// ai for professions — journalists

AI for Journalists — Roadmap

39 topics across fundamentals, verification, audio and sources, documents, data, the newsroom, ethics, and an advanced track. Organized by skill, not by tool.
A curated path, not written by a journalist: it is built on sources from people who are.
Total
39
Done
0
Progress
0%
Left
39
Est. time
~20 h

Fundamentals

fundamentals 0/5 ▾

How an LLM writes: predicting the next word

Tokens, training data and probability. Why fluent text is not the same as true text.

Hallucinations: invented quotes, figures and sources

The model fills gaps with plausible text. Real cases of made-up quotes and references, and what they teach.

What AI is and isn't: a course without code

Machine learning, neural networks and their limits, explained without math.

Chat, search-grounded or local: picking the tool

Knowledge cutoffs, answers with sources, and what happens to whatever you paste in.

Prompting for reporting tasks

Give context, a role and a format; ask for sources and for what it is unsure about. Iterate instead of taking the first answer.

Verification

verification 0/7 ▾

Verification basics: source, date, place

Lateral reading and tracing the original. The method doesn't change because the content is synthetic.

Reverse image search and the InVID-WeVerify plugin

Keyframes, metadata, magnifier and reverse search across engines, from one browser extension.

Spotting AI-generated images and video

Light, physics, text and context checks. Old tells like hands no longer work: think in probabilities, not proof.

Cloned voices and synthetic audio

A five-second clip is nearly impossible to judge by ear. Trace where it came from first, then look at the spectrogram.

AI detectors, and why they are not enough

False positives on human text, misses on edited content. A detector score is a clue, never the evidence.

Checking the AI's own output

Every name, figure, quote and citation goes back to a primary source before it is published.

Content provenance: C2PA and Content Credentials

Signed metadata that travels with the file. What it proves, and what its absence does not.

Audio & sources

sources 0/5 ▾

Transcribing interviews with AI

Speed vs accuracy, speaker labels, names and accents. Always check quotes against the audio.

Local transcription to protect sources

Whisper on your own machine: the recording never leaves it.

What not to paste into a cloud AI

Retention, training on your data and legal requests. Treat a chat window like an email to a third party.

Translation and reporting across languages

Machine translation for reading, a person for publishing. Idioms, names and legal terms are where it breaks.

Interview and background prep

Briefs, timelines and question lists as a starting point you verify, never as a source.

Documents & leaks

documents 0/4 ▾

Summarizing long documents without losing facts

Ask for page references, compare against the original and look at what the summary leaves out.

PDFs, OCR and tables

Getting text and tables out of scanned files before any model touches them.

Large document sets and leaks

Search, entity extraction and clustering over thousands of files, the way ICIJ works.

Entities and networks: people, companies, money

Extracting names and linking them across documents. Every connection still needs a human check.

Data

data 0/5 ▾

AI as a spreadsheet and SQL assistant

Formulas, cleaning and queries explained step by step. Check the result on a sample you can do by hand.

Open data portals

Finding official datasets and reading their documentation before analyzing anything.

Scraping with AI-written scripts

Let the model write the scraper; you check what it collects and respect the site.

Charts and maps

Picking the right chart, not decorating it. AI can draft; the editor decides what the data says.

Newsroom

newsroom 0/5 ▾

Headlines, summaries and SEO drafts

Options to choose from, never the final copy. Watch for details that were not in the story.

Automated stories from structured data

Elections, sports, markets: templates fed by data. Older than LLMs, and more reliable.

Newsletters, archives and repackaging

Turning published work into new formats, with the original as the only source.

What AI does to news audiences

Search summaries, fewer clicks and trust. Read the audience data before planning around it.

How newsrooms actually use it

Case studies: what worked, what was dropped and why.

Ethics & policy

ethics 0/4 ▾

Newsroom AI policies

What to allow, what to ban and who signs off. Read real policies before writing your own.

Telling readers when AI was used

Disclosure that informs the reader instead of covering the newsroom.

Copyright, training and licensing

Publishers vs AI companies: lawsuits, licensing deals and what they mean for your archive.

Bias and representation

Who the model leaves out: languages, regions, gender. Test it on your own beat.

Advanced: your own tools

advanced 0/4 ▾

Running models locally

Open models on a laptop with Ollama: privacy, cost and quality trade-offs.

Calling a model from a script

Python and an API: the same task over hundreds of documents, reproducibly.

Search over your own archive (RAG)

Answers grounded in your documents, with citations you can check.

Measuring the error rate before trusting a pipeline

A hand-checked sample tells you how often it is wrong. Publish nothing faster than you can check it.