{"id":"eval-mkt-blog-draft-claims-vs-survey","audience":"marketing","category":"evaluation","track":"evals","technique":"verification","title":"Check an AI blog intro against your own survey report","free":false,"task":"An AI drafted a blog intro from your customer survey report, and every stat has to match before it's published. Write exactly 5 numbered lines, one per claim, each labeled Matches, Wrong or Not in report. For each Wrong line, add the report's figure in quotation marks. Judge only by the report. Plain text, under 100 words.","materialLabel":"The survey report and the AI's draft (attached to your prompt automatically)","material":"SURVEY REPORT - Thistle Pay \"State of Freelance Invoicing\" (Sept, n = 1,040 freelancers in the US)\n- 62% said they were paid late at least once in the past year.\n- The average late payment arrived 18 days after the due date.\n- 41% send invoices from a spreadsheet or word processor.\n- 27% have ever charged a late fee.\n- Freelancers who send reminders within 3 days of the due date reported getting paid 9 days faster on average.\n\nAI-DRAFTED BLOG INTRO (claims numbered by you)\n(1) In our survey of over 1,000 US freelancers, 62% were paid late at least once last year. (2) The average late payment showed up nearly a month after the due date. (3) Almost half still invoice from a spreadsheet or document. (4) Only 17% have ever charged a late fee. (5) Most freelancers say late payments are their number one stress.","rules":[{"type":"numberedItems","count":5},{"type":"plainText"},{"type":"maxWords","max":100}],"mistakes":[{"title":"Asking for a proofread","why":"The AI fixes commas and style and never checks the numbers against the report.","fix":"Say: \"Label each claim Matches, Wrong or Not in report.\"","checks":["r0","r2"]},{"title":"Letting rounding slide","why":"'Nearly a month' sounds close enough to the AI, but 18 days isn't.","fix":"Add: \"Wrong includes rounding that changes the meaning.\"","checks":["r0","r1"]},{"title":"No proof for corrections","why":"The AI says a stat is off without giving the real figure.","fix":"Add: \"For Wrong, quote the report's figure.\"","checks":["r1"]},{"title":"A rewritten intro","why":"The AI 'helps' by rewriting the paragraph.","fix":"Be exact: \"Exactly 5 numbered lines, one per claim, nothing else.\"","checks":["numberedItems","maxWords","plainText"]}],"promptHints":[{"label":"Gives exact labels","patterns":["matches","wrong","not in report","label"]},{"label":"Makes the report the judge","patterns":["only.{0,30}(report|survey)","against the (report|survey)"]},{"label":"Asks for a quote as evidence","patterns":["quote","exact words","evidence"]},{"label":"Says exactly how many","patterns":["\\b(5|five)\\b.{0,25}(line|claim)","exactly \\d+"]},{"label":"Asks for plain text","patterns":["plain","no bold","no markdown","no formatting","no headings"]}],"hint":"AI drafts round numbers up and add claims that sound like your data. Give the AI three labels and make the report the only judge.","rubricLabels":["Labels claims 1 and 3 Matches, claims 2 and 4 Wrong, and claim 5 Not in report","The Wrong lines give the report's figures in quotation marks: 18 days (not nearly a month) and 27% (not 17%)","Uses only the three labels Matches, Wrong and Not in report, with no rewritten intro, headline ideas or intro sentence"]}