What AI
can
replace (or compress)
- Curation: Build a personalized feed across blogs, forums, newsletters, filings, preprints, local papers—AI can summarize and dedupe.
- Synthesis: Juxtapose competing takes, extract points of agreement/dispute, and turn long threads into briefs.
- Drafting/production: Turn research into clear memos, explainers, or scripts; translate and localize niche sources.
- Discovery: Surface non-obvious sources (regional outlets, industry reports, municipal minutes) you’d never see on a front page.
What AI
can’t
replace
- Original reporting & access: On‑the‑ground interviews, FOIA work, courtrooms, war zones, whistleblowers.
- Accountability layers: Editors, legal review, corrections policies, reputational risk that keeps standards tight.
- Primary data quality: Many AI answers mirror their training data; if that data is wrong or skewed, so is the output.
- Context & stakes: Knowing when a “small” detail is the whole story often requires beat expertise.
A practical playbook to go “post‑mainstream” without going post‑truth
1) Build a primary‑source first diet
- Government/agency pages (legislation, budgets, statistics)
- Company filings and transcripts
- Preprints/journals and datasets
- Local outlets and trade publications
Use mainstream stories as cross‑checks, not your first stop.
2) Use AI with guardrails
Give instructions like:
- “Synthesize today’s developments on X from at least five non-duplicative sources; list each source with a one‑line credibility note. Separate facts from opinion. Flag any claim that lacks a primary document.”
- “Contrast mainstream framing vs. independent outlets on X. Show three concrete factual disagreements and who provides evidence for each.”
- “Extract all numbers with provenance and recreate the calculation that leads to the headline figure.”
- “Generate the three strongest counterarguments to this conclusion and the evidence required to falsify each one.”
3) Demand receipts
- Ask for links to primary documents (filings, bills, datasets, transcripts).
- Require the model to label each sentence as {reported by, inferred from, speculative}.
- Have it output a claims‑table: claim → evidence link → date → confidence → who benefits if true.
4) Triangulate quickly
- Run the same prompt through two different models; compare disagreements.
- Sample one mainstream outlet, one trade/indie, and one primary doc for the same story. Where they diverge is where you dig.
5) Red‑team your feed
- Periodically instruct: “Identify my blind spots. Which credible sources consistently disagree with my current set?”
- Rotate in dissenting sources on purpose; echo chambers feel efficient but decay judgment.
6) Know when mainstream still adds value
- Breaking crises, complex investigations, litigation, and war coverage often rely on resources only large orgs have.
- Use their work as a scaffold; validate key facts against the underlying documents they reference.
7) Keep an evidence ledger
Maintain a simple sheet with: topic, top claims, doc links, last verified date, remaining unknowns. It prevents “vibes” from becoming conclusions.
Quick-start prompt pack (copy/paste)
- Multi‑source brief
“Give me a 300‑word brief on [topic] from the last [N] days using at least [5] distinct sources. Separate facts/opinions, add links, and list 3 things we don’t yet know.”
- Disagreement map
“Map the top 5 points of disagreement on [topic]. For each: who says what, evidence cited, how to resolve, and a one‑line why this matters.”
- Number check
“Extract every statistic related to [metric] in [article/text]. Recompute from source documents and show your math. Flag inconsistencies.”
- Bias audit
“Audit my source list for ideological, geographic, and incentive bias. Propose 10 credible additions that push in different directions.”
If you want, tell me 3–5 subjects you care about and your preferred depth (e.g., “daily 5‑min brief” vs. “weekly deep dive”). I’ll generate a tailored, source‑backed digest that leans on primary and indie outlets—and clearly marks where mainstream cross‑checks still help.