How we choose, verify and publish.
Our method is documented so you can judge the quality yourself — and know exactly when AI is involved, when a human read it last, and what happens if we get something wrong.
- Primary source required
- Always
- Editorial review
- Before publish
- Corrections
- Logged openly
- Coverage
- Sweden · EU
From signal to published article
Six steps, every article — no shortcuts. Automation level is labelled on every published story.
- 01 · Signals
Every day we collect signals from official sources (European Commission, ENISA, NIST, MSB, IMY), research institutions, vendor newsrooms and trusted aggregators. Sources are scored by historical accuracy and independence.
Around 400 sources in the weighted pool.
- 02 · Selection
A signal becomes a story when (a) it is new, (b) it is relevant to European readers, (c) at least one primary source exists, and (d) it fits our beats: Regulation, Tools, Models, Safety, Ethics, Research.
We do not publish rumours or unconfirmed leaks.
- 03 · Drafting
An LLM (typically GPT-5 or Gemini 2.5 Pro) writes a structured draft from the primary source and contextual research. The template requires TLDR, what changed, why it matters, who is affected, EU status and cited sources.
A draft is never a finished article.
- 04 · Editorial review
An editor verifies every factual claim against the source, checks quotes, edits language and assesses EU availability. We read the primary source — not just the summary.
Nothing publishes until a human has signed off.
- 05 · Publish & index
On publish we notify search engines via IndexNow (Bing, Yandex, DuckDuckGo) and Google Indexing API. The article is added to our news sitemap, category feeds (RSS + JSON Feed) and relevant topic indexes.
Machine-readable metadata: schema.org, dateModified, dateline.
- 06 · Follow-up
Major stories are updated when new facts arrive — dateModified is refreshed so search engines see the text as live. Factual errors are corrected transparently on /rattelser with a visible note on the article.
We never let a mistake stand uncorrected.
How we weight sources
Not all sources are equal. This is how we grade them, from highest trust to lowest.
| Tier | Examples | Weight | How we use it |
|---|---|---|---|
| S — Primary | EU regulations, regulator decisions, peer-reviewed research, official press releases. | 1.0 | Sufficient as a sole source. Always cited with a link. |
| A — Vendor | OpenAI, Anthropic, Google, Mistral, Microsoft — first-party posts and release notes. | 0.8 | Acceptable for product facts. Never the sole source for regulatory or safety claims. |
| B — Established journalism | Reuters, FT, Bloomberg, DN, SvD, Politico, TechCrunch. | 0.6 | Complements a primary source. We never link blindly — we read their sources too. |
| C — Analysts & researchers | Gartner, Stanford HAI, Ada Lovelace Institute, EPFL, KTH, RISE. | 0.5 | Context and assessment — clearly labelled as opinion or analysis. |
| D — Social signals | Posts on X, LinkedIn, Reddit, Hacker News. | 0.2 | Only as a first-signal lead. Never a sole source in a published story. |
AI in the newsroom — what it does and doesn't do
We use AI daily. That's the point. But the rules for when, how and what are strict — and we label every article accordingly.
What AI is allowed to do
- Draft structured articles from primary sources.
- Summarise long documents (e.g. AI Act articles) that an editor then verifies.
- Suggest headlines, TLDRs and categorisations.
- Extract data points (versions, prices, EU availability) that an editor confirms.
- Translate between Swedish and English — always human-reviewed.
What AI must never do
- Invent quotes, numbers or sources.
- Publish without a human reading it last.
- Write opinion pieces, analysis or columns.
- Replace editorial judgement in ethically sensitive stories.
- Report on personal data or ongoing criminal investigations.
Automation level on every article
Every published story carries one of three labels so you immediately see how it was made.
- Written by a human
A journalist wrote the piece. AI may have been used for research or translation.
- AI draft, editor-reviewed
The draft is AI-generated. An editor verified every factual claim against the source before publish.
- AI-assisted data update
Automated update of structured data (prices, versions, EU status). Verified by an editor at the next manual sweep.
What you can rely on
Four pillars hold the method up: editorial independence, open corrections, financial transparency and a traceable publishing chain.
Editorial independence
No advertiser, sponsor or vendor influences selection or angle. Sponsored content is clearly labelled — and always separate from editorial.
Open corrections log
All factual errors are corrected on /rattelser with a timestamp, what was wrong and what is right. We never delete an article to hide a mistake.
Financial transparency
Aheadline is funded by subscriptions and Hasty AB. We openly list vendors we use and disclose commercial relationships with companies we cover.
Traceable chain
Every article carries author, reviewer, publish date, last-updated date and a short changelog when material has been revised.
Frequently asked questions about the method
The questions readers, researchers and other newsrooms ask most often.
- Do you use AI to write your articles?
- Yes, often as drafts from primary sources. But an editor verifies every factual claim and signs off before publish. The automation level is visible on every article.
- How quickly do you correct errors?
- We aim for within 24 hours of confirmation. The correction is logged on /rattelser and the article carries a visible correction note — we never remove an article quietly.
- Are you independent from vendors like OpenAI, Anthropic, Google?
- Yes. We use their products for research and operations, but they have no editorial input. When we have commercial relationships with companies we cover, we disclose it in the article.
- How do you make sure AI drafts don't hallucinate?
- Every factual claim must have an identifiable primary source in the draft. The editor checks against the source. Uncertain quotes or numbers are cut — better a gap than a fabrication.
- Can you cover a specific topic I'm working on?
- Email the editors. We don't take contract journalism, but we welcome tips and questions that shape what we prioritise next.
See something that looks wrong? Tell us.
We're grateful for every proofreader. Send an email with the article link and what caught your eye — we'll reply once it's been dealt with.
Email the editors