The Basics
Artificial intelligence (AI)
Software that performs tasks that typically require human thinking — like understanding language, recognizing images, or making predictions.
In the newsroom: AI is already in your spell-checker, transcription tools, and content recommendation systems.
Generative AI
AI that creates new content — text, audio, images, or video — rather than just analyzing existing content.
Examples: ChatGPT, DALL-E, ElevenLabs. This is what most people mean when they say “AI” right now.
Large language model (LLM)
The technology behind tools like ChatGPT and Claude. Trained on massive amounts of text, an LLM predicts what words should come next (rather than “knowing” the answer) to generate coherent responses.
Think of it as: A very fast, very well-read autocomplete that can hold a conversation.
Training data
The text, images, or other content an AI model learned from. The model’s information, biases, and blind spots largely come from here.
Why it matters: If training data was biased or had a cutoff date, its outputs will reflect that.
Working with Tools
Prompt
The instruction or question you give to an AI tool. Better prompts get better results.
Tip: Be specific. “Summarize this 800-word story in 3 sentences for a social post” beats “make this shorter.”
Hallucination
The AP Stylebook defines this as when AI “produces falsehoods or inaccurate or illogical information.” This could include made-up quotes, sources, or statistics.
Rule of thumb: Never publish AI-generated facts without independent verification. Treat AI output like an eager intern: useful, but needs verification.
Context window
How much text an AI model or tool can “see” and work with at once — like its short-term memory. Larger windows handle longer documents.
Practical note: If you paste a long transcript into a tool and it seems to forget the beginning, you may have hit its limit.
Prompt engineering
The practice of crafting prompts strategically to get more useful, accurate, or consistent AI outputs.
Newsroom use: Develop a reusable prompt template for interview transcription summaries or story briefs.
Knowledge cutoff
The date after which an AI model has no information. Ask about events after its cutoff and it may hallucinate or simply say it doesn’t know.
Example: A model trained through early 2024 won’t know about elections, legislation, or leadership changes after that date.
Retrieval-Augmented Generation (RAG)
A technique that connects an AI tool to a specific database or document library, so it pulls real, current information before generating a response.
Newsroom use: An AI tool built on your station’s archive can answer questions based on actual past coverage.
Ethics, Policy & the Bigger Picture
AI transcription
Automated speech-to-text using AI. Tools like Descript, Otter.ai, and Whisper turn audio into editable text quickly.
Heads up: Accuracy varies with accents, crosstalk, and audio quality. Always review before publishing or quoting.
AI-assisted editing
Using AI tools to speed up audio or video editing — removing filler words, noise reduction, auto-leveling, or generating show notes.
Tools to know: Adobe Podcast, Descript, Cleanfeed, and Auphonic are common in public media workflows.
Deepfake
AI-generated audio or video that convincingly depicts someone saying or doing something they never did.
For journalists: Treat any audio or video of a public figure saying something surprising as unverified until confirmed.
AI bias
When an AI system produces skewed results because its training data reflected historical inequities or underrepresented certain groups.
Newsroom relevance: An AI story-suggestion tool trained mostly on legacy media may undervalue certain communities or story types.
Intellectual property (IP) & copyright
Legal rights to creative work — an open question with AI, since most models were trained on copyrighted material without explicit permission.
Watch this space: Litigation and legislation around AI and copyright is actively evolving.
Transparency / disclosure
Telling your audience when and how AI was used in reporting or production. SPJ’s Code of Ethics holds that “ethical journalism means taking responsibility for one’s work and explaining one’s decisions to the public.”
Industry norm emerging: Many newsrooms now append disclosures like “AI tools were used to transcribe interviews in this story.”
Human-in-the-loop
A workflow design where a human reviews, approves, or corrects AI outputs before they’re used or published.
Best practice: For journalism, “human in the loop” isn’t optional. AI assists; journalists decide.
Synthetic voice / voice cloning
AI that generates realistic speech from text — either a generic voice or a clone of a real person’s voice.
Ethical line: Using synthetic voice for efficiency is different from cloning a journalist’s voice without consent.
Agentic AI (agents)
AI that can take sequences of actions on its own — browsing the web, running searches, filing forms — rather than just answering questions and prompts.
On the horizon: Newsrooms are experimenting with agents that monitor sources or draft alerts automatically.
AI use policy
A newsroom’s written guidelines for when AI tools can and cannot be used, what approval is required, and what must be disclosed.
Ask your newsroom: A 2024 AP study found nearly 3 in 4 journalists had tried generative AI on the job — yet only about 20% of local newsrooms have a public policy. If yours doesn’t, today’s a good day to start the conversation.
Remember: AI tools are only as good as the humans using them. Your editorial judgment, your sourcing instincts, and your knowledge of your community are things no model can replicate.
Sources: AP Stylebook, Generative AI entry (stylebook.ap.org) · Poynter AI Ethics Starter Kit (poynter.org/ai-ethics-journalism) · SPJ Code of Ethics (spj.org/ethics) · Associated Press / American Journalism Project, AI in Newsrooms Study, 2024 · MIT Sloan Teaching & Learning Technologies, Glossary of Generative AI Basics (mitsloanedtech.mit.edu) · IBM Research, What is Retrieval-Augmented Generation? (research.ibm.com) · IBM Think, What is Agentic AI? (ibm.com/think)
Near + Now: New York City
Wednesday, September 9, 2026 · Google, 85 Tenth Avenue, Chelsea, NYC
Register Now — $150Questions? conference@pmja.org
PMJA Near + Now sponsors
Sponsorship makes Near + Now possible — and helps us keep registration costs low and scholarships available at every event. We’re grateful for partners who believe in what this kind of accessible professional development can do.
-
-
Corporation for Public Broadcasting
401 9th St NW
Washington, DC 20004Series funding partner -
