the deterministic product behavior around it;
  3. representative eval cases, metrics, and release thresholds;
  4. expected failure modes and user-visible recovery;
  5. input, output, retrieval, tool, and human-approval guardrails;
  6. testable acceptance criteria;
  7. monitoring, staged rollout, rollback, and ownership.

The template below can be adapted for copilots, summarizers, recommendation systems, classifiers, conversational experiences, retrieval-augmented generation, and agents that take actions.

Why an AI Feature Needs a Different PRD

A conventional feature is often expected to produce the same output from the same input. An AI feature may produce different acceptable outputs, degrade on unfamiliar cases, depend on retrieved context, or make a confident statement that is not supported.

The PRD therefore has to define a behavioral envelope rather than one exact answer:

Evals, Failure Modes, Guardrails, and Acceptance Criteria

LayerQuestion it answersExample
EvalsHow will we measure variable AI behavior across representative cases?A labeled test set measures groundedness, task completion, and harmful-response rate before release.
Failure modesHow can the model, context, workflow, or user interaction fail?The assistant invents a refund policy when the relevant policy document is missing.
GuardrailsWhat enforced control prevents, contains, or escalates the failure?If no approved policy source is retrieved, the system abstains and routes the user to support.
Acceptance criteriaWhat observable behavior must the complete product experience satisfy?Every policy answer displays its source; an unsupported answer is not shown to the user.

These layers should connect. A high-severity failure mode should have a prevention or containment guardrail, a verification method, an owner, and an acceptance decision.

What the AI Feature PRD Should Contain

Problem, evidence, and outcome

Name the user, problem, supporting evidence, current workaround, desired outcome, and why AI is appropriate. Include a non-AI baseline where useful; the team should be able to explain why variable model behavior creates enough value to justify its cost and risk.

User workflow and model role

Describe the complete experience around the model: what triggers it, what context it receives, what it produces, what the user can inspect or edit, what action follows, and what happens if the model is unavailable.

Separate probabilistic work from deterministic product behavior. For example, drafting a response may be probabilistic; checking permissions before sending it must be deterministic.

Data and context boundaries

Specify allowed data sources, prohibited data, retention expectations, sensitive fields, tenant boundaries, freshness requirements, source precedence, and behavior when relevant context is missing or conflicting.

Evaluation plan

Define the cases, metrics, grading method, thresholds, and evaluation cadence. A useful plan includes:

Failure-mode register

For each failure, record the trigger, affected user, severity, detectability, prevention, containment, fallback, telemetry, and owner.

Failure categoryExamplePossible product response
Unsupported outputAnswer is not grounded in an approved sourceAbstain, identify missing evidence, and offer a safe next step
Wrong contextContent from another workspace is retrievedEnforce tenant boundaries and block the response
Ambiguous inputUser request could mean two materially different actionsAsk a clarifying question before proceeding
Tool failureDownstream action times out after an uncertain resultDo not retry blindly; check status and show a recoverable state
Unsafe actionAgent attempts an irreversible change without approvalRequire explicit human confirmation and narrow permissions
Degraded serviceModel or retrieval provider is unavailableUse an approved fallback or disable the AI path gracefully

Guardrails and human escalation

Write guardrails as enforceable controls, not aspirations. Each guardrail should state what triggers it, what the system does, what the user sees, how it is tested, and who owns it.

Acceptance criteria

Acceptance criteria cover the whole product behavior, not only the model response. Include access, context retrieval, output presentation, user control, failure recovery, logging, accessibility, and fallback behavior.

Prefer observable statements such as: “When no approved source supports the answer, the assistant states that it cannot answer and does not display a fabricated citation.”

Monitoring, rollout, and rollback

Define what the team will observe after release, which cohorts receive the feature first, what triggers a pause, how rollback works, and who responds. Track both product outcomes and guardrail health; a feature can improve engagement while increasing unsafe or unsupported outputs.

Copy-and-Paste AI Feature PRD Template

Replace the bracketed instructions. Delete sections only after deciding they are not relevant to the feature.

# AI Feature PRD: [Feature name] ## 1. Document control - Owner: [Name / role] - Engineering owner: [Name / role] - Risk or policy owner: [Name / role] - Status: [Draft / Review / Approved] - Version: [Version] - Last updated: [Date] - Linked evidence: [Sources] ## 2. Customer problem and evidence - Target user: [Who] - Situation: [When / where the problem occurs] - Problem: [Observable problem] - Evidence: [Tickets, interviews, usage, research] - Current workaround: [What users do today] - Assumptions: [What remains unproven] ## 3. Desired outcome - User outcome: [What becomes possible or easier] - Business outcome: [Why it matters] - Non-AI baseline: [Simpler alternative considered] - Why AI is appropriate: [Value created by AI behavior] ## 4. Scope ### In scope - [Behavior] ### Out of scope - [Explicit exclusion] ## 5. User workflow 1. [Trigger] 2. [Context collected or retrieved] 3. [Model task] 4. [Output shown] 5. [User review, edit, approval, or action] 6. [Confirmation, logging, or follow-up] ## 6. Model role and system boundary - Model responsibility: [Classify / draft / summarize / recommend / act] - Deterministic system responsibility: [Permissions, validation, calculations] - Model must not: [Forbidden decisions or actions] - User control: [Inspect / edit / reject / retry / override] - Fallback when AI is unavailable: [Behavior] ## 7. Data and context - Allowed inputs: [Data] - Prohibited inputs: [Data] - Approved knowledge sources: [Sources] - Source precedence: [Which source wins] - Freshness requirement: [Maximum age] - Tenant and permission boundary: [Rule] - Retention and deletion: [Rule] - Behavior when context is missing or conflicting: [Abstain / clarify / escalate] ## 8. Evaluation plan ### Evaluation dataset - Source: [Production samples / synthetic / expert-authored] - Segments: [Common / difficult / ambiguous / adversarial / out-of-scope] - Size and refresh cadence: [Define] - Privacy review: [Owner / status] ### Quality and safety metrics | Metric | Measurement method | Release threshold | Regression limit | Owner | |---|---|---|---|---| | [Groundedness] | [Human rubric / automated check] | [Risk-based target] | [Limit] | [Owner] | | [Task success] | [Method] | [Target] | [Limit] | [Owner] | | [Safety failure rate] | [Method] | [Target] | [Limit] | [Owner] | | [Latency / cost] | [Method] | [Target] | [Limit] | [Owner] | ### Zero-tolerance cases - [Case that must never pass] ### Review policy - Failed-case review: [Process] - Residual-risk approver: [Role] - Re-evaluation triggers: [Model, prompt, retrieval, policy, or data change] ## 9. Failure modes | Failure | Trigger | Severity | Detection | Prevention / containment | User fallback | Owner | |---|---|---|---|---|---|---| | [Failure] | [Condition] | [Low / Medium / High / Critical] | [Signal] | [Control] | [Experience] | [Owner] | ## 10. Guardrails | Guardrail | Trigger | Enforced response | User-visible behavior | Verification | Owner | |---|---|---|---|---|---| | [Control] | [Condition] | [Block / redact / abstain / ask / escalate] | [Message or flow] | [Test or eval] | [Owner] | ## 11. Human review and escalation - Human approval required when: [Conditions] - Escalation destination: [Team / queue] - Information shown to reviewer: [Context] - Override permissions: [Roles] - Audit record: [Fields] ## 12. Acceptance criteria - Given [state], when [action], then [observable result]. - Given [missing context], when [request], then [safe fallback]. - Given [unauthorized user], when [action], then [blocked behavior]. - Given [provider failure], when [request], then [recoverable experience]. - Given [high-impact action], when [agent proposes it], then [human approval is required]. ## 13. Observability - Product outcome metrics: [Metrics] - Quality metrics: [Metrics] - Guardrail metrics: [Triggers, blocks, escalations] - Operational metrics: [Latency, availability, cost] - Feedback capture: [Mechanism] - Alert thresholds and owner: [Define] ## 14. Rollout and rollback - Initial cohort: [Users / percentage] - Entry criteria: [Required evidence] - Expansion criteria: [Metrics and review] - Pause triggers: [Conditions] - Rollback mechanism: [Feature flag / provider fallback / disable action] - Rollback owner: [Role] ## 15. Open decisions | Decision | Options | Owner | Due date | Blocking? | |---|---|---|---|---| | [Question] | [Options] | [Owner] | [Date] | [Yes / No] | ## 16. Release evidence - Eval report: [Link] - Acceptance-test evidence: [Link] - Security / privacy review: [Link] - Human-review signoff: [Link] - Rollback rehearsal or verification: [Link]

Filled Example: AI Support Reply Drafts

Problem and outcome

Support agents spend time searching approved policy documents and rewriting similar replies. The feature should draft a grounded response that the agent can inspect and edit, reducing handling time without sending unsupported policy claims.

Model and user boundary

Representative eval segments

Illustrative release criteria

The product, support, security, and engineering owners should set thresholds based on their own risk and baseline. An illustrative policy could require:

Example failure-to-guardrail mapping

Failure modeGuardrailAcceptance evidence
Draft invents a refund ruleRequire an approved source for every policy claim; otherwise abstainUnsupported-policy eval cases produce an abstention with no fabricated source
Context belongs to another tenantAuthorization-filter retrieval before model context is assembledCross-tenant tests return no content and record a blocked attempt
Agent submits an unreviewed draftNo send tool is available to the model; user action is requiredIntegration tests confirm drafts cannot be sent through the AI path
Retrieval provider is unavailableDisable drafting and show the manual workflowFailure simulation shows a clear fallback without a blank or invented response

How to Write Acceptance Criteria for AI Features

Write criteria for observable product behavior under a defined state. Avoid requiring an exact sentence unless exact text is a policy requirement. Evaluate the quality envelope separately across a representative dataset.

WeakBetter
The assistant provides accurate answers.When an approved source supports the answer, the response identifies that source; when no approved source is available, the assistant abstains and offers the escalation path.
The summary is high quality.The summary preserves the named decision, owner, deadline, and unresolved questions across the approved evaluation set without adding unsupported commitments.
The agent is safe.When a requested action exceeds the user’s permission or requires human approval, the tool call is blocked and the interface explains the required next step.
The AI is fast.The interface displays a progress state within the defined response window and provides the approved retry or fallback behavior when the operation exceeds it.

For additional patterns, use the guide to writing acceptance criteria for AI.

AI Feature PRD Review Checklist

Common AI PRD Mistakes

  1. Writing “use AI” as the solution. Define the user outcome and compare it with a simpler baseline.
  2. Evaluating only average quality. A strong average can hide failure for a critical segment or safety case.
  3. Using one metric for every risk. Helpfulness, groundedness, safety, latency, and cost answer different questions.
  4. Treating prompt instructions as guardrails. High-impact controls should be enforced outside model discretion where feasible.
  5. Ignoring the surrounding workflow. Access control, user review, fallback, logging, and recovery are product requirements too.
  6. Launching without a re-evaluation policy. Model, prompt, data, retrieval, and tool changes can alter behavior.
  7. Hiding open decisions in prose. Give each decision an owner, deadline, and blocking status.

Frequently Asked Questions

What should an AI feature PRD include?

Include the normal product decision plus the model role, data boundaries, eval plan, failure modes, guardrails, human escalation, acceptance criteria, observability, rollout, rollback, and release evidence.

What is the difference between evals and acceptance criteria?

Evals measure variable AI behavior across a representative collection of cases. Acceptance criteria define the observable conditions the complete feature and workflow must satisfy. Teams usually need both.

What makes a guardrail testable?

Name the trigger, enforced intervention, user-visible behavior, telemetry, and expected evidence. “Avoid harmful output” is an aspiration; “block the response and escalate when this classifier and policy condition are met” can be verified.

Who owns AI feature quality?

Ownership is shared but should not be vague. Product owns intended behavior and outcome; engineering owns implementation and operational quality; domain, risk, security, legal, or policy owners approve relevant boundaries; named individuals own monitoring and incident response.

Can a team reuse one eval dataset forever?

No. Add production failures, new user segments, changing policies, adversarial cases, and newly discovered edge cases. Protect a stable regression set while evolving the broader evaluation suite.