Back to Home

How We Rate AI News

Every story on Agentic Daily carries two independent ratings. The verdict rates the evidence behind the claims — not our opinion on whether the news is positive or negative. The conviction rates how strongly our desk stands behind its own read of the story. They move independently, and the combinations are where the honesty lives. Here is exactly how we decide.

Verdicts — rating the evidence

Breakthrough

Criteria

  • -An externally supported change that alters what professionals can build, buy, deploy, regulate, finance, or defend
  • -Never on a company claim alone — needs peer review, a regulatory or legal document, or multiple independent primary sources
  • -Covers capability and method advances AND market-shaping events: a model recall, a precedent-setting regulatory action, or a landmark deal can qualify
  • -Not merely a scale-up of existing techniques with marginal gains
Example

A new architecture that halves inference cost at equal quality, with published benchmarks and reproducible code.

Incremental

Criteria

  • -A real improvement, but builds on existing approaches without a fundamental shift
  • -Performance gains are modest or limited to narrow use cases
  • -Useful for practitioners but unlikely to reshape the competitive landscape
  • -Typically announced via blog post or product update rather than research paper
Example

A model update that improves coding benchmarks by 5% while keeping the same architecture.

Overhyped

Criteria

  • -Marketing claims substantially exceed what the evidence supports
  • -Benchmarks are cherry-picked, non-standard, or lack independent verification
  • -Framed as a breakthrough when the underlying technique is well-established
  • -Press coverage amplifies the claim without critical analysis
Example

A product launch claiming "AGI-level reasoning" based on a proprietary, non-reproducible internal benchmark.

Verified

Criteria

  • -Key factual claims have been independently confirmed by our team
  • -Data points cross-referenced against primary sources (papers, filings, official announcements)
  • -Quotes and attributions verified for accuracy and context
  • -Does not imply the news is good or bad — only that the facts check out
  • -Magnitude-neutral: reserved for pure fact events (deals, shutdowns, filings, rulings) where a capability framing does not apply — never used as a hedge between Breakthrough and Incremental
Example

A reported funding round confirmed against SEC filings and official company statements.

Confidence — how sure we are

Below each story sits a proof line: Evidence · Confidence. The verdict rates how big the change is. Confidence (High / Medium / Low) rates how sure we are, and it is driven by source strength, not by how strongly we hold our view.

High — peer-reviewed research, regulatory or legal documents, or multiple independent primary sources.

Medium — a single primary source plus reputable reporting, or a strong claim pending document or methodology linkage.

Low — a company claim only, an unreproduced benchmark, or a self-designed benchmark.

Confidence is independent of the verdict. A Breakthrough can be Medium confidence: the change is real and market-shaping, but the documents are not all linked yet. That independence is why Verified is never a hedge — we can confirm a story is true while being plain about how strong the evidence behind our read is.

Our Process

1

Source Review

Every story is traced to its primary source: the original paper, press release, SEC filing, or official announcement. We do not rate stories based on secondary coverage alone.

2

Claim Assessment

We evaluate the core claims against available evidence. For technical claims, we look for reproducible benchmarks, peer review, or open code. For business claims, we cross-reference public filings and data.

3

Editorial Decision

A verdict and a conviction level are assigned based on the criteria above. Both reflect the evidence available at publication time and may be updated as new information emerges.

Corrections & Updates

Verdicts reflect the evidence available at publication time. If new information emerges that changes our assessment, we update the verdict and note the change at the top of the article. A verdict change is a feature of the rating system, not an embarrassment. If you believe a rating is inaccurate, contact us at support@agenticdaily.ai.