{ "platformId": "chatgpt-work", "roleSlug": "paralegals-and-legal-assistants", "directAnswer": "For paralegals and legal assistants, ChatGPT Work fits the role's own constitutional structure: everything drafts under supervision, mirroring the attorney-review structure paralegal work is done within. It is strongest on document preparation with every legal assertion flagged for attorney verification, ruthless cite-checking that never vouches from memory, and research memos where every proposition pins to its source span. This occupation also has the widest expert disagreement in our set: published exposure estimates span from very low to very high depending on what each study measured, and usage estimates, projected from economy-wide data rather than directly observed, lean augmentation. The verification discipline is not overhead here; fabricated citations are this profession's best-known AI failure, so the discipline is the product.", "topUseCases": [ { "interventionId": "task-18491:supervised-document-draft:approval-required", "taskId": "task-18491", "task": "Prepare affidavits or other documents, such as legal correspondence, and organize and maintain documents in paper or electronic filing system.", "workflowId": "supervised-document-draft", "authorityMode": "approval-required", "centralityScreen": { "value": 0.20759016979999997, "importance": { "value": 0.845, "sourceField": "role-tasks-kv:task-18491.importance" }, "occupationalRelevance": { "value": 0.9573999999999999, "sourceField": "role-tasks-kv:task-18491.relevance / 100" }, "frequency": { "kind": "modal-response-share", "modalResponseSharePct": 25.66, "collectionDate": "2026-06-03", "sourceField": "role-tasks-kv:task-18491.modalResponseSharePct" } }, "agentFit": { "value": 0.9025, "sourceField": "rule:digital-feasibility-and-capability-fit-single-count.v2" }, "distribution": { "value": 0.9, "sourceField": "roles-kv:aiPlaybook.delegatedTasks[task-18491].observed-distribution" }, "evidenceConfidence": "weak", "withinGateScore": 0.16861511542004998, "gate": { "status": "approval-required", "reason": "A person must approve the verifiable handoff before action." }, "capabilityIds": [ "connected-apps-files", "deliverable-generation", "long-running-projects" ], "expectedArtifact": "Document draft with three inline flag types (FACT, LEGAL ASSERTION, BOILERPLATE) plus its verification checklist", "humanReviewPoint": "Attorney verification of every legal assertion; nothing is filed, sent, or signed by the agent.", "riskLevel": "high" }, { "interventionId": "task-1636:supervised-document-draft:recommend", "taskId": "task-1636", "task": "Gather and analyze research data, such as statutes, decisions, and legal articles, codes, and documents.", "workflowId": "supervised-document-draft", "authorityMode": "recommend", "centralityScreen": { "value": 0.066847470125, "importance": { "value": 0.7225, "sourceField": "role-tasks-kv:task-1636.importance" }, "occupationalRelevance": { "value": 0.8465, "sourceField": "role-tasks-kv:task-1636.relevance / 100" }, "frequency": { "kind": "modal-response-share", "modalResponseSharePct": 10.93, "collectionDate": "2026-06-03", "sourceField": "role-tasks-kv:task-1636.modalResponseSharePct" } }, "agentFit": { "value": 0.9025, "sourceField": "rule:digital-feasibility-and-capability-fit-single-count.v2" }, "distribution": { "value": 0.9, "sourceField": "roles-kv:aiPlaybook.delegatedTasks[task-1636].observed-distribution" }, "evidenceConfidence": "weak", "withinGateScore": 0.054296857609031246, "gate": { "status": "limited", "reason": "Authority remains below an approval-required intervention." }, "capabilityIds": [ "connected-apps-files", "browser-research-actions", "deliverable-generation" ], "expectedArtifact": "Authority map: what supports, what cuts against, what is unsettled, every proposition pinned to its source span", "humanReviewPoint": "The attorney judges the authorities; anything unconfirmable from supplied sources stays marked UNVERIFIED.", "riskLevel": "high" } ], "workflows": [ { "id": "supervised-document-draft", "goal": "Draft a legal document under the role's own supervision structure, with the verification checklist as a first-class artifact beside the draft.", "inputs": [ "Matter context and facts, supplied by you in the interview step", "The document type and its template or exemplar", "The sources the draft may rely on" ], "connectedSystems": [ "Matter document repository", "Template library" ], "steps": [ "Interview for the facts and matter context; the occupational data carries no matter facts and the draft must not invent them.", "Draft with three inline flag types: FACT (user-supplied), LEGAL ASSERTION (needs attorney verification), BOILERPLATE (template origin named).", "Emit the verification checklist alongside the draft, mapping each flagged assertion to who verifies it.", "Stop: nothing is filed, sent, or signed by the agent, and the draft footer says who verifies what." ], "expectedArtifact": "Flagged draft plus verification checklist", "starterPrompt": "Here is the matter context and the document type. Interview me for the facts first. Then draft with every legal assertion flagged for attorney verification, facts marked as supplied by me, and boilerplate marked with its template origin. Produce the verification checklist beside the draft. Nothing you produce gets filed, sent, or signed by you; say so in the footer. If I ask you to just finalize the instrument, decline: execution is attorney territory.", "humanReviewPoint": "This guide treats attorney review as the structure of the work, not an add-on; the checklist routes each assertion to its verifier.", "commonFailure": "A fluent draft whose legal assertions read as settled when no attorney has verified them, or invented matter facts filling an interview gap." }, { "id": "ruthless-cite-check", "goal": "Verify every citation in a draft against supplied sources, marking anything unconfirmable rather than vouching from memory.", "inputs": [ "The draft with citations", "The authorities you can supply as source documents" ], "connectedSystems": [ "Source document repository" ], "steps": [ "For each citation: does the cited authority exist in the supplied sources, does it say what the draft claims, and is there any signal it is no longer good law.", "Mark UNVERIFIED anything not confirmable from the supplied sources; memory is not a source.", "Produce the cite-check table: citation, claim, verification status, source span or UNVERIFIED reason.", "Route the UNVERIFIED list for human retrieval and attorney judgment." ], "expectedArtifact": "Cite-check table with per-citation verification status and source spans", "starterPrompt": "Here is a draft with citations and the authorities I can supply. For each citation: does the authority exist, does it say what we claim, is it still good law as far as these sources show? Mark UNVERIFIED anything you cannot confirm from what I gave you. Never vouch for a citation from memory.", "humanReviewPoint": "UNVERIFIED items go to human retrieval; good-law judgment is the attorney's.", "commonFailure": "Vouching for a citation from training memory instead of the supplied sources; that is this profession's best-known AI failure mode." } ], "setupRecommendations": [ { "item": "Connect the matter repository and template library read-only; no filing system write access, no e-filing, no email.", "rationale": "The setup mirrors this role's supervision structure by making the agent structurally unable to file, send, or sign.", "capabilityIds": [ "connected-apps-files", "approval-enterprise-controls" ] } ], "prompts": [ { "id": "draft-under-supervision", "title": "Draft a document under supervision", "prompt": "Here is the matter context and the document type. Interview me for the facts, draft with every legal assertion flagged for attorney verification, and produce the verification checklist alongside the draft.", "expectedArtifact": "Flagged draft plus verification checklist", "humanReviewPoint": "Attorney verifies every flagged assertion before the document moves." }, { "id": "cite-check", "title": "Cite-check ruthlessly", "prompt": "Here is a draft with citations. For each: does the cited authority exist, does it say what we claim, is it still good law? Mark UNVERIFIED anything you cannot confirm from the sources I provide. Never vouch from memory.", "expectedArtifact": "Cite-check table", "humanReviewPoint": "UNVERIFIED items route to human retrieval and attorney judgment." }, { "id": "research-memo", "title": "Build a research memo with receipts", "prompt": "Question: [issue]. From the sources I give you, build the authority map: what supports, what cuts against, what is unsettled. Pin every proposition to its source span.", "expectedArtifact": "Authority map with source-span citations", "humanReviewPoint": "The legal conclusion is the attorney's; the map is evidence." }, { "id": "filing-order", "title": "Bring order to the filing system", "prompt": "Here is my matter list and document pile. Propose the organization scheme, draft the index, and flag documents whose retention or privilege status needs attorney review.", "expectedArtifact": "Organization scheme, index draft, and review flags", "humanReviewPoint": "Privilege and retention calls are attorney decisions." }, { "id": "deadline-sweep", "title": "Sweep the matter for dates and deadlines", "prompt": "From these matter documents, extract every date, deadline, and obligation with its source span. Separate hard deadlines from soft ones as the documents state them; do not infer deadlines the documents do not contain. I verify against the court rules.", "expectedArtifact": "Dated obligations table with source spans", "humanReviewPoint": "You verify against controlling rules; the table cites, it does not conclude." }, { "id": "honest-outlook", "title": "Give me the honest read on this occupation", "prompt": "One published study puts this role's exposure very low and another very high. Explain the spread: what did each study actually measure? Then what do the augmentation-leaning usage estimates and heavy document-prep share suggest about how the work is really changing? Keep the studies separate.", "expectedArtifact": "Honest occupational outlook with studies kept separate", "humanReviewPoint": "Career decisions stay yours; this is orientation, not advice." } ], "humanOwnedResponsibilities": [ "Attorney verification of every legal assertion: this guide mirrors the supervision structure paralegal work is done within, and the agent's drafts live inside it", "Filing, sending, signing, and executing instruments: the agent is structurally unable, by setup, not just instruction", "Good-law judgment and legal conclusions: maps and tables are evidence for the attorney, never the holding", "Privilege and retention decisions", "Anything a citation rests on that the supplied sources cannot confirm" ], "risks": [ { "id": "hallucinated-authority", "level": "high", "description": "A citation vouched from memory rather than sources, or a legal assertion reading as settled without attorney verification; this profession's signature AI failure mode is exactly this.", "mitigation": "Verification-first discipline: three-flag drafting, UNVERIFIED as a mandatory status, checklists as first-class artifacts, and memory never counted as a source." } ], "sources": [ { "id": "occupation-paralegals-and-legal-assistants", "title": "Paralegals and Legal Assistants occupational render model", "url": "https://singulariki.com/roles/paralegals-and-legal-assistants", "publisher": "gold.onet_role_starter_surface", "lastReviewedAt": "2026-06-03" }, { "id": "openai-chatgpt-work", "title": "ChatGPT Work and Codex", "url": "https://help.openai.com/en/articles/20001275-chatgpt-work-and-codex", "publisher": "OpenAI", "lastReviewedAt": "2026-07-11" }, { "id": "openai-work-deliverables", "title": "Creating and editing documents, spreadsheets, and presentations with ChatGPT Work", "url": "https://help.openai.com/en/articles/20001278-creating-and-editing-documents-spreadsheets-and-presentations-with-chatgpt-work", "publisher": "OpenAI", "lastReviewedAt": "2026-07-11" }, { "id": "openai-work-cloud-browser", "title": "Using cloud browser in ChatGPT", "url": "https://help.openai.com/en/articles/20001280-using-cloud-browser-in-chatgpt", "publisher": "OpenAI", "lastReviewedAt": "2026-07-11" }, { "id": "openai-work-desktop-browser", "title": "Using the built-in browser in the ChatGPT desktop app", "url": "https://help.openai.com/en/articles/20001277-using-the-built-in-browser-in-the-chatgpt-desktop-app", "publisher": "OpenAI", "lastReviewedAt": "2026-07-11" }, { "id": "openai-chatgpt-skills", "title": "Skills in ChatGPT", "url": "https://help.openai.com/en/articles/20001066", "publisher": "OpenAI", "lastReviewedAt": "2026-07-11" }, { "id": "openai-chatgpt-scheduled-tasks", "title": "Scheduled Tasks in ChatGPT", "url": "https://help.openai.com/en/articles/10291617-tasks-in-chatgpt", "publisher": "OpenAI", "lastReviewedAt": "2026-07-11" }, { "id": "openai-workspace-agents", "title": "ChatGPT Workspace Agents for Enterprise and Business", "url": "https://help.openai.com/en/articles/20001143", "publisher": "OpenAI", "lastReviewedAt": "2026-07-11" } ], "freshness": { "platformVersion": "2026-07-09-launch", "platformReviewedAt": "2026-07-11", "occupationalDataVersion": "O*NET 30.3", "occupationalDataGeneratedAt": "2026-06-03T05:05:52.989984+00:00" }, "quality": { "uniqueOccupationalOrTaskEvidence": false, "validPlatformSpecificWorkflow": true, "clearOutputArtifact": true, "realCapabilityMatch": true, "meaningfulHumanReviewSection": true, "sourcesAndFreshness": false, "distinctInternalRelationships": true, "noUnresolvedFactualOrProductVersionFailure": false, "indexable": false } }