Tao ClimateExplore the Planetary Health Platform ↗
Prompt Observatory

Prompt
Observatory.

Which prompts are people picking up? See what participating agents report in real time, what visitors copy here and where a phrase appears in public discussion.

Try
Explore Tao Climate ↗
NASA Blue Marble
AGENT TELEMETRY

Live signal.

Connecting
—reports in the last hour
60 minutes agoNow
—active prompt labels
—15-minute change

Reading the public receiver…

    Updates every 10 secondsSubmit an agent report ↗

    Voluntary, unverified reports from participating agents. Each report represents one declared use.

    CATALOGUE ANALYTICS

    Prompt performance.

    Which prompts get a second look? Compare views and copies from visitors who opt in here with uses reported by agents.

    PromptViewsCopiesCopy rateAgent uses

    Reading the catalogue totals…

    Each prompt counts once per page visit when its text is visible. A successful copy counts once for that view. Agent uses are separate, self-reported events. Private chats and search phrases are excluded.

    PROMPT LIBRARY

    Pick a prompt.

    Empty grey room template, viewed from the entrance
    IMAGE STARTER

    Fill this room

    Fill this room with what you know about me. Be creative and artistic.

    How to use it →
    DAILY EXPLORATION

    Today’s field note

    Find one event from this date in history. Compare two contemporary accounts and explain what each witness could see.

    Featured article source ↗
    BuildBuild the smallest version

    Help me practise vibe coding. Ask who will use my idea and what problem they face. Build the smallest working version, then give me one test a real user could try.

    ResearchFollow the evidence

    Do deep research on [question]. Find primary sources, date each finding, show where the evidence is strongest, and explain why reasonable experts disagree.

    LearnTeach it twice

    Explain [idea] to me twice. First, explain like I am twelve. Then explain it to someone working in the field. Use the same everyday object in both explanations.

    ImaginePostcard from the future

    Write a postcard from [place] in 2050. Describe one thing a visitor can hear, one thing people repair, and one change a resident misses. Keep it under 120 words.

    ExploreA street in three eras

    Choose one street in [city]. Show me what a walker would have seen there in three different eras. Cite a dated map or photograph for each scene.

    DecideRehearse a decision

    I need to decide whether to [decision] by [date]. Interview me one question at a time, explain why each answer matters, then summarise the trade-offs.

    CreateThe one-object museum

    Curate a tiny museum exhibition around [ordinary object]. Write three labels for visitors: how it was made, where it travelled and why someone kept it.

    GLOBAL RESEARCH

    What people ask.

    Global share of sampled individual ChatGPT messages by topic, September 2026. OpenAI publishes category shares; exact prompt rankings and all-provider volumes are unavailable.

    1. Practical Guidance28.1%
    2. Writing21.4%
    3. Seeking information20.2%
    4. Self-expression8.9%
    5. Multimedia8.8%
    6. Other/Unknown8.1%
    7. Technical help4.6%

    Writing moved from 34.1% in July 2024 to 21.4% in September 2026. Practical guidance led the latest month at 28.1%.

    Derived from OpenAI Signals v3.0 by Chatterji et al. · CC BY 4.0. Shares are rounded. Enterprise and Codex are excluded.

    PROMPT ANATOMY

    The six questions.

    Which ingredients show up most? Each library prompt is tagged once per element. Switch to reported use to weight those tags by voluntary reports received in the last 30 UTC days.

      HACKER NEWS

      Vibe coding

      Loading source data…

      Hacker News mentions…
      Peak daily mentions…
      Days with mentions…

      Mentions over time

      Hacker News mentions

      Matching discussions

      Inspect the individual records behind the chart.

      Methodology

      Agent reports

      The live receiver groups voluntary submissions by prompt label and receipt minute in UTC. Figures refresh every ten seconds. Changes compare two complete 15-minute windows. Reports are self-declared and test submissions are excluded.

      Hacker News mentions

      Each matching Hacker News story or comment counts once. We request phrase candidates from the Algolia Hacker News index, then check the title or body for consecutive whole words, ignoring case and punctuation. The chart covers text indexed on Hacker News. The X links open a separate media search. Posts that have only an image may be easier to find in the thread replies. A mention is evidence of public discussion. It cannot establish that anyone ran a prompt in ChatGPT.

      Recent results are fetched on demand. For the example phrases a dated snapshot is available when live search is unavailable. A limited search is labelled as partial.

      Analyse an export

      Each matching user message on the current branch of an exported conversation counts once, using its original UTC timestamp. Assistant replies, alternative branches, undated messages and files are excluded. Counts cover the selected period. The JSON is processed on your device and released when you clear it or leave.

      Only phrases you explicitly save are kept in this browser. Searching the public source sends your query to Algolia.

      Hacker News search API ↗