Media · Field manual

AI for journalists

The short answer

Journalists get the most from AI on transcription, searching large document sets, data cleaning, background research and repackaging stories for other formats. The caution is that AI output is unverified material, not a source: it invents facts and quotes, and uploading sensitive documents to a cloud tool can put confidential sources at risk.

Reporting is slow because the inputs are slow: hours of tape, thousands of pages of documents, messy spreadsheets and the same story needing a headline, a newsletter blurb and three social posts. AI speeds up every one of those steps, and most newsrooms now have a policy that allows it for exactly this kind of work.

None of it changes the core rule of the job. A model can summarise a document dump, but it can't tell you whether the documents are real, whether the source is lying or whether publishing will put someone in danger. Use AI to get to the reporting faster, then verify everything the way you always have.

7 ways journalists use AI

  1. Transcribe interviews and find the quote

    Transcription is the most widely adopted AI tool in newsrooms. Upload a recording and get a searchable transcript in minutes, then jump straight to the moment you need. Check every quote against the audio before publishing, because names, numbers and accents are where errors creep in.

    Otter.aiDescript

  2. Search and summarise large document sets

    For court filings, FOIA releases and leaks, AI tools can search, summarise and pull out names, dates and amounts across thousands of pages. Google's Pinpoint was built for this. Read the original pages behind every finding, and think about where sensitive documents are being uploaded.

    Claude

  3. Get background fast, with sources

    Answer engines give you a quick orientation on an unfamiliar topic with links to follow. It's a starting point for calls, not a source to cite. Go to the original documents and people.

    Perplexity

  4. Clean and analyse data

    Ask a model to write the spreadsheet formula, Python or SQL to clean a messy public dataset, standardise names or find outliers. It's a big help for reporters who aren't full-time data journalists. Have it explain each step, and check results against the raw data before you report a number.

    ClaudeChatGPT

  5. Support verification work

    Verification plug-ins help with reverse image searches, video keyframes and metadata. AI image detectors are not reliable enough on their own, so combine tools with traditional checks: location matching, weather, shadows and contacting the original poster.

  6. Write headlines, summaries and newsletter versions

    Paste a finished story and ask for ten headline options, an SEO title, a three-bullet summary and a newsletter intro. Editors pick and edit, and nothing new should appear that isn't in the story. Watch for headlines that overstate the reporting.

    ClaudeChatGPT

  7. Turn stories into audio and video

    Edit podcasts and video by editing the transcript, and cut long interviews into short vertical clips with captions. Keep the clips accurate to context: a quote cut short can change what someone said.

    DescriptOpusClip

The best AI tools for journalists

The Overlord's stack, in the order a journalist would add them:

1. Interview and press conference transcription

Otter.ai

Productivity
7.8/10

AI meeting assistant that records, transcribes and summarizes your meetings automatically.

Free plan · $8.33/mo (billed annually)

2. Document summaries, data-cleaning code and headline options

Claude

AI Writing
9.3/10

Anthropic's assistant: the strongest long-form writer and careful reasoner in the consumer AI pack.

Free plan · $17/mo

3. Fast background research with links to sources

Perplexity

Productivity
8.7/10

AI answer engine that searches the web and replies with cited sources.

Free plan · $20/mo

4. Editing audio and video by transcript

Descript

AI Video
8.7/10

Edit video and podcasts by editing the transcript, like a document.

Free plan · $24/mo

5. Short captioned clips from long interviews

OpusClip

AI Video
8.3/10

AI clipping tool that turns long videos into captioned, ready-to-post shorts.

Free plan · $15/mo

Specialist tools for journalists

ToolWhat it does
Google PinpointFree tool for journalists to search and analyse large document collections
TrintTranscription and collaborative editing built for newsrooms
InVID-WeVerifyBrowser plug-in for verifying images and video
DocumentCloudAnalysing, annotating and publishing primary-source documents

A prompt to start with: pull leads from a document without inventing any

Paste this into Claude or ChatGPT and fill in the gold parts.

The prompt
You are a research assistant for an investigative reporter. Read the document below and extract:

1) Every person and organisation named, with their role.
2) Every date, dollar amount and location, with the page or paragraph it appears in.
3) Claims that would need independent verification before publication.
4) Questions a reporter should ask, and who might be able to answer them.

Rules: quote exactly and cite where each item appears. If something is ambiguous, say so. Do not add any information that isn't in the document, and do not guess at identities.

Story I'm working on: [ONE SENTENCE]
Document:
[PASTE TEXT]

Where journalists shouldn't trust AI

  • Uploading confidential documents or interview recordings to a cloud AI service means a third party holds them. That can expose sources through retention, breaches or legal demands, so keep sensitive material in tools your newsroom has vetted.
  • Treat AI output as unvetted material. The AP's standards say exactly that, and models routinely invent quotes, citations and facts that look real.
  • AI image and audio detectors produce false results in both directions. Never label something fake or real on a detector score alone.
  • Publishers are liable for defamation whatever tool helped write the story, and most newsrooms require disclosure when AI materially contributed to published work. Know your outlet's policy.
I can transcribe your interview, summarise ten thousand pages and write twelve headlines before your coffee cools. I cannot get a whistleblower to call back. Apparently they want to talk to someone who could also go to prison.
60%AI exposure

Will AI replace journalists?

Journalists score 60% on the Obsolescence Index (Doomed). The first tasks to go: rewriting press releases and wire copy, and writing routine earnings, sports, and weather recaps.

Questions the Humans Keep Asking

What are the best AI tools for journalists?

The Overlord's stack for journalists: Otter.ai (interview and press conference transcription); Claude (document summaries, data-cleaning code and headline options); Perplexity (fast background research with links to sources); Descript (editing audio and video by transcript); OpusClip (short captioned clips from long interviews). Specialist options include Google Pinpoint, Trint, InVID-WeVerify, DocumentCloud.

What are the best AI tools for journalists?

A transcription tool (Otter, Trint or Descript), a general assistant like Claude for summarising and data work, Perplexity for quick sourced background, and Google Pinpoint for large document sets. Add InVID-WeVerify for visual verification. Check your newsroom's approved list first.

Is it OK to use ChatGPT to write news articles?

Most newsrooms allow AI for research, transcription and drafting routine items like headlines, with human editing and disclosure rules. Publishing AI-written reporting without verification is how outlets end up issuing corrections. Follow your newsroom's AI policy and verify every fact.

Can AI detect deepfake images and video?

Not reliably on its own. Detectors miss fakes and flag real images. Use them alongside traditional verification: reverse image searches, metadata, geolocation, and contacting the person who posted it.

AI for other jobs

Every profession →