'We Did Not Trust It' Is Not a Review Record
When a newsroom rejects an AI claim, that decision needs a record. How to document the rejection so the review is defensible if the claim resurfaces elsewhere.
Who this is for
Editors, managing editors, newsroom teams, fact-checking teams — Editors and managing editors who need a structured process for documenting the rejection of AI-sourced claims so the decision is defensible and reviewable
The problem
The decision not to publish is as important as the decision to publish. When a newsroom rejects an AI-sourced claim because evidence was insufficient, model agreement did not substitute for corroboration, or the allegation could not be independently verified — the reasoning should be documented. Without a record, the rejection is a judgment call with no audit trail.
If the story surfaces later in another outlet, if a source challenges the decision, or if a similar claim appears with new evidence, there is no record of what was found and why it was not sufficient. The gap is not malice — it is that most newsrooms have workflow for publication but not for documented rejection.
How ConvergePanel helps
ConvergePanel's verification output provides the evidence layer for the rejection record. The log of what was submitted, what models found, where they disagreed, and what evidence was reviewed becomes the basis for the documentation. The journalist or editor adds the editorial conclusion and the specific reason the claim was not sufficient for publication.
How they compare
| Record field | What to document | Why it matters |
|---|---|---|
| Claim | Exact wording of the AI-sourced claim | Establishes what was under review |
| Sources supplied | Every source the AI cited or implied | Documents the evidentiary basis as received |
| Evidence reviewed | What was independently checked and what it showed | Proves the rejection was based on review, not avoidance |
| Model analysis | ConvergePanel output: agreement score, disagreements, evidence gaps | Provides structured documentation of AI research quality |
| Weakness identified | Specific reason the claim was insufficient: missing corroboration, contradicted by evidence, subject denial | States the editorial conclusion directly |
| Unresolved questions | What remained open at the time of rejection | Documents known uncertainty rather than concealing it |
| Reviewer | Name and role of the editor who made the decision | Establishes editorial accountability |
| Decision and reason | Reject / hold / refer to legal, with stated rationale | Creates the formal record of the outcome |
| Follow-up | Whether the claim will be revisited, monitored, or referred | Prevents the rejected claim from disappearing without resolution |
How it works
- 1Document the original claim: what was it, when was it submitted, who submitted it
- 2List every source the AI cited or implied as evidence
- 3Record the verification steps taken: what was checked, by whom, and what ConvergePanel returned
- 4Note model agreement and disagreement on the core claim
- 5State the specific weakness that prevented publication: missing corroboration, contradicted evidence, insufficient sourcing, subject response that changes the claim
- 6Document any open questions that were not resolved before the rejection
- 7Identify the reviewer who made the rejection decision and their role
- 8Record the editorial decision, the stated reason, and the date
- 9Note any required follow-up: whether the claim will be monitored, whether a source will be contacted again, whether it can be revisited with additional evidence
Use cases
- Creating a defensible record of why a newsroom chose not to publish an AI-sourced allegation
- Building an institutional audit trail that shows the editorial process for rejected claims
- Preparing documentation in case a rejected claim later appears in another publication
- Training journalists and editors on what constitutes a sufficient rejection record
Why Documentation of Rejection Matters
When a newsroom investigates a claim and decides not to publish, two things are true simultaneously: the claim was reviewed seriously, and it was found insufficient. Both parts of that record matter. The review record demonstrates due diligence. The rejection reason demonstrates editorial judgment. Together, they create a defensible position if the claim surfaces later.
Without documentation, 'we looked at it and decided it wasn't publishable' is a statement with no supporting evidence. With documentation, it is a reviewable decision that can be examined against the record of what was found.
What Is Not a Sufficient Rejection Record
- 'We did not trust the source' — states a conclusion without the basis
- 'The AI said it was unverifiable' — documents a tool output, not an editorial review
- 'It felt wrong' — editorial instinct is valid but cannot stand alone as a documented reason
- 'We couldn't confirm it' — does not state what was checked or why confirmation failed
- 'The legal team said no' — is a decision point, not a review record; the legal team's concerns should be documented
- No record at all — leaves the newsroom with no defense if the claim surfaces elsewhere
How ConvergePanel Supports the Rejection Record
When a claim is run through ConvergePanel, the output includes model agreement, per-model evidence, disagreements, and the points where evidence is absent or contested. This structured output becomes the documented evidence review — more precise and reviewable than 'we searched and didn't find support.'
The ConvergePanel run does not make the rejection decision. The editor makes that decision. ConvergePanel provides a structured, exportable record of what was found in the AI research phase — one component of a complete rejection record.
Frequently asked questions
Does every rejected claim need to be documented?
Not every claim requires the same level of documentation. A claim rejected immediately because it was obviously unsubstantiated may need a short note. A claim that was seriously investigated before rejection — involving documentary review, source contacts, and editorial review — requires a complete record. Apply higher documentation standards to claims that involve allegations, that carried significant investigation time, or that could resurface.
How long should a rejection record be retained?
Retention depends on editorial policy and jurisdiction, but as a general principle: retain rejection records for claims that involved significant investigation for at least as long as the claim could plausibly resurface or lead to a challenge. For allegations involving named individuals, retention should align with the limitation period relevant to your jurisdiction.
What if the same claim appears in another publication after we rejected it?
A documented rejection record shows your newsroom reviewed the claim and found it insufficient. That record becomes relevant to your editorial response. If new evidence appears in the other publication that was not available during your review, document that the evidence is new and decide whether to revisit the claim under the same review process.
Can the rejection record be created after the fact?
A contemporaneous record — created at the time of the rejection — is more credible than one reconstructed later. After-the-fact documentation can omit what was not remembered, alter emphasis in light of later events, and cannot demonstrate that the decision was based on what was known at the time. Create the record when the decision is made.
Does ConvergePanel generate the rejection record automatically?
ConvergePanel generates an exportable record of the AI research phase: what was queried, what models found, where they agreed and disagreed. That record needs to be combined with the editorial components — reviewer name, editorial reason, decision, and follow-up — to constitute a complete rejection documentation. The editorial judgment is added by the journalist or editor, not by the tool.
What if the story is published anyway, with the open question still unresolved?
That's a different record. This page covers documenting a decision not to publish. If the decision is to proceed with publication while a specific fact remains open, see how to document unresolved facts before publication — it covers the qualification language, reviewer sign-off, and follow-up tracking for that scenario instead.
Explore related pages
ConvergePanel provides AI-assisted verification for informational purposes only. Not forensic analysis. Not legal evidence.
More in Governance
AI Governance Workflow for Enterprise Teams
Enterprise AI governance: automatic policy checks, peer review workflows, and full audit trails. ConvergePanel makes AI verification auditable.
AI Peer Review for High-Stakes Workflows
Use AI peer review to compare models, surface disagreement, document review notes, and create decision receipts for serious work.
AI Trust Dashboard for Decision Support
Use trust signals, model agreement, disagreement, source review, and audit trails to support AI-assisted decisions.