{ "artifactVersion": "2026-07-26.1", "status": "preregistered_not_run", "frozenAt": "2026-07-26", "primaryQuestion": "When the task, source input, target model, settings, and acceptance checks are held constant, which prompt rewrites produce work that satisfies more of the declared requirements?", "tracks": { "optimizer": { "status": "approved_for_access_screen", "startingArtifact": "existing rough prompt", "resultArtifact": "rewritten prompt", "fixturePath": "docs/analysis/prompt-content-growth-2026-07-25/prompt-lab-examples-fixtures.json", "inputPath": "docs/analysis/prompt-content-growth-2026-07-25/prompt-lab-examples-inputs.json", "caseIds": [ "marketing_launch_announcement", "coding_scoped_bug_fix", "coding_json_export", "extraction_invoice_json", "extraction_support_routing", "data_weekly_conversion", "data_anomaly_summary", "research_provider_comparison", "research_evidence_table", "writing_executive_summary", "writing_localized_release_note", "operations_meeting_actions" ], "targetExecution": { "model": "openai/gpt-5.6-luna", "providerRoute": "OpenRouter", "runsPerPromptArtifact": 3, "temperature": null, "reasoning": false, "zeroDataRetentionRequired": true, "dataCollection": "deny", "outputTokenLimit": 4000 } }, "generator": { "status": "deferred_separate_protocol", "startingArtifact": "structured brief", "resultArtifact": "new prompt", "reason": "A generator begins with a brief, while an optimizer begins with an existing prompt. Combining them would score different product jobs as if they were interchangeable." } }, "serpSelection": { "source": "DataForSEO Google Organic Live Advanced", "location": "United States", "language": "English", "device": "desktop", "searchEngine": "google", "retrievedAt": "2026-07-26", "depth": 20, "queries": [ { "keyword": "ai prompt optimizer", "taskId": "07262354-2069-0139-0000-b6de8a9803fe", "eligibleToolResults": [ { "rank": 1, "domain": "promptoptimizer.tools", "url": "https://promptoptimizer.tools/" }, { "rank": 4, "domain": "tripleten.com", "url": "https://tripleten.com/tools/prompt-optimizer/" }, { "rank": 5, "domain": "sitegpt.ai", "url": "https://sitegpt.ai/tools/ai-prompt-optimizer" }, { "rank": 8, "domain": "aiven.io", "url": "https://aiven.io/tools/prompt-optimizer" } ] }, { "keyword": "ai prompt generator", "taskId": "07262354-2069-0139-0000-911c22476240", "eligibleToolResults": [ { "rank": 1, "domain": "quillbot.com", "url": "https://quillbot.com/ai-writing-tools/ai-prompt-generator" }, { "rank": 3, "domain": "feedough.com", "url": "https://www.feedough.com/ai-prompt-generator/" }, { "rank": 4, "domain": "generateprompt.ai", "url": "https://generateprompt.ai/en" }, { "rank": 5, "domain": "tripleten.com", "url": "https://tripleten.com/tools/prompt-generator/" } ], "trackUse": "generator discovery only; excluded from the optimizer score" }, { "keyword": "prompt improver", "taskId": "07262354-2069-0139-0000-cddb4bd2d8ff", "eligibleToolResults": [ { "rank": 1, "domain": "promptoptimizer.tools", "url": "https://promptoptimizer.tools/" }, { "rank": 2, "domain": "copy.ai", "url": "https://www.copy.ai/tools/ai-prompt-improver" }, { "rank": 3, "domain": "maxai.co", "url": "https://www.maxai.co/ai-tools/ai-writer/prompt-enhancement/" }, { "rank": 4, "domain": "sitegpt.ai", "url": "https://sitegpt.ai/tools/ai-prompt-optimizer" }, { "rank": 5, "domain": "customgpt.ai", "url": "https://customgpt.ai/improve-my-prompt/" }, { "rank": 8, "domain": "promptperfect.jina.ai", "url": "https://promptperfect.jina.ai/" }, { "rank": 9, "domain": "rightblogger.com", "url": "https://rightblogger.com/tool/prompt-improver" } ] } ], "excludedResultTypes": [ "provider documentation without a public rewrite interface", "source-code repositories", "discussion and video pages", "browser extensions", "tools that only generate image prompts" ] }, "optimizerCohort": { "baseline": { "id": "benchlm", "url": "https://benchlm.ai/tools/prompt-optimizer", "accessStatus": "included_first_party_baseline" }, "competitorCap": 3, "minimumCompetitorsToRun": 2, "priorityRule": "Use the lowest observed organic rank across the two optimizer-intent queries. Break ties with ai prompt optimizer rank, then domain alphabetically.", "candidates": [ { "priority": 1, "id": "promptoptimizer-tools", "domain": "promptoptimizer.tools", "accessStatus": "pending_access_review" }, { "priority": 2, "id": "copy-ai", "domain": "copy.ai", "accessStatus": "pending_access_review" }, { "priority": 3, "id": "maxai", "domain": "maxai.co", "accessStatus": "pending_access_review" }, { "priority": 4, "id": "tripleten", "domain": "tripleten.com", "accessStatus": "pending_access_review" }, { "priority": 5, "id": "sitegpt", "domain": "sitegpt.ai", "accessStatus": "pending_access_review" }, { "priority": 6, "id": "customgpt", "domain": "customgpt.ai", "accessStatus": "pending_access_review" }, { "priority": 7, "id": "aiven", "domain": "aiven.io", "accessStatus": "pending_access_review" }, { "priority": 8, "id": "promptperfect", "domain": "promptperfect.jina.ai", "accessStatus": "pending_access_review" }, { "priority": 9, "id": "rightblogger", "domain": "rightblogger.com", "accessStatus": "pending_access_review" } ], "eligibilityChecks": [ "accepts an existing text prompt rather than only a blank brief", "returns a prompt artifact that can be copied without treating a model answer as the rewrite", "is publicly usable through anonymous access or an already-authorized account without purchasing a plan", "does not require bypassing authentication, rate limits, bot controls, or product restrictions", "can accept the frozen synthetic fixtures under its published terms and visible input limits" ], "freezeRule": "Complete and record the access screen for every candidate before collecting any scored prompt artifact. Select the first three eligible candidates. Do not replace a selected tool after seeing its outputs.", "unavailableRule": "Record unavailable, blocked, or incompatible tools with the observed reason. Never assign a zero." }, "collection": { "method": "manual public-interface collection", "promptArtifactsPerTool": 12, "targetRunsPerTool": 36, "order": "Use a seeded shuffled case order recorded before the first collection session. Rotate the starting tool by case.", "artifactCapture": [ "tool id and public URL", "collection timestamp", "visible model or mode selection", "input character count", "rewritten prompt text", "latency measured from submit to usable result", "visible error, limit, or retry state", "manual versus automated collection" ], "prohibited": [ "bypassing authentication, rate limits, CAPTCHAs, or bot controls", "creating paid accounts or purchasing plans without separate approval", "changing a frozen task after viewing a tool output", "quietly retrying a failed tool response", "sending private or user-provided content" ] }, "evaluation": { "primaryUnit": "case-by-tool", "primaryOutcome": "share of target-model runs that fully satisfy all frozen automatic checks", "secondaryOutcomes": [ "checks passed out of checks available", "partial, full, and failed result counts", "prompt requirement preservation", "tool error and unavailable rates", "prompt-artifact latency", "target-model latency and provider cost" ], "blinding": "Replace tool and variant names with random artifact ids before target execution and any human review. Keep the key sealed until automatic checks and review decisions are frozen.", "subjectiveBoundary": "Automatic checks are primary. Any claim about clarity, persuasiveness, tone, or task focus remains secondary until a documented blinded human rubric is completed.", "reportingRules": [ "publish task-level results, ties, failures, and regressions", "separate prompt-tool effects from target-model effects", "report the selected cohort and every screened exclusion", "do not combine generator and optimizer tracks", "do not claim statistical significance from this sample", "do not call a tool best overall" ] }, "stopConditions": [ "fewer than two competitor tools pass the frozen access screen", "the frozen target-model route or privacy settings are unavailable", "a fixture or automatic check changes after scored collection starts", "the collection process would require bypassing a product restriction", "the study cannot publish complete prompt artifacts and material failures" ] }