Investigative journalism finds and verifies facts of public relevance, particularly when an important claim is difficult to establish from routine announcements. It is not a licence to treat suspicion as guilt: documents, interviews, chronology and independent checking must support the published account.
This Gold Standard guide takes readers from a reportable question to a line-by-line evidence check, with original fictional procurement and rate-comparison examples. It explains source protection, public response, legal review and the limits that should remain visible when evidence is incomplete.
Chapter index
From lead to records and original evidence · Chapters 1–8
- What makes an investigation different from a headline
- Begin with a specific answerable public question
- Build an evidence ledger, not a folder of rumours
- Lawful public records are a starting point, not a verdict
- Source reliability depends on access and motivation
- Build a timeline without inferring causation from sequence
- Worked case: contract payments and the hidden assumptions
- Data tables have units, populations and missing rows
Calculation, corroboration and careful reporting · Chapters 9–16
- Worked example: raw complaint counts versus rates
- Interviews test records as well as tell human stories
- Right of reply is part of verification
- Confidentiality and source safety require professional judgement
- Independent corroboration means different chains of evidence
- Provenance determines whether a document is meaningful
- Data anomalies are questions, not verdicts
- Cross-border work adds legal and contextual differences
Editorial safeguards and learning transfer · Chapters 17–24
- Editing should challenge the strongest version of the story
- Legal and safety review can alter publication choices
- Write the report around verified findings
- Disclose methods and meaningful limitations
- Corrections and accountability continue after publication
- Independent practice: test four reporting claims
- Transfer the method when the evidence format changes
- Frequently asked questions and connected learning
1. What makes an investigation different from a headline
Investigative journalism is sustained reporting that develops evidence about matters of public significance, particularly when important facts are difficult to establish or have not been adequately scrutinised. It is not defined by dramatic language, a reporter’s suspicion or the number of accusations a story contains. The characteristic work lies in the careful question, independent evidence, systematic verification and accountable publication.
A routine announcement can report that an institution introduced a new policy. An investigation may ask whether the policy was implemented as promised, what records reveal, who was affected and which explanations survive checking. The answer might confirm the official account, reveal an error or find that the evidence is insufficient. Reporting honestly about an unproven allegation is different from proving one.
The Global Investigative Journalism Network’s fact-checking guide stresses scrutiny of each claim and fair opportunities to respond. This guide develops that method using fictional records and explicit evidence limits, without treating mere suspicion as a discovery.
2. Begin with a specific answerable public question
‘The entire agency is corrupt’ is not an adequate reporting hypothesis. ‘Did a particular procurement decision follow the published approval process during this period?’ is narrower and makes source requirements visible. A useful question identifies the actor, action, relevant dates, authority and information that could disprove the concern.
A newsroom may begin with a tip, complaint or unusual dataset. Those are leads rather than established findings. Before assigning blame, identify what could have happened for innocent administrative reasons: delayed paperwork, a contract variation, a different accounting basis or an inaccurate source record. The alternatives determine what must be checked.
Write a short question memo containing the claim, public relevance, known facts, uncertainties and possible sources. It keeps the investigation from expanding into unrelated personal details just because the reporter finds them interesting. The Research root explains how a defined question improves method selection.
3. Build an evidence ledger, not a folder of rumours
Every consequential statement in an investigative draft should connect to a record, observation or source whose basis of knowledge is understood. An evidence ledger can track claim, source, date, provenance, corroboration, unresolved contradiction and current verification status. This is an editorial tool for accountability, not a substitute for judgement.
A document may show what an institution recorded, but not automatically whether the record was complete or correctly described. An eyewitness may know what happened in one room but not the motive of someone elsewhere. Two accounts that copied the same original claim are not independent corroboration. Ask how each piece of evidence came to exist.
The GIJN introduction to finding sources and public records advocates combining documents and interviews rather than relying exclusively on either. For each finding, consider which alternative explanation it rules out and which explanation remains plausible.
4. Lawful public records are a starting point, not a verdict
Depending on jurisdiction and the question, relevant material may include procurement notices, budgets, meeting minutes, annual reports, regulatory filings or court records available through lawful public channels. Access rules, publication practices and privacy protections vary. A public record does not authorise publication of every personal detail it contains.
A reporter should identify the issuing organisation, document type, publication date, version and scope. A draft tender, executed contract, amended contract and payment record can refer to different stages and amounts. Treating one as interchangeable with another can manufacture an apparent contradiction.
Use publicly available information and authorised requests. Never bypass access controls or pressure someone to obtain material unlawfully. A responsible investigation can produce important findings by patiently reconciling ordinary documents and asking clear questions of the accountable institution.
5. Source reliability depends on access and motivation
A source can be sincere and still be mistaken. Ask what the person personally observed, what they learned from others, which records they saw and why their perspective matters. Distinguish a fact witness, expert interpreter, whistleblower, spokesperson and interested party. These roles overlap, but each supplies a different kind of evidence.
Motivation is a factor for evaluation, not a reason to dismiss someone automatically. A dissatisfied employee may possess authentic records; an official may have accurate operational knowledge; either may omit important context. The report should check the underlying claim with independent material whenever possible.
Confidentiality decisions deserve serious editorial review. A promise to protect someone’s identity cannot be made casually or presumed to eliminate legal or practical risk. Reporters should seek qualified guidance on source safety, applicable laws and responsible record handling before encouraging vulnerable people to disclose information.
6. Build a timeline without inferring causation from sequence
Timelines help reveal what was known or done at particular dates. But an event occurring before another does not prove it caused the later event, and a document’s publication date may differ from its signature, effective or transaction date. Investigative accounts must state which clock each date describes.
Imagine a fictional contract approved in March, an invoice recorded in April and a payment made in May. That sequence can be routine. If a complaint was filed in June, the timing alone does not show the complaint caused or prevented a payment. What matters is the actual decision process, authority and supporting records.
Record uncertainties directly on the timeline, including when a document is missing or its version is unclear. A neat graphic that silently fills those gaps can become more misleading than a text note admitting incomplete evidence. The History root offers complementary source-context habits.
7. Worked case: contract payments and the hidden assumptions
Consider a fictional procurement file. Three payment entries refer to one project: 18,000, 22,000 and 30,000 monetary units. Their sum is 70,000. A reporter also finds an original contract summary stating an initial ceiling of 65,000. The arithmetic difference is 5,000, but the numbers alone do not demonstrate fraud, misconduct or unlawful expenditure.
The project may have authorised amendments, taxes, a different accounting scope, partial refunds or entries that belong to other periods. Conversely, some discrepancies can remain serious after those explanations are checked. The reporter needs the executed contract, variation records, payment basis and appropriate response from the accountable institution before choosing a headline.
Under the fictional assumptions, the responsible statement is ‘Three recorded payments total 70,000, compared with the original 65,000 ceiling; the status of amendments has not yet been established.’ This preserves a real observation while identifying precisely which missing evidence might change its meaning.
8. Data tables have units, populations and missing rows
Spreadsheets can make a story appear objective because they contain many numbers. Yet an export can omit cancelled transactions, divide one contract across several codes or change classifications between years. Reconciliation requires checking field definitions, date ranges, duplicate rows and which records are not available.
Before adding values, decide whether the figures are commitments, invoices, actual payments or planned budgets. Summing them as one expenditure amount may count the same money several times. Inspect sample records against original documents and maintain a transparent method for any transformations.
A calculation can be independently reproducible while its interpretation is still uncertain. The Statistics and Information Science roots help clarify definitions and data structure. The journalistic task is to communicate the evidence without laundering an inappropriate calculation into a conclusion.
9. Worked example: raw complaint counts versus rates
Suppose a fictional service reports twelve complaints from sixty transactions in Period A and eight complaints from twenty transactions in Period B. The complaint count falls from twelve to eight, but the corresponding rates are 12 ÷ 60 = 20% and 8 ÷ 20 = 40%. A headline about a one-third reduction in complaints would hide the different transaction volumes.
This comparison also has limitations. The definition of a complaint, its severity, willingness to report and whether transactions were comparable may have changed. A doubled complaint rate does not by itself prove why service quality changed. The small sample in Period B may also create substantial statistical uncertainty.
Publish both counts and denominators with the relevant dates. A competent report can state that recorded complaints fell in number while the percentage of transactions generating complaints rose under the specified definitions, without inventing a simple causal explanation.
10. Interviews test records as well as tell human stories
Documents may show an official process, while participants may explain how it operated in practice. Interviews help identify missing records, clarify procedures and test whether a numerical pattern corresponds to people’s experiences. The interviewee’s narrative still requires verification and context, especially when allegations affect others.
Distinguish personal observation from inference, rumour and interpretation. Ask when and where the event occurred, which document or person can corroborate it and what details remain uncertain. Avoid leading questions that assume the alleged misconduct has already been proved. The reporter should not coach someone to provide a preferred answer.
Use interviews to challenge the evidence ledger, not only to supply vivid quotations after the conclusion is fixed. A reluctant or inconvenient response may reveal the most important flaw in the original hypothesis. The GIJN guide to background research and public records emphasises interplay between interviews and documents.
11. Right of reply is part of verification
People and institutions facing serious factual allegations should ordinarily receive a meaningful chance to address the specific claims before publication, with due consideration for the investigation’s safety and circumstances. A vague message saying ‘Any comments?’ may fail to put the actual issue fairly. A responsible request distinguishes the facts at issue and offers a reasonable route to respond.
Responses can supply missing amendments, explain methods or identify errors, and journalists should check those assertions with the same care applied to the original allegation. A denial is not proof the concern was unfounded, just as a failure to reply is not proof it was true. The final story should accurately explain what was asked and what response was received.
The GIJN fact-checking guidance explains the value of detailed and fair engagement with subjects. Where safety risks make ordinary contact inappropriate, reporters need specialist editorial and legal judgement rather than a universal checklist.
12. Confidentiality and source safety require professional judgement
A source may fear retaliation or exposure, but anonymous reporting carries editorial and practical risks. Before receiving sensitive material, journalists need suitable organisational protocols and qualified advice. No communication channel should be advertised as guaranteeing safety, and no promise of anonymity should be made without understanding what the reporter can realistically protect.
An investigation can sometimes use public or lawfully authorised records without obtaining unnecessary identifying information. Limit circulation of personal details within the reporting team and use appropriate access controls. Avoid asking someone to breach laws or professional duties merely to strengthen a story.
The correct question is not how to acquire the largest possible collection of secret records. It is which evidence is needed, how it can be obtained responsibly and how people who could be harmed will be protected. These ethical limits are part of the quality of the investigation, not obstacles to be dismissed.
| Item | Amount |
|---|---|
| Three recorded payments | 18,000 + 22,000 + 30,000 = 70,000 |
| Original stated contract ceiling | 65,000 |
| Difference requiring documentary explanation | 5,000 |
| Period | Complaints | Transactions | Rate |
|---|---|---|---|
| A | 12 | 60 | 20% |
| B | 8 | 20 | 40% |
13. Independent corroboration means different chains of evidence
Three news items repeating one anonymous post are not three independent confirmations. Corroboration becomes stronger when sources have genuinely separate access to the relevant facts, such as a primary document, a firsthand observation and a separately collected institutional record. Their agreement matters only if they refer to the same event, date and definition.
An official answer may verify a procedural fact but not the effect of a decision on affected people. An eyewitness may describe an encounter but not a board’s intent. Researchers should identify each source’s actual scope and avoid treating apparent consistency as proof that all aspects of the story are settled.
When evidence conflicts, preserve the contradiction until it can be explained. Do not quietly drop the less convenient record. A good evidence ledger makes the unresolved question visible and identifies which additional lawful check would change the conclusion.
14. Provenance determines whether a document is meaningful
A screenshot of a page may omit a revision note, date or surrounding table. A file name may not reveal whether the document is final. Before relying on a record, ask who issued it, when it became effective, whether it was amended and whether a newer version exists. The provenance problem is particularly serious when old consultation material circulates as current policy.
Use original official publications where accessible and document any limitations in access. If only a partial extract can be verified, describe it as partial. Quoting a sentence accurately does not make it representative of a full document; surrounding clauses can qualify what the sentence means.
An investigation of historical material also needs chronology. The History root develops source criticism. Journalistic verification adds the urgency of determining what can responsibly be reported now without mistaking a draft, announcement and implemented decision for the same event.
15. Data anomalies are questions, not verdicts
An unusual payment, sudden score change or missing record can be worth investigating. It can also reflect a data-entry error, different reporting periods or a legitimate administrative change. Calling an anomaly ‘proof of fraud’ skips the very work that distinguishes investigation from accusation.
Reproduce every calculation with the appropriate units and denominators, then check underlying source records. Where an apparent pattern depends on one outlier, investigate that observation rather than deleting it or building the whole story on it. A graph may help identify questions, but it cannot establish motive or unlawful intent.
Write cautious interim language: ‘The records show a discrepancy that has not yet been explained’ is not the same as ‘The institution misused funds.’ Publish the stronger claim only when sufficient credible evidence supports it and the applicable legal terms are used accurately.
16. Cross-border work adds legal and contextual differences
Investigations can involve organisations, documents or people across jurisdictions. Public-access rules, data-protection duties, reporting restrictions and defamation laws can differ. A disclosure lawful in one country may create serious legal or safety concerns elsewhere. A team should not assume that a familiar domestic practice applies unchanged across borders.
Names, currencies, accounting periods and institutional responsibilities also vary. Apparent discrepancies can arise from conversion dates or different definitions of financial years. Translate documents carefully and verify technical terms before asserting equivalence between agencies or regulatory statuses.
The International Law and Translation roots supply related context. Specialist local legal advice may be necessary for consequential publication; a general educational article cannot determine the law of every jurisdiction.
17. Editing should challenge the strongest version of the story
A good editor asks what would make the main conclusion false. Are there documents the reporter has not examined? Are dates consistent? Could the chosen denominator make a numerical claim misleading? Has the subject’s strongest documented counterargument been fairly evaluated? These questions should be answered before a dramatic opening is perfected.
The GIJN investigative editing guide emphasises comprehensive fact checks and coherent explanation. A narrative that is exciting but depends on unsupported transitions is not ready for publication. The editor must verify what connects each scene, figure or quotation to the next conclusion.
Retain a record of substantive edits and reasons. If a claim is narrowed after a response or corrected record, the revised story should reflect that change throughout, including headline and graphics. A caveat hidden at the end does not repair an opening that still asserts the original stronger claim.
18. Legal and safety review can alter publication choices
Before publication, a high-risk investigation may require qualified legal, editorial and security review. The relevant issues can include defamation, privacy, confidentiality obligations, reporting restrictions, source protection and foreseeable threats. The purpose is to improve accuracy and minimise unnecessary harm, not to suppress verifiable public-interest reporting reflexively.
A risk review asks which names, images and granular records are essential to the story. Sensitive details can sometimes be omitted while preserving an accountable description of institutional decisions. Other details may be legally and publicly important and require careful evidence. There is no single universal rule.
An editor should know when the investigation lacks the evidence needed to publish a consequential accusation. Delaying, narrowing or abandoning that claim can be the responsible outcome. The Society of Professional Journalists’ Code of Ethics provides complementary principles for truthfulness, independence, harm reduction and accountability.
19. Write the report around verified findings
The structure of an investigative story should make the established finding understandable early, then show the records, people and context needed to evaluate it. A dramatic chronology is not always the clearest order. Readers may first need to know which institution, contract or decision is being discussed and what the records actually say.
Distinguish allegation, official response and independently verified fact through attribution and careful language. A quotation can reveal the source’s position, but it does not necessarily establish the underlying claim. A graphic showing a discrepancy must identify the amounts and accounting basis rather than draw an accusatory visual conclusion unsupported by the records.
The Journalism and Writing roots support clear public explanation. Editorial craft should help readers examine evidence, not make weak evidence feel decisive.
20. Disclose methods and meaningful limitations
An investigative piece should explain where central information came from, how records were compared and what could not be verified, while protecting legitimate confidential sources or restricted material. An assertion such as ‘Our investigation reveals’ is not itself a method. Readers need enough transparent detail to understand why the publication reached its conclusion.
For data-driven reporting, specify the time window, population, denominators, excluded records and method of calculation. For interviews, identify relevant roles and the nature of firsthand knowledge where doing so is responsible. A summary can be readable without exposing private identities unnecessarily.
Do not claim that a publicly accessible dataset is complete unless its collection and coverage justify the assertion. Limits can be substantial and still allow a useful, narrower finding. Accuracy is preserved when the story states the scope it can support.
21. Corrections and accountability continue after publication
A published investigation may later receive new evidence, corrections or relevant responses. A responsible newsroom distinguishes a factual correction from a subsequent update and makes important changes visible. The headline, chart and explanatory text must be reconciled; repairing one paragraph while leaving a false summary in place is insufficient.
Set an accessible channel for factual challenges and document how they were reviewed. A complainant’s objection does not automatically invalidate a verified story, but it can point to a misunderstood document or omitted fact. Correct the record proportionately when evidence demonstrates an error.
The SPJ Code of Ethics treats accountability as part of professional practice. A serious correction can strengthen an information process when it reveals a commitment to evidence rather than preserving appearances.
22. Independent practice: test four reporting claims
Claim one: ‘The project exceeded its original 65,000 ceiling by 5,000.’ Recorded payments total 70,000, but what else must be checked? Look for contract amendments, the precise accounting basis, refunds and relevant approvals. Claim two: ‘Complaints dropped from twelve to eight, so service quality improved.’ With transaction counts changing from sixty to twenty, the recorded rates actually change from 20% to 40%.
Claim three: ‘Three websites independently confirmed an allegation.’ Each copied the same original social-media post. The supposed corroboration is not independent. Claim four: an institution declines an interview. Does its silence prove the allegation? No. The story should describe the request fairly and continue evaluating the underlying evidence.
For each answer, name which source or method could change the conclusion. This is the difference between identifying a red flag and establishing what it means.
23. Transfer the method when the evidence format changes
Return to the fictional procurement discrepancy. Instead of a downloadable table, the initial lead is now a photograph of part of a printed invoice. The main public question remains, but the verification method must change: the image may omit the page’s context or reflect a draft, duplicate or different accounting period.
Seek the complete lawful source and verify how the document fits the original contract and payment system. Do not treat image appearance or a machine-generated transcription as proof that the record is authentic. A responsible account may need to say that the available extract is insufficient for a conclusion.
Next imagine the same project crosses two countries with different currencies and reporting periods. Recalculate the comparison only after aligning dates, units and legal context. Sound principles transfer, but their application changes with the evidence.
24. Frequently asked questions and connected learning
Does an unexplained discrepancy prove corruption? No. It is a question requiring source reconciliation and fair investigation. Is an anonymous source automatically unreliable? No, but the reporter must understand the basis of knowledge, corroboration and safety concerns. Is one official denial enough to close every inquiry? Not necessarily; the relevant evidence must be evaluated on its merits.
Can investigative journalism use public records alone? Some findings can be developed from records, but people and context often help establish meaning and resolve apparent contradictions. Must every allegation be published? No. Sufficient evidence, lawful reporting, public interest and proportionate harm all matter.
For deeper reading, connect Journalism, Media Literacy, Research, Statistics and Ethics. The GIJN fact-checking guide is a primary professional reference. All worked cases here are fictional and imply nothing about real people or institutions. Return to The Gold Standard of Education for the skill of revising a claim when the evidence changes.