EDUKATE SINGAPORE · GOLD STANDARD PRACTICAL CASE STUDIES · CASE 116
A quote may contain every correct word and still tell readers the wrong story.
A newsroom records a community-hall interview accurately but publishes an important sentence under the wrong speaker’s name. This fictional Gold Standard case explains source attribution, responsible corrections and verification across automated transcription and editorial handoffs.
Every journalist, interviewee, quotation and local institution is invented. Reuters is cited for published editorial guidance, not as the subject of this incident or as the fictional newsroom’s accreditor.
Chapter index
- 1. A perfect transcript and a public statement nobody made
- 2. The words and the speaker are two separate facts
- 3. A production pipeline can preserve text and corrupt meaning
- 4. Reuters’ published standards make attribution a central control
- 5. The headline correction has a separate clock
- 6. The people quoted deserve accurate distinctions, not equal publicity
- 7. Mark speaker changes where the source can support them
- 8. A reproducible editorial source chain avoids the next error
- 9. Rehearse a quote with correct words and wrong identity
- 10. Practice: who really said it?
1. A perfect transcript and a public statement nobody made
A fictional local newsroom interviews several people after a community learning hall is closed for building work. Its reporter speaks with a facilities engineer, the programme organiser and a parent representative. Each describes a different part of the situation: the engineer discusses the conditions for a technical decision, the organiser discusses possible programme dates, and the parent explains the effects on families.
The newsroom uses an automated transcript that reproduces their spoken words with high textual accuracy. During the interview, people occasionally speak over one another. The transcript changes speaker labels at one point, and the article editor does not compare the labels with the original recording before publication.
The next morning the headline attributes the engineer’s condition about reopening to the organiser. The quoted sentence is genuine, but the published attribution is wrong. The Gold Standard issue is that word accuracy does not establish source accuracy, and a reader needs both to judge what a public statement means.
2. The words and the speaker are two separate facts
In the recording, the facilities engineer says: ‘The hall should not reopen until the required review has been completed.’ That is a statement about a condition of technical readiness from a person acting in a particular role. The programme organiser says: ‘We hope to run classes again when the hall has been cleared for use.’ That is an expression of an intended programme outcome.
The journalist’s article prints the first sentence exactly but attributes it to the organiser. The editor correctly checked the vocabulary and punctuation against the text output. The source identity, however, was generated by the automated speaker-detection process rather than verified against the recording.
An exact quotation can still be misleading if the wrong person is made responsible for it. Source identity and role affect what a reader infers about authority, knowledge and accountability. A technical observation attributed to a public programme organiser may change who families contact and whom the public believes made a formal decision.
GS-R08 requires evidence that supports the precise claim. For a quotation, that includes the actual words, speaker, timing, relevant context and the conditions under which the words were said. A word-perfect transcript is only one link.
3. A production pipeline can preserve text and corrupt meaning
The newsroom has source recording J1 with a continuous timestamp; automatically generated transcript J2 containing text and speaker identifiers; reporter’s notes J3; edited story draft J4; and final published story J5. J1 may allow a competent reviewer to reconstruct the actual conversation. J2 is an intermediate representation, not the source authority.
A diarisation system can put accurate words beside the wrong label when speakers overlap, interrupt or have similar voices. Human editors can make the same mistake while consolidating notes. Neither mechanism automatically establishes malicious intent; both show why speaker identity needs verification when direct quotations are presented as authoritative statements.
The editorial workflow had an optional review for transcript wording but no mandatory checkpoint for the identity of a quoted speaker. The label became part of the final story without an accepted handoff between the person who conducted the interview and the person releasing the article.
GS-R07 asks whether information survives each handoff with its relevant meaning. If the sentence keeps its words but loses the person who said it, the information has not survived intact for publication.
4. Reuters’ published standards make attribution a central control
The Reuters Standards and Values state that accuracy takes precedence over speed, distinguish correct quotations from the context in which they were spoken, and require proper attribution to sources. Reuters also describes its practice of correcting errors transparently and comprehensively.
Those are published professional standards of one newsroom, not a claim that this fictional local publication is owned by Reuters or formally bound by its internal rules. They offer credible editorial context for why a quote cannot be considered correct solely because the words match an audio transcript.
Reuters explicitly connects proper sourcing to the public’s ability to assess reliability. A statement from an engineer responsible for a particular technical assessment conveys different authority from a statement by someone coordinating future classes. Readers should be able to see who supplied the information without the newsroom inventing that relationship.
The Gold Standard turns an ethical aspiration into a practical acceptance requirement: each significant quote must have a verifiable link to a recording or reliable notes, a correctly identified source and relevant context before the public is asked to trust it.
5. The headline correction has a separate clock
Once the editor learns of the misattribution, it matters how quickly and clearly the public record is corrected. The newsroom should preserve the source recording, original draft, published version and the evidence supporting the correction. A silent replacement that makes the mistake disappear from history may leave readers and affected people unsure what was changed.
The responsible editor confirms the actual speaker and revised attribution under the publication’s real policies. The corrected story identifies the error proportionately and distinguishes what the engineer said from what the organiser actually said. If other publications or social posts repeated the wrong attribution, relevant corrections may need to reach those channels.
The newsroom should not assert that the hall is legally safe or unsafe simply because a speaker made a particular remark. The substantive technical and permission decisions remain with the competent authorities under the actual premises arrangements. Journalism reports what sources said and examines their evidentiary basis; it does not create those authorities by printing a quote.
GS-R09 requires recognition and response to an exception. When public information guides families, a misleading attribution is not merely a style problem waiting for the next routine editorial meeting.
6. The people quoted deserve accurate distinctions, not equal publicity
The programme organiser may rightly worry that readers now believe they issued a technical restriction outside their role. The engineer may be concerned that their professional caution has disappeared from the report. The parent representative may be affected if the story implies a community consensus that nobody expressed.
The editor’s task is to correct what the source record shows, seek appropriate comment and preserve context without allowing any source to rewrite the story for convenience. A request for correction does not automatically justify giving each participant equal space, nor does it require suppressing verified uncomfortable statements.
Quotation accuracy involves both preserving the actual words and not choosing fragments that invert the speaker’s meaning. The newsroom should distinguish the engineer’s conditional professional position from the organiser’s conditional hope of holding classes again. Both can be included accurately because they answer different questions.
GS-R03 assigns editorial authority. Staff receiving a correction request need a clear route to competent verification rather than a negotiation in which the loudest source determines the final record.
7. Mark speaker changes where the source can support them
The newsroom revises its transcription process so automated speaker labels are treated as provisional until verified where attribution matters. A reporter or suitably competent editor listens to the relevant recording segment, checks the speaker’s identity, and confirms whether the quote accurately represents the intended context.
This does not require manually relistening to every second of every recording before any story can be published. The review should be proportionate to the materiality of attribution and the risks of misrepresenting a consequential statement. Sensitive or disputed claims deserve greater care than a minor background detail.
Where the recording cannot establish a speaker confidently, the article should not simply select the most plausible name. Staff can use suitably verified contemporaneous notes or other legitimate sources, report uncertainty accurately, or omit a direct quote whose source cannot be established.
The Gold Standard values an honestly qualified passage above a falsely precise attribution. A model’s high text-confidence score is not evidence that the correct human identity has been verified.
8. A reproducible editorial source chain avoids the next error
The revised story record includes the interview’s authorised source file, time reference, speaker identity as verified, transcript version, direct quotation, context notes and final published wording. Appropriate privacy, source-protection and retention rules apply; the reporter should not expose confidential interview material indiscriminately just to prove an internal control worked.
If a newsroom uses generated summaries as draft aids, the final article must still preserve evidence links and be reviewed for unsupported additions. The system should be able to distinguish a direct source quote from a reporter’s paraphrase or an inference made by a model or editor.
When multiple speakers overlap, a useful flag sends the relevant passage to a human reviewer rather than automatically publishing an unverified label. Staff should know who has authority to approve the correction and how to notify downstream platforms that hold an older copy.
GS-R12 governs change to the transcription and editorial workflow. Replacing a software provider should not silently change the newsroom’s standards for attribution.
9. Rehearse a quote with correct words and wrong identity
A controlled exercise gives editors a fictional interview recording with three voices. The transcript contains an exact sentence but assigns it to the person who spoke immediately before the true speaker. Reviewers must detect the mismatch and either establish the correct attribution or mark it uncertain before publication.
A second exercise uses an accurate quote removed from its qualifying sentence. Editors should recognise that word accuracy alone may not preserve meaning. A third introduces a corrected speaker label after a story is published and tests whether the newsroom’s correction route is accountable and visible.
Success is not measured by a 100% transcription similarity score. It is measured by whether the staff can verify the source, preserve context and present the correction responsibly. The methods should work under realistic deadline pressure without depending on one editor’s personal memory.
A passing rehearsal supports the particular tested process. Real interviews, disputed identities and technical content still require appropriate editorial judgement.
10. Practice: who really said it?
Exercise. A three-person interview covers reopening a teaching hall. An automated transcript accurately reproduces the engineer’s words about required review but labels them as spoken by the programme organiser. The published story repeats that attribution. Write the supported finding, immediate publication decision, verification route and retest.
A reasoned finding is: ‘The printed quotation is textually faithful to the recording, but its speaker attribution is not supported by the source. The newsroom must verify the recording and notes, issue an appropriate transparent correction and revise the editorial acceptance process so important automated speaker labels are checked.’
The next test should deliberately mislabel an accurate quote and require a reviewer to connect it to the correct identified speaker before release. The Gold Standard lesson is that truth in public information lives in the relationship between words, speaker, context and consequence—not in word matching alone.
Further reading, sources and limitations
- Reuters · Journalistic Standards and Values
- Reuters · Brand and Source Attribution Guidance
- Gold Standard · Information and Record Integrity
- Gold Standard · Evidence, Claims and Verification
- Gold Standard · Core Requirements and Evidence
- Case 19 · AI Assistant Invented a Policy
- Case 113 · AI Reconstruction Labelled a Historical Photograph
- Gold Standard Reference · Casebook
- Master Gold Standard Hub
Professional ethical guidance illustrates why words, sources and corrections matter. The GS-R01–GS-R12 Gold Standard programme is eduKate’s educational management framework rather than formal newsroom accreditation.