AI disclosure

What AI does here

Tārake uses generative AI (Google Gemini via the Lovable AI Gateway) for one specific job: writing short summaries — around 300 words — of stories that have already been published by a reputable outlet. The original source is always linked from the story page and clearly attributed.

AI also generates a short "Why this matters" note and a small kupu (te reo) glossary drawn only from terms already present in the source. It does not invent facts, quotes, iwi affiliations, dates or names.

What AI does not do

• AI does not translate Tārake into te reo Māori. All translations are done by Dr Vincent Ieni Olsen-Reeder, a certified translator.
• AI does not write reader-submitted long-form pieces — those come from the community.
• AI does not moderate reports or set editorial policy — the Editor does.
• AI does not infer whakapapa or ancestry. Iwi affiliations only appear when the source states them or the person publicly self-identifies.

Daily audio bulletin (Gemini Pro TTS)

Once a day Tārake generates a short spoken bulletin of that day's headlines using Google Gemini 2.5 Pro text-to-speech, called through the Lovable AI Gateway. The audio you hear on the homepage is machine-narrated, not a human reader.

What Google receives: only the bulletin script we build on our server — a bilingual mihi, the headline and summary of each featured story, source attribution, and a short pronunciation guide for words like Māori, Aotearoa, Tārake, iwi, whānau and kaupapa. Plus the model and voice configuration (Gemini 2.5 Pro TTS, voice Kore).

What Google does not receive: no reader IP addresses, cookies, accounts or analytics; no story IDs, URLs, images or database rows; no contributor or admin details; no PII of any kind. Only the public headline and summary text already visible on the site is sent, together with the pronunciation guide.

Google's terms for paid Gemini API usage state that prompts are not used to train their models. If you notice a mispronounced name or word in a bulletin, please flag it via the corrections page.

The safety gates

Every AI-curated story goes through automated checks before it can appear on the site:

1. Source gate — the source URL must resolve and be from an allowed publisher.
2. Positivity gate — negative, tragic or harm-focused framings are rejected.
3. Cultural gate — content that risks breaching tikanga or misrepresenting Māori is rejected.
4. Fact-check pass — the AI summary is compared back against the source and given a verdict and a confidence score.

A story only auto-publishes if the verdict is verified and confidence is at least 0.85. Anything less is held for human review. Stories that fall below our confidence threshold, that we cannot verify, or that go stale without readers are removed from search engine indexing.

Limitations

AI systems can and do make mistakes. If you spot something wrong in an AI-curated summary — a mis-stated fact, a mangled name, a missing macron — please tell us via the corrections page or email kiaora@tarake.maori.nz. We treat AI errors as our errors and fix them the same way.

Where our data sits

Tārake currently runs on offshore infrastructure. Moving the platform onto servers in Aotearoa is a priority for us and we aim to complete that by 2028.