Insights from the political corpus

Actor trajectories Methodology →

Per-actor weekly mention volume across all channels (news, commentary, audio, social) over the trailing 6 weeks. The chart shows each actor’s total mentions per week; the table shows channel breakdown and stance sentiment from commentary sources.

Agenda setting Methodology →

Which channel surfaces topics first? The donut shows the overall first-mover distribution across all multi-channel topics. The table below shows channel share-of-voice per topic — where each topic’s coverage actually lives. Topics dominated by one channel (e.g. 80% social / 5% news) surface gaps between public conversation and press coverage.

Channel share-of-voice

Each cell shows what percentage of this topic’s total normalised attention comes from each channel. Raw counts are adjusted for channel volume — so social’s thousands of posts and news’s tens of articles are placed on a comparable scale. The small number underneath is the raw item count.

Topic lifecycle Methodology →

How long do topics persist? Each topic is classified as persistent (still near peak), decaying (fading), flash (one-week spike), or resurgent (bounced back after a dip). Half-life estimates how many weeks until volume halves from peak.

Money & narrative Methodology →

Organisations with recorded donations or Meta ad spend, alongside their media footprint. Does money correlate with coverage? Note: correlation is not causation — parties with more money tend to be larger parties that naturally attract more coverage.

Follow the money: third-party & union spend Methodology →

Disclosed money footprint per organisation — party donations received plus independent advertising spend across the currently-observable sources. Filter by left/right to compare totals and spread; trade unions are highlighted. Independent spend is legally separate from parties — alignment is not coordination.

These are still different disclosure windows, not one comparable period — read each bar as “disclosed money we can currently see for this actor,” not spend within a single timeframe:

• Donations — Electoral Commission >$20,000 register, disclosed in real time but only published in election years; non-election-year totals are sparse by design, not zero. We are currently in the 2026 election year.
• Meta ads (Facebook/Instagram) — rolling, every political/issue ad in our corpus from 2020 to now.
• Google ads (Search/YouTube) — cumulative advertiser totals over the lifetime of Google’s NZ political-ads programme.

Regulated third-party promoter returns — which carry the newspaper and other print spend — are shown separately below, because the only ones published so far are from the 2023 election and don’t belong in a live current-cycle view. See the methodology for each source’s exact coverage and caveats.

Last election: regulated promoter spend (2023)

Registered third-party promoter expense returns from the 2023 General Election — the only source with a per-medium breakdown, including the newspaper and print spend no digital ad library captures. This is historical: it covers the regulated period 14 July–13 October 2023, only promoters who spent over $100,000, and is not current-cycle. It is shown on its own and never summed into the live figures above. The 2026 returns are filed after the election.

Echo chambers & framing divergence Methodology →

Topics where left-leaning and right-leaning commentary sources both have coverage (minimum 5 items per side), with per-stance breakdowns. The divergence score compares the proportion of critical coverage on each side, normalised for volume: 0% means both sides criticise the topic at the same rate; 100% means one side is entirely critical while the other is not at all. Raw stance counts are shown so you can judge the sample size yourself. Source orientation labels come from our hand-curated taxonomy.

Cross-channel anomaly clusters Methodology →

Topics that triggered volume spikes in two or more channels within the trailing 6 weeks. Each card shows the peak spike ratio per channel (e.g. ×5.2 means 5.2× the prior 4-week average). Synchronised anomalies are stronger signals than isolated spikes.

Talk radio exclusives Methodology →

Topics discussed on talk radio (transcribed audio) that are absent or minimal in written media. These represent the parallel conversation happening on air that journalists haven’t picked up — or can’t cover.