How we use AI

The tools we use, what they do, what they don’t do, and how a named human reviewer signs off on every story.

Status: Draft, needs human review — not yet assigned to a reviewer. All policies will be reviewed by a California media-law attorney before launch.

Status

This policy is a first draft. AI usage in journalism is evolving fast, and this policy will evolve with it. Material changes will be timestamped and versioned.

What this policy commits us to

Pier and Point is an AI-assisted hyperlocal newsroom. That means AI tools help us draft, summarize, transcribe, and translate — but they do not decide what to publish, and they do not stand in for a human reporter.

Every AI-assisted story carries a permanent, visible disclosure banner naming the tool, the reviewer, and the sources. Disclosure is not buried in a footnote or hidden behind a "more info" toggle.

The tools we use, by name

We disclose our stack so readers can judge it for themselves.

Drafting: Kimi K2.6 routed through the Vercel AI Gateway with Zero Data Retention enabled. Confidential source material never goes through Moonshot direct.
Fact-checking: Claude Sonnet 4.6 via the Anthropic commercial API with the Citations API enabled. Citations API returns source-attributed responses so we can audit which document passages support which claims.
Transcription: WhisperX with whisper-large-v3 and pyannote diarization, run locally on the editor’s workstation. Council and board recordings are transcribed locally and never leave the local network unless the story requires us to share a transcript publicly.

We may add or change tools over time. When we do, we will update this page and timestamp the change.

What AI does at Pier and Point

Summarizes primary documents (council agendas, planning packets, public-records responses) into structured drafts.
Translates stories between English and Spanish where we publish bilingually.
Transcribes public meeting audio so we can quote speakers accurately and quickly.
Suggests headlines, subheads, and dek copy that a human editor selects, edits, or rejects.
Cross-checks factual claims against the documents they were drawn from, flagging weak or unsupported assertions for human review.

What AI does NOT do at Pier and Point

These are categorical, not soft preferences:

No fabricated quotes. Every direct quote in a Pier and Point story comes from a real person, a real public meeting, a real public document, or a real interview.
No AI photorealism of real Ventura events, people, or places. Following AP and Reuters, we do not publish AI-generated images that depict real things photorealistically. AI illustrations, when used, are clearly captioned as such.
No AI voice cloning. We do not synthesize a voice to imitate a real person, ever.
No AI-only-sourced reporting. Every story rests on at least one primary document, public record, named human source, or first-hand observation. AI may help us read or summarize that material — it does not stand in for it.
No AI-generated reviews or ratings. Per the FTC Final Rule on Reviews and Testimonials (16 CFR Part 465), we do not publish AI-fabricated consumer reviews of restaurants, businesses, or services.

Human in the loop, for real

Every story Pier and Point publishes has a named human reviewer who:

Reads the AI-drafted text in full.
Verifies every factual claim against a primary source.
Confirms the arrestee / charged / convicted distinction is applied correctly. (See naming.)
Confirms a right-of-reply attempt was made for every named private individual.
Signs off on the AI prompt log being attached to the editorial record.

This is not an aspirational checklist. The publish action is gated on the reviewer signing off; an AI cannot one-click publish a Pier and Point story.

Disclosure language

The disclosure banner on each AI-assisted story names the tool and the reviewer specifically, not generically. Example:

Drafted with Kimi K2.6 from a council agenda packet and meeting transcript. Reviewed by Finn Bennett before publication. Sources: [list].

Generic boilerplate ("we use AI tools in our newsroom") does not appear in place of specific disclosure.

Privacy and source confidentiality

Confidential source material is never sent to a Chinese-jurisdiction model API directly. We route through Vercel AI Gateway with Zero Data Retention.
The Anthropic commercial API is used in its no-training default with 7-day retention. We do not use the consumer Claude.ai web app for production newsroom work.
Local transcription (WhisperX) is used for any audio that includes off-the-record material.

Questions

If you have a question about how a specific story was produced, email hello@pierandpoint.com with the URL and we’ll answer it.