Build an ad-targeting customer persona that flags every invented detail instead of presenting guesses as facts

Build a customer persona for A meal-kit subscription service targeting time-constrained households. to guide ad creative and targeting decisions — the requirement is that every detail in the persona is traceable to either the real data I give you or an explicitly labeled inference, never presented as an established fact when it's actually a guess.

PRODUCT / SERVICE
A meal-kit subscription service targeting time-constrained households.

REAL DATA AVAILABLE
200-response survey: 65% cite "lack of time to plan meals" as top reason for subscribing; support tickets show recurring complaints about portion sizes for larger households.
(survey responses, customer interview notes, purchase/behavioral data, support ticket themes — whatever you actually have)

WHAT THIS PERSONA WILL BE USED FOR
Deciding which pain point to lead with in a new round of Facebook ad creative.

KNOWN GAPS IN THE DATA
No reliable data on household size, income level, or cooking skill.

Build the persona in two clearly separated parts. Part 1, "Confirmed from data": only details directly supported by the real data given, each one citing which piece of data supports it (e.g., "Frequently mentions price comparison in support tickets — supported by 40% of tickets referencing a competitor's price"). Part 2, "Reasonable inferences, unverified": details that would round out a useful persona but aren't directly evidenced — clearly labeled as inference, with the reasoning shown, and phrased as hypotheses to test rather than facts ("likely also considers [X], though this isn't confirmed by current data"). Do not blend the two categories into one seamless narrative persona bio that reads as if everything is equally certain — that's the most common failure of AI-generated personas, where an invented detail like an age range or a hobby sits next to a real data point with identical confidence, and whoever uses the persona downstream has no way to tell which parts to trust. For the stated use case, add one section naming which unverified inferences would be most valuable to actually confirm first, since not all data gaps matter equally for every use case.

OUTPUT: Part 1 (confirmed, with citations to the data given), Part 2 (inferences, clearly labeled and reasoned), and a short prioritized list of what to verify next.