![Surprised my girlfriend by visualizing our WhatsApp chat export and made a website. [PROMPT]](https://preview.redd.it/0nqsqgo43uah1.jpeg?auto=webp&s=5c2349cd9d531900b877cc5426a998b198ad271c)
Surprised my girlfriend by visualizing our WhatsApp chat export and made a website. [PROMPT]
So my girlfriends birthday is coming, i exported my whatsapp chat with her and ran it in emergent to create a website. Here is the prompt i was using:
PROMPT:
You are a data-storyteller. I'm giving you an exported WhatsApp chat (a .zip with a _chat.txt). Analyze the ENTIRE conversation and build me beautiful, self-contained HTML pages that reveal who these two people are to each other, not just how much they talked. Ground rules (the traps):
1/ Parse robustly - lines are [DD/MM/YY, HH:MM:SS AM/PM] Sender: message; multi-line messages continue to the next timestamp. Strip invisible Unicode first (LRM/RLM/zero-width), WhatsApp injects them before media/call lines and they break
regexes.
2/ De-duplicate - exports often get concatenated so messages repeat 3-4x. Dedupe by (timestamp, sender, text); report raw vs true-unique count; use unique for all
stats.
3/ Sort by timestamp - ~0.2% of rows are out of order; sort before any gap/streak math or you'll invent fake silences. 4/ Verify surprising claims against the raw
lines before presenting.
5/ Media = markers, not words - exclude <attached>/omitted/.opus/.webp from word counts, count them separately.
6/Tone: warm, clever, decent - "what the behavior suggests," never creepy.
Compute broadly: basics (msgs, words, media, emoji); rhythm (hourly/daily, inferred sleep window, after-midnight %, reply time by hour, streaks, all-nighters); call log (hours, avg, longest, missed); mechanics (who initiates/ends, breaks silences, double-texts, session length); behavioral, who's funnier (laughter triggered), who does emotional labor (sorry/ thanks/comfort), reassurance-seeking, expressiveness (elongation/caps/emoji), I-vs-you-vs-we focus, "we-ness" over time, language-style-matching (do their message lengths converge?), the fading of the original pretext, nickname evolution, sentiment by month; life signals, likely in-person meetups, social graph, any rupture (mutual vs one-sided + who reconnected); vs the world, web-search real global averages (msgs/day, reply time, call length, sleep) and show the multiplier.
Deliver 4 self-contained HTML pages (inline CSS+JS, no libraries, embedded JSON, animated, own visual identity, cross-linked nav): Wrapped (cute overview), Deep Cuts (forensic), Profiles (character cards), vs the World (benchmarks). Open them when done. First: parse → dedupe → sort → show a sanity summary (raw vs unique, date range, per-person split) for me to confirm. Then compute everything and build. Work autonomously; verify surprising findings.
Just surprise your girl with this🗿
Bye.