Skip to main content
AdCheckNew Zealand

Transparency

Methodology

How we analyse political advertisements, assess claims, select evidence, and calculate truthfulness scores.

What we analyse

We analyse political advertisements, campaign videos, social media posts, leaflets, posters, radio scripts, emails, and other campaign material from all parties, candidates, campaign groups, unions, lobby groups, and advertisers using the same standard rubric.

We do not decide whether we agree with the political message. We analyse only the claims, evidence, context, framing, and persuasion techniques.

Analysis pipeline

  1. 01Media extraction — transcript, OCR, visual description, timeline
  2. 02Claim extraction — factual, statistical, policy, implied, and visual claims
  3. 03Evidence retrieval — checking claims against credible sources
  4. 04Claim assessment — verdict per claim with explanation
  5. 05Scoring — four category scores plus weighted overall score
  6. 06Human review — every published report is reviewed by our team

Scoring rubric

Overall truthfulness is a weighted average of four category scores:

Factual accuracy (40%)

Are the checkable claims in the ad accurate?

  • 10 — All checkable claims are accurate
  • 5 — Mixed accuracy, some important problems
  • 0 — Mostly false

Context fairness (30%)

Does the ad give fair context and avoid distortion?

  • 10 — Fair context, avoids distortion
  • 5 — Selective but not grossly misleading
  • 0 — Highly distorted or deceptive

Evidence quality (20%)

Are claims backed by strong official or independent evidence?

  • 10 — Strong official or independent evidence
  • 5 — Some evidence, but incomplete or partisan
  • 0 — No credible evidence

Visual/audio fairness (10%)

Do imagery and sound fairly support the claims?

  • 10 — Imagery and sound fairly support claims
  • 5 — Emotionally loaded but not deceptive
  • 0 — Imagery or sound strongly misleads

Verdict labels

AccurateMostly accuratePartly trueMisleadingHighly misleadingUnsupportedMostly falseFalseUnverifiable

The final verdict label considers the seriousness of the main claim, not just the numeric score.

Evidence source hierarchy

  1. 1.Official policy documents from the party/campaign being discussed
  2. 2.Electoral Commission guidance or filings
  3. 3.Government agencies (Stats NZ, Treasury, IRD, Ministry reports, Parliament)
  4. 4.Independent official bodies
  5. 5.Reputable news organisations
  6. 6.Academic or policy research
  7. 7.Party or campaign statements from the advertiser
  8. 8.Secondary commentary (clearly labelled)

Party tracker

Party scores are the average overall truthfulness score of all published, human-reviewed analyses for that party. Confidence levels reflect sample size:

  • Low confidence: 1–4 analysed ads
  • Medium confidence: 5–14 analysed ads
  • High confidence: 15+ analysed ads

Human review & corrections

AI-generated analyses are drafts. No report is published without human review. Parties and campaigns can submit a right of reply. Anyone can request a correction. Accepted corrections are noted on the report.

What we avoid

We use neutral language. We prefer “misleading,” “unsupported,” or “false” over inflammatory terms. We do not rank parties by honesty — only by the truthfulness of ads we have analysed.

Browse analyses