Newscastor.ai · Public policy

Editorial Ethics & Corrections Policy

What we hold ourselves to, what we ask of every newsroom that publishes through Newscastor, and exactly what happens when something is wrong.

1. Why this policy exists

Newscastor.ai is a production tool, not a newsroom. It has no reporters, no editors, and no independent knowledge of the world. Every newscast it produces is only as accurate as the journalism a partner newsroom gives it — and AI systems can introduce errors that were not in the source material.

Audiences do not make that distinction. If a Newscastor-produced segment is wrong, the harm lands on the newsroom's name, on the people in the story, and on public trust. This policy exists so that everyone involved — Newscastor, partner newsrooms, and the audience — knows who is responsible for what, what standards apply, and exactly what happens when something is wrong.

It is modeled on the published standards of The Washington Post, The New York Times, and the Associated Press, adapted for a platform where a machine sits between the reporting and the audience.

2. Who this applies to

Why the line is drawn at self-declaration. Audiences treat "news" differently from everything else, and so do we. An account that calls itself a news organization is asking the public to extend it the trust that label carries. This policy is the price of that label. Accounts that don't claim it — entertainment, opinion, creator content — are not claiming that trust and are not held to newsroom standards; they're held to PollinatorStudio's base policy. Claiming the news label falsely, or dropping it to escape this policy after publishing as news, is a §11 violation.

Whose policy governs editorial judgment? The newsroom's. Newscastor is a production tool. A newsroom's ethics, sourcing, fairness, and corrections standards belong to the newsroom and its counsel — not to a vendor. We therefore do not ask newsrooms to adopt KSG's editorial standards. We ask two things: that they acknowledge the Platform Baseline (what the machine guarantees regardless of anyone's opinion), and that they put their own editorial policy on file so their audience, and we, can see what they hold themselves to. A newsroom that has no written policy may adopt the template in Appendix D as its own document, under its own name, after its own review.

Signup condition: No Partner Newsroom account may publish until an authorized representative has completed the News Organization Declaration, provided the newsroom's editorial policy (or adopted the template), and acknowledged the Platform Baseline (Appendix A). All of this is logged (who, when, which version). Material changes to the Baseline require re-acknowledgment.

2a. Platform Baseline

These are guarantees about how Newscastor.ai operates. They apply to every news segment on the platform and cannot be switched off by any account, because they describe the pipeline, not editorial judgment.

1. A named human approves every news segment before it publishes. The approval record (who, when, version) is retained. There is no auto-publish for news.
2. AI disclosure is always present and cannot be removed. Spoken and/or on-screen, plus metadata: that the newscast was produced with AI, from whose reporting, and how to report an error.
3. Nothing is invented. The pipeline does not generate facts, quotes, names, numbers, dates, images of real events, or statements by real people that are not in the newsroom's source material. Hedge language in the source ("police say," "alleged," "unconfirmed") is preserved.
4. No real person's voice or likeness without documented consent. Synthetic presenters are clearly fictional.
5. Corrections create a new version; the original is retained and marked superseded. Nothing is silently overwritten. Anyone can report an error (§8).
6. Pipeline errors are disclosed. If our systems introduce an error, we tell every affected newsroom within 24 hours and publish a notice.
7. We do not train models on newsroom content without separate, explicit, written consent.

Everything beyond this list — what counts as a source, how to handle anonymous sourcing, conflicts of interest, sensitive subjects, opinion labeling, how fast to correct — is the newsroom's policy to set. Sections 5, 7, and 9 below describe how the platform supports those decisions and what we recommend; the newsroom's own policy governs where it differs.

3. Core principles

These are in order of precedence. When two conflict, the earlier one wins.

1. Accuracy over speed. A late newscast is a missed opportunity. A wrong one is a harm. We never trade the second for the first.
2. Human responsibility. A named human at the Partner Newsroom is editorially responsible for every newscast published. AI does not hold responsibility; people do.
3. Nothing invented. Newscastor does not generate facts, quotes, names, numbers, dates, images of real events, or statements by real people that do not appear in the source reporting. Ever.
4. Fairness. Following the Post's standard: no story is fair if it omits facts of major importance, includes irrelevant information at the expense of significant facts, or misleads by emphasis or tone. Subjects of serious allegations get an opportunity to respond in the source reporting before it is broadcast.
5. Independence. Editorial content is not shaped by advertisers, sponsors, funders, platform partners, or KSG's own business interests. Sponsored or paid content is labeled as such and kept visually and audibly distinct.
6. Transparency. Audiences are told, clearly and every time, that a newscast was AI-produced, from whose reporting, and how to report an error.
7. Accountability. Errors are corrected promptly, visibly, and in a way that is as prominent as the error. We keep a public record. We do not quietly fix things.

4. Division of responsibility

This is the section that matters most for a platform, so it is deliberately explicit.

4.1 Partner Newsroom is responsible for

• The accuracy, sourcing, fairness, and legality of all source material submitted to Newscastor.
• Reviewing and approving every newscast before publication (Section 5.3). "Auto-publish without review" is not available for news content.
• Designating at least one Responsible Editor — a real, named person — per account.
• Filing corrections when it learns of an error, within the timelines in Section 7.
• Keeping its editorial policy on file with us current, and honoring it. Where it is stricter than anything here, it governs.

4.2 Newscastor.ai is responsible for

• Designing and operating its AI systems so they do not add, remove, or distort facts present in the source material, and testing for that continuously.
• Making it technically impossible to publish a news segment without a human approval step.
• Applying AI-disclosure labels automatically and making them impossible to strip.
• Providing the corrections tooling (Section 7.4) and the public corrections log.
• Responding to error reports that reach Newscastor directly, and routing them to the Partner Newsroom the same business day.
• Investigating and disclosing any error that originated in the AI pipeline rather than the source material (a pipeline error), including notifying every Partner Newsroom that may have been affected.
• Suspending accounts that violate this policy (Section 12).

4.3 What Newscastor will not do

• Act as the editor of record for any Partner Newsroom's content.
• Adjudicate disputes about the underlying journalism. That belongs to the newsroom and, where applicable, its ombudsman or press council.
• Publish original "reporting" generated by AI from sources the Partner Newsroom has not vetted.

5. Editorial standards for AI-produced newscasts

5.1 Source-of-record rule

Every factual statement in a newscast must trace to a specific piece of reporting the Partner Newsroom has published or formally approved. Newscastor keeps that linkage (story → script → rendered segment) and exposes it to the Responsible Editor. If a statement cannot be traced, it cannot air.

5.2 No fabrication, no synthesis of people

• Quotes: Only verbatim quotes from the source material, attributed as in the source. No paraphrase presented as a quote. No AI-generated "what they might have said."
• Voices and likenesses: Newscastor's synthetic anchors are clearly fictional presenters. We do not clone the voice or face of any real journalist, public figure, or private person without that person's documented, revocable consent — and never to put words in their mouth that they did not say.
• Imagery: Generated or altered images and video are never presented as depictions of real events, places, or people. B-roll is either licensed real footage, clearly abstract/illustrative, or labeled "illustration." This follows the AP rule against AI images that purport to show real events.
• Numbers and names: Preserved exactly. The pipeline is tested specifically against numeric drift, date drift, and name substitution.

5.3 Mandatory human review

Before any news segment publishes, the Responsible Editor (or an Authorized User they designate) must:

1. Read the generated script side-by-side with the source reporting.
2. Confirm every factual claim traces to the source.
3. Preview the rendered audio/video for mispronunciations that change meaning, mis-emphasis, tone problems, and any visual that could mislead.
4. Approve by name. The approval record (who, when, version) is retained.

5.4 Tone and emphasis

AI systems tend toward confident, smooth delivery. That is a risk when the underlying reporting is hedged, developing, or contested. Scripts must preserve the source's uncertainty language ("police say," "according to," "has not been confirmed," "alleged"). Stripping hedges is treated as a factual error.

5.5 Sensitive subjects

Partner Newsrooms must apply their own standards and, at minimum, the following: do not identify minors, victims of sexual violence, or confidential sources unless the source reporting did so deliberately and lawfully; do not broadcast unverified identifications of suspects; treat suicide, self-harm, and mass-casualty events with established reporting guidelines (e.g., avoid method detail). Newscastor may add friction (a required second approval) on segments flagged by these categories.

6. Sourcing, attribution, and transparency to the audience

Every newscast carries, in both the rendered media and its metadata:

• AI disclosure: "This newscast was produced with AI by Newscastor.ai from reporting by [Partner Newsroom]." Spoken at the open or close, and shown on screen for video. Cannot be removed by the partner.
• Source attribution: the originating story/stories, with links where the distribution channel allows.
• Responsible Editor: the newsroom's editorial contact for the segment.
• Error reporting path: how to report a problem (Section 8).
• Version and timestamp: so a corrected re-render is distinguishable from the original.

Partner Newsrooms may not describe Newscastor output as human-anchored, live, or "reported by" the synthetic presenter.

7. Corrections policy

It follows the Post's three-tier structure (correction / clarification / editor's note) and adds the two things a media pipeline needs: re-rendering and a pipeline-error class.

7.1 Definitions

7.2 What gets corrected

Any error of fact, large or small, in any element of the newscast: script, spoken audio, on-screen text, captions, graphics, imagery, metadata, or attribution. "It was only the lower-third" is not a reason to skip a correction. Matters of opinion, editorial judgment, or disagreement with the underlying reporting are not corrections and are referred to the Partner Newsroom.

7.3 Timelines

These are proposed defaults, not industry mandates — there is no regulator-set clock for corrections in U.S. journalism. They are set to be faster than print-era norms because distributed audio/video keeps spreading while a correction waits.

7.4 How corrections are made (the mechanics)

1. The original is not silently overwritten. The corrected segment is a new version. The original remains in the archive, marked superseded, visible to the Responsible Editor and, on request, to the public.
2. Re-render. The corrected script is re-rendered. The new version carries a spoken and on-screen note: "This segment has been corrected. [What was wrong, what is right.]"
3. Correction card. A written correction is attached to the segment wherever it is hosted, with the date and the nature of the error — stated plainly, without euphemism. Following the Times' practice, the correction says what the error was, not just what the truth is.
4. Downstream channels. Where a segment was pushed to third-party platforms (podcast feeds, social, OTT), the Partner Newsroom replaces or annotates it there too. Newscastor's tooling generates the correction text for each channel.
5. Public log. Every correction, clarification, editor's note, and takedown is listed on the Partner Newsroom's public Newscastor corrections page and on Newscastor's platform-wide log at newscastor.ai/corrections. Pipeline errors are listed platform-wide.

7.5 What a correction notice says

Correction (YYYY-MM-DD): An earlier version of this segment said [X]. [Y] is correct. [One sentence on the cause, if known: "The error was in the source reporting" / "The error was introduced during AI script generation and has been reported to Newscastor."]

Clarifications and editor's notes follow the same plain-language form (templates in Appendix B).

8. Reporting an error

Anyone — audience member, story subject, another journalist — can report a suspected error.

• To the Partner Newsroom: via the contact shown on every segment.
• To Newscastor directly: corrections@newscastor.ai or the form at newscastor.ai/report-an-error. Newscastor routes the report to the Partner Newsroom the same business day and tracks it to resolution.
• Reporters of errors receive a reply with the outcome. We do not require a reason, an account, or proof of identity to file a report. Reports are never used to identify or retaliate against the reporter.

8a. Fact-check assist

Newsrooms may connect Google's Fact Check Tools API to their account and turn on fact-check assist. When it is on, the platform extracts factual claims from each generated script and looks them up in the Google Fact Check database — the set of fact-checks that publishers such as the Associated Press, Reuters, PolitiFact, and local fact-checkers have published with ClaimReview markup. Matches (publisher, rating, date, link) are shown to the Responsible Editor in the review screen before approval. Nothing is changed automatically. Every lookup is logged per segment.

What it is and isn't. This is a lookup against existing fact-checks, not an independent verification of the newsroom's reporting. A claim with no match is not thereby true; a claim with a match is not thereby false. The human reviewer decides. For that reason, on-air and written wording must say that claims were checked against the Google Fact Check database — never that content was verified by Google. The platform blocks the latter phrasing.

Cost and control. The integration uses the newsroom's own Google API key (bring-your-own-key), so any usage cost or quota is the newsroom's, not pooled. Fact-check assist is off until a key is connected and can be turned off at any time. It is strongly encouraged during onboarding and never required.

Disclosure line. Newsrooms may use a platform-generated intro that reflects the account's actual settings, for example: "The claims in this newscast were checked against the Google Fact Check database before broadcast, and a human editor reviewed every story. To read our ethics policy, how we use AI, and how we check facts, visit [newsroom website]." With assist off, the line omits the fact-check clause. The website spoken in the line is the newsroom's, where its own policy (§2) is published.

9. Independence, conflicts, and paid content

The first item is a platform rule; the rest describe what the newsroom's own policy should cover and what we recommend.

• Sponsored segments must be labeled "Sponsored" or "Paid content" in audio and on screen and are never interleaved with news in a way that blurs the boundary. They are excluded from the news corrections log but still subject to accuracy rules and consumer-protection law.
• KSG's own interests. Newscastor and KSG do not edit, rank, suppress, or promote partner content based on KSG's commercial, political, or community relationships. Any KSG-affiliated entity that becomes a Partner Newsroom is held to the same rules and identified as affiliated.
• Partner Newsroom conflicts are governed by the newsroom's own policy, in line with the Post's and Times' standard: no gifts, payments, or favors from subjects of coverage; disclosed outside interests; no coverage of entities from which payment is accepted.

10. Privacy, source protection, and data handling

• Newscastor does not train models on Partner Newsroom content without explicit, separate, written consent.
• Partner Newsrooms must not submit unpublished confidential-source material, legally restricted documents, or personal data beyond what the published story contains. Roughly half of newsroom AI policies now warn against feeding sensitive material into third-party AI tools; this policy makes it a rule.
• Newscastor's subprocessors (model providers, rendering, hosting) are listed publicly and updated when they change.

11. Prohibited uses

A Partner Newsroom may not use Newscastor to:

• Publish content that has not gone through human review.
• Present AI-produced content as human-reported or live.
• Clone a real person's voice or likeness without documented consent.
• Generate imagery presented as real events or real people.
• Produce content that impersonates another news organization, government agency, or emergency service.
• Strip, hide, or obscure AI-disclosure labels or corrections.
• Publish content it knows to be false.
• Declare itself a news organization when it is not, withdraw the declaration to avoid this policy after publishing as news, or put a policy on file it does not actually follow.
• State or imply that content was verified by Google or by Newscastor. Claims are checked against the Google Fact Check database; they are verified by the newsroom.

12. Enforcement

Most problems are honest mistakes, and the first response is always help, not punishment — that is the KSG Compliance Partnership posture. But repeated or willful violations have consequences, because audience trust is a shared asset.

Partners may appeal any enforcement action to KSG in writing. Decisions and reasons are recorded.

13. Review and change log

• This policy is reviewed at least every 6 months and whenever a pipeline error reveals a gap.
• Every change is dated in the public change log. Material changes trigger re-acceptance by Partner Newsrooms.
• Questions, challenges, and suggestions: ethics@newscastor.ai.

Appendix A — Newsroom onboarding gate

Shown only to accounts that select a news organization type at signup. Other accounts see the PollinatorStudio acknowledgment instead. Everything below is logged with timestamp, IP, and Baseline version.

Step 1 — Organization type (required)

Daily or weekly newspaper · Radio news · Public media · Television news · Digital-native news outlet · Nonprofit newsroom · Wire service / syndication · Student or university news organization · Trade, industry, or specialty news publication · Other / not a news organization (→ Creator path)

Would your listeners expect what they hear to be accurate, verified news? If yes, you're a news organization here — regardless of size.

Step 2 — Declaration

Step 3 — Your editorial policy (required)

One of: - [ ] URL to our published editorial ethics / AI policy: __ - [ ] Upload our policy (PDF or document) - [ ] We do not yet have a written policy. We are adopting the Newscastor Newsroom Editorial Policy Template (Appendix D) as our own document, under our name, and will publish it at ____ within 30 days.

Where we will link our policy for our own audience (footer, about page, newscast description): ______

Step 4 — Platform Baseline acknowledgment

• We have read the Platform Baseline (§2a) and understand these seven guarantees apply to every news segment we publish through Newscastor and cannot be disabled.
• We understand a named human at our organization must approve every news segment before it publishes, and we designate the Responsible Editor above.
• We will not remove or obscure AI-disclosure labels or corrections, clone a real person's voice or likeness without documented consent, or present generated imagery as real.
• We will not describe content as "verified by Google" or "verified by Newscastor."
• We confirm we are a news organization as declared above and understand a false declaration is grounds for termination (§§11–12).

Step 5 — Fact-check assist (encouraged, not required)

• Connect our Google Fact Check Tools API key now (we've read §8a: lookup, not verification; our key, our usage)
• Skip for now — remind me

Typed name: __ Title: _ Organization: Date: ___ I have authority to bind this organization: ☐ Yes

Appendix B — Notice templates

Correction

Correction (DATE): An earlier version of this segment [stated / showed] . The correct ___ is . [Cause sentence.]

Clarification

Clarification (DATE): This segment [omitted / did not make clear] ___. [Added context.]

Editor's Note

Editor's Note (DATE): This segment [did not meet our standards because ___]. [What we did: re-reported / withdrawn / re-rendered.] We regret the error.

Pipeline Error Notice (Newscastor)

Pipeline Error Notice (DATE): Between ___ and ___, Newscastor's script-generation system [description of defect, e.g., rounded figures / dropped attribution]. ___ segments across ___ partner newsrooms were affected. All have been corrected or withdrawn. [Root cause and fix.]

Appendix C — How this maps to the standards we studied

Appendix D — Newsroom Editorial Policy Template

A newsroom that has no written editorial policy may adopt this template as its own document, under its own name, after its own review. It is offered as a starting point, not a requirement; newsrooms with an existing policy should use theirs. The template covers: accuracy and sourcing (named sources preferred; two independent sources for confidential-source stories); fairness and the opportunity to respond; hedge language and developing stories; sensitive subjects (minors, victims, suspects, suicide and self-harm, mass-casualty events); opinion labeling; conflicts of interest, gifts, and outside interests; sponsored content separation; corrections tiers (correction, clarification, editor's note, retraction) and timelines; how audiences report errors; and how AI tools are used and disclosed. The template is published at newscastor.ai/policy-template and is maintained alongside this policy.

Related: Newscastor.ai Corrections & Error Reporting · PollinatorStudio.ai AI Use, Transparency & Accountability Policy · How we build
© 2026 Kenwood Solutions Group · Newscastor.ai is a PollinatorStudio.ai product serving news organizations · ethics@newscastor.ai · corrections@newscastor.ai