{"id":"mkt-x-email-test-hypotheses","track":"marketing","audience":"marketing","category":"brainstorming","technique":"grounding","title":"Write A/B test hypotheses backed by your past results","free":false,"task":"You're planning next quarter's email tests at Harrow & Hale, a men's shoe brand. Write exactly 3 numbered hypotheses, each one sentence in the shape If we..., then... will rise, because... (citing a past result from the notes), under 30 words. Each tests only one change, none touches discounts, and reply with only the plain-text list.","materialLabel":"Past test results (attached to your prompt automatically)","material":"HARROW & HALE - email test log (last 2 quarters)\n\nTest 1: Subject with product name vs. without\n- \"The Calder boot is back\" vs. \"Back in stock\" -> product name won, open rate 34% vs. 27%\n\nTest 2: Send time 7am vs. 7pm\n- 7pm won on clicks: 3.1% vs. 2.4%\n\nTest 3: One product vs. grid of 6 products\n- Single product won on clicks: 3.8% vs. 2.2%\n\nTest 4: Lifestyle photo vs. plain product photo\n- No real difference (2.9% vs. 2.8%), not significant\n\nConstraints for next quarter:\n- No discount tests. Leadership wants to stop training customers to wait for sales.\n- Each test should change one thing only.\n\nIdeas floating around:\n- \"Redesign the whole template\" (too many changes)\n- \"Test 20% vs. 30% off\" (breaks the no-discount rule)","rules":[{"type":"numberedItems","count":3},{"type":"plainText"}],"mistakes":[{"title":"No hypothesis shape","why":"The AI writes vague test ideas without a prediction or reason.","fix":"Show it: \"If we..., then... will rise, because...\"","checks":["r0"]},{"title":"Not grounding the reason","why":"The AI makes up reasons from general email advice.","fix":"Add: \"Each because cites a result from our test log.\"","checks":["r1"]},{"title":"Testing what's off-limits","why":"Discount tests and redesigns are in the notes, and the AI picks them up.","fix":"Add: \"One change per test. No discounts, no redesign.\"","checks":["r2"]},{"title":"Getting a planning doc","why":"The AI writes a table with bold headers and success metrics.","fix":"Add: \"Exactly 3 numbered sentences, plain text.\"","checks":["numberedItems","plainText"]}],"promptHints":[{"label":"Sets the hypothesis shape","patterns":["if we","then","because"]},{"label":"Cites past results","patterns":["past (test|result)","test log","cite"]},{"label":"One change each","patterns":["one (change|thing|variable)","single"]},{"label":"Rules out discounts","patterns":["discount","% off"]},{"label":"Asks for only the list","patterns":["only the (numbered )?list","no (intro|notes|explanation)","nothing else","plain text"]}],"hint":"A hypothesis needs a prediction and a reason. Tell the AI the shape, to cite your own results, and what's off the table.","rubricLabels":["Every hypothesis is one sentence in the shape 'If we..., then... will rise, because...' and under 30 words","Each 'because' cites a real result from the log (such as product-name subjects 34% vs. 27%, 7pm clicks 3.1% vs. 2.4%, single product 3.8% vs. 2.2%), and none builds on the lifestyle photo test, which showed no difference","Each hypothesis changes only one thing, none involves discounts or a template redesign, and the reply is only the numbered list"]}