{"id":"search-fin-ap-automation-findings-confidence","audience":"finance","category":"ai-search","track":"research","technique":"verification","title":"Rate a deep-research report's ROI findings by confidence","free":false,"task":"A deep-research tool gave you 4 findings about invoice-automation software before a budget request. Label each High, Medium or Low using your rules: High = 2 or more independent sources from 2024 or later; Medium = exactly 1 independent source from 2024 or later; Low = only older sources or only vendor-written sources. Write exactly 4 numbered lines, each with the label and the source count, years and whether any are vendor-written. Plain text, under 110 words.","materialLabel":"The deep-research findings and their sources (attached to your prompt automatically)","material":"DEEP-RESEARCH REPORT: \"Does AP automation pay off for mid-size companies?\"\n\nFinding 1: Automation cuts cost per invoice by about half.\nSources: Controllers Council benchmark (2025); Hartwell University study (2024).\n\nFinding 2: Teams process invoices 3x faster.\nSources: InvoiceFlow customer case study (2025) - InvoiceFlow sells AP software.\n\nFinding 3: Late-payment fees drop sharply.\nSources: Finance Ops Journal (2024).\n\nFinding 4: Duplicate payments fall to near zero.\nSources: PayPilot white paper (2023) - PayPilot sells AP software; Okoro, Accounting Review (2020).\n\nReport summary (written by the tool): \"Strong, consistent evidence supports AP automation across all areas.\"\n\nOur context: 4-person AP team, ~3,000 invoices a month, budget request due Nov 20.","rules":[{"type":"numberedItems","count":4},{"type":"maxWords","max":110},{"type":"plainText"}],"mistakes":[{"title":"Trusting the report's summary","why":"The AI repeats \"strong, consistent evidence\" because the report says so.","fix":"Say: \"Rate each finding yourself using my rules.\"","checks":["r2"]},{"title":"Not giving your rules","why":"The AI rates a vendor case study as solid because it has specific numbers.","fix":"Add: \"High = 2+ independent sources from 2024 on; Medium = 1; Low = older or vendor-only.\"","checks":["r0","r1"]},{"title":"Labels with no proof","why":"Your CFO will ask why each finding got its label.","fix":"Add: \"Each line shows source count, years and any vendor sources.\"","checks":["r2"]},{"title":"Long or formatted output","why":"AI adds an intro, bold labels and tips you didn't ask for.","fix":"Add: \"Only the numbered lines, each under 20 words, plain text.\"","checks":["numberedItems","maxWords","plainText"]}],"promptHints":[{"label":"Gives your confidence rules","patterns":["high\\s*=","2 or more|2\\+|two or more","2024"]},{"label":"Flags vendor sources","patterns":["vendor","independent","sells"]},{"label":"Asks for proof on each line","patterns":["source count","years","how many sources"]},{"label":"Sets the shape","patterns":["\\b(4|four)\\b.{0,20}lines","exactly","numbered"]},{"label":"Asks for plain text","patterns":["plain","no bold","no markdown"]}],"hint":"A research report's summary grades itself. Give the AI your rules and make it show the evidence behind every label.","rubricLabels":["Finding 1 is labeled High (2 independent sources, 2024 and 2025) and Finding 3 is labeled Medium (1 independent source, 2024)","Finding 2 is labeled Low (only a vendor-written source) and Finding 4 is labeled Low (a vendor white paper and a 2020 study, nothing independent from 2024 on)","Every line gives the source count, years and whether any source is vendor-written, and the reply doesn't repeat the report's claim of strong evidence across all areas"]}