Publish AI news quickly without manufacturing certainty.
AI news changes quickly, but speed does not excuse weak sourcing. A credible creator separates the event date from the article date, distinguishes a company claim from independent evidence, labels inference and updates the audience when facts change.
1. Build a source ledger before writing
Start with primary sources: official documentation, model cards, release notes, product announcements, research papers, regulatory text and direct demonstrations. News articles can add context, but they should not replace the source behind a technical claim.
Research [TOPIC] for a creator news report. For every source record: - URL and publisher - publication date - date the event happened - exact claim supported - whether it is a company claim, documented capability, independent test or analysis - geographic or account availability - pricing or plan limitations - known limitations - what remains uncertain Do not draft a headline until every headline claim is traceable.
2. Classify claims
A useful ledger labels information as confirmed fact, provider claim, measured result, informed inference or unknown. Avoid converting “rolling out” into “available to everyone.” Avoid presenting a benchmark as proof of all real-world use cases. Record model versions because a capability may not apply to an older product with a similar name.
3. Write a script that serves the viewer
Do not begin with company adjectives. Begin with the change and the practical consequence. A strong structure is: what happened, what is actually new, evidence, limits, who benefits, who should wait, and the next test.
- Cold open: the most important practical change, stated without exaggeration.
- Context: what existed before and why the difference matters.
- Evidence: documentation, demonstration or measured result.
- Constraints: access, cost, geography, model limits and failure cases.
- Creator test: one reproducible experiment the viewer can perform.
- Outlook: what to monitor without pretending to predict certainty.
4. Use transparent visuals
Prefer product interfaces, charts, timelines, source excerpts used within rights, and your own demonstrations. AI-generated B-roll can illustrate an abstract concept, but it must not appear to document a real event that did not happen. Add an on-screen “AI illustration” or “reconstruction” label when realism creates ambiguity.
Never generate a realistic executive quote, public demonstration, disaster, protest, medical event or product test and present it as footage. Never make a public figure appear to announce a release.
5. Treat voice and translation as editorial work
Use your own voice or a voice for which you have explicit permission. Maintain a pronunciation list for model names, companies and Myanmar terms. When translating, preserve uncertainty and attribution. A sentence that means “the company says” must not become “it is proven” in another language.
6. Edit for comprehension
Show the source or demonstration when the claim is spoken. Keep citations readable. Use a neutral audio bed and avoid alarm effects that inflate minor changes. Include version numbers, access limitations and dates on screen when they are essential.
7. Publish a correction path
Include source links, article update time, disclosure and a visible method for reporting errors. Use a pinned comment for material updates after publishing. If a provider changes access or corrects a claim, update both the description and the website article.
News prompt checklist
- Did every factual sentence survive source review?
- Are event date and publication date separated?
- Are company claims attributed?
- Are synthetic scenes clearly illustrative?
- Are pricing and availability current and scoped?
- Does the title promise exactly what the evidence supports?