Create a claim inventory

Read the draft once without polishing it. Highlight every statement that a reasonable reader could challenge: numbers, dates, rankings, quotations, causes, superlatives, product capabilities, legal requirements, and descriptions of research. Include claims that feel obvious. Familiarity is not evidence, and stale facts often hide in introductory sentences. Place each claim in a simple table with its location, importance, source status, and owner. Mark high-risk claims—health, money, law, safety, or reputational allegations—for specialist review. This inventory prevents attractive wording from distracting you from the article’s factual load.

Match the source to the claim

Use primary sources when possible: the original paper, official dataset, regulatory text, court document, company documentation, or direct interview. A search snippet is a discovery aid, not evidence. Secondary reporting can provide context and help interpret technical material, but trace important facts back to their origin. Check the publication date, the date of the underlying data, geography, sample, definitions, and version. A statistic about “workers” may refer only to surveyed office employees in one country. If the source cannot support the exact scope of the sentence, narrow the sentence.

Verify quotations and numbers in context

For a quotation, inspect the surrounding passage or listen to the recording. Confirm speaker, wording, date, and whether edits preserve meaning. For numbers, reproduce simple calculations when possible and check units. Percentage and percentage-point changes are not interchangeable. An average may conceal a wide range, and a forecast is not an observed result. Compare a surprising value with another credible source or earlier edition. If two reputable sources disagree, investigate definitions instead of selecting the more dramatic figure. Readers benefit from knowing why estimates differ.

Calibrate language to evidence

Fact-checking also changes verbs and qualifiers. A controlled experiment may support stronger causal language than an observational survey. One company’s internal report does not establish an industry-wide trend. Replace “proves” with “suggests” when that is what the evidence allows, and state meaningful limitations without burying them. Avoid vague shields such as “experts say” or “studies show.” Name the institution or describe the body of research. If evidence is mixed, say so and explain the source of uncertainty. Honest qualification builds more trust than artificial confidence.

A useful workflow makes quality easier to repeat, not merely easier to describe.

Keep an editorial audit trail

Record the supporting URL, document title, access date, and relevant page or section beside each claim. Note whether the claim was confirmed, revised, removed, or sent for further review. Before publication, have someone other than the drafter spot-check the highest-impact items. After publication, provide an easy correction channel and update material errors transparently. For fast-changing topics, schedule a review date rather than allowing an article to age silently. The audit trail saves time during updates and makes the newsroom less dependent on one person’s memory.

Make the method your own

Run the workflow on a small, real assignment and keep a short decision log. Record the starting problem, the instruction you gave the tool, what the first result missed, and why you accepted the final version. This takes only a few minutes, but it separates repeatable learning from accidental success. After three projects, review the notes for patterns. You may discover that better source material matters more than longer prompts, or that a particular review step catches most quality problems. Keep the useful pattern and remove the ceremony. A personal workflow should become simpler as your judgment improves.

Small experiment: Apply one idea from this guide within the next seven days. Keep the scope narrow enough to finish, compare the result with your previous approach, and write down one change you would make next time. Note what surprised you and which assumption proved wrong. Share the test with one trusted reader if the context allows. Practical evidence is a better teacher than an endlessly refined plan, and a completed test gives your next creative decision a firmer foundation.

Try this in your next session

1. Choose one real project rather than a hypothetical exercise.
2. Change one variable at a time and save the result.
3. Write a short note explaining why the selected version works.

Final takeaway

A reliable article is not one with a citation attached to every paragraph. It is one whose claims have the right scope, source, context, and level of certainty. AI can help identify statements to inspect and organize the evidence, but it cannot take responsibility for publication. A disciplined claim inventory and audit trail give editors the control that fluent generation can otherwise obscure.