For Journalistss and Reporters
Protect Your Judgement: A Guide for Journalistss Using AI Tools
AI tools like Perplexity and Claude can summarise 50 sources in minutes, but a summary is not reporting. You risk building stories on AI-digested research instead of the primary sources themselves, which erodes the one skill no tool can replace: knowing what matters and why. The bigger threat is losing the instinct that comes from actually talking to people, listening for what they do not say, and recognising when a story is more than its facts.
These are suggestions. Your situation will differ. Use what is useful.
Do Your Own Source Work Before Using AI for Efficiency
Start with primary sources and direct reporting before you ever open Perplexity or Claude. Spend real time with the documents, the people, the data yourself. Only after you have formed your own judgement about what the story is should you use AI to help verify patterns or fill gaps. When you reverse this order, AI becomes a proxy for your reporting instinct instead of a tool that serves it.
- ›Conduct your core interviews and read your key documents before asking an AI to summarise related research
- ›Use Claude to spot-check your findings against what it has seen in training data, not to find your findings for you
- ›When using Otter.ai for transcription, listen to at least one full interview yourself before reading the transcript
Recognise What AI Summaries Remove From Your Understanding
A Perplexity summary of 30 sources tells you what those sources say. It does not tell you which source is contradicting the others, which reporter has a particular angle, or which experts have changed their position since their last statement. These gaps matter because they are where real stories live. When you read the actual sources, you build judgement about credibility and significance that no AI can transfer to you.
- ›After reading an AI summary, always spot-check at least three of the sources it cited to see what context it omitted
- ›Note which organisations or voices appear most often in AI summaries, then ask yourself whether that reflects importance or just volume of coverage
- ›When a source is quoted by an AI tool, read the full original to understand its actual position, not the extracted line
Use First-Draft Help Without Letting It Replace Your Editorsial Eye
ChatGPT and Claude can write a serviceable first draft from your reporting notes faster than you can type. This is genuinely useful. What is not useful is publishing that draft without major reconstruction. AI will make your story accurate but generic. It will hit all the facts but miss the colour, the tension, the specific human detail that makes readers care. Your job is to read every sentence and ask: did I actually report this? Does this sound like reporting or like a textbook?
- ›Feed AI only your own notes and reporting, never published stories or competitor coverage
- ›Rewrite every quote the AI generates from your notes, because the original words from your interview will be sharper
- ›Read the AI draft aloud and delete any sentence that does not sound like something a human learned through reporting
Protect Source Relationships by Staying in the Conversation
Your sources trust you because you have spent time understanding their world and calling them back with follow-up questions. If you use AI to summarise their previous statements instead of reading them yourself, and you ask AI what to ask them next, you are outsourcing the relationship. Sources can sense when a journalist has actually done the work versus when they are reading from a sheet. Younger journalists especially need to build this skill now, or they will never develop it.
- ›Before each source call, read their previous statements yourself, not an AI summary of them
- ›Write your own questions based on your reporting, then use Claude to help you spot gaps or follow-ups you missed
- ›Keep a running note of what each source has told you over time so you can recognise when their story has shifted
Make News Judgement Your Non-Negotiable Human Work
Deciding what matters, why now, and for whom is the core of journalism. This is where you fail if you hand your thinking to an engagement algorithm or an AI that optimises for clicks. An AI tool will not tell you that a small policy change affects real people more than celebrity gossip. It will not recognise when a story is important because it fills a gap in public understanding rather than because it is trending. This judgement is what separates a journalist from a content machine.
- ›Before you pitch a story, write one sentence answering why this matters to someone who does not already know they care
- ›Use engagement metrics to test your news judgement, not to replace it
- ›When an AI tool suggests a different angle or emphasis, ask yourself whether it is following what readers already know or leading them toward something they need to know
Key principles
- 1.Read your primary sources yourself before using AI to help you understand them better.
- 2.Let AI write fast, then rewrite fast, because your editorial instinct is what makes the story true even when the facts are correct.
- 3.Build source relationships through actual conversation, not by asking an AI what you already know about them.
- 4.Decide what matters based on reporting and human impact, not on what an algorithm predicts will get clicks.
- 5.Use AI as a research assistant that works for your judgement, never as a substitute for it.
Key reminders
- When using Perplexity for research, always ask it to list its sources, then read at least the headline and first paragraph of each one yourself before trusting the summary
- Otter.ai transcription is reliable for accuracy, but listen to the audio yourself at least once to catch tone, hesitation, and emphasis that the transcript removes
- Feed ChatGPT your raw reporting notes with timestamps from interviews, then edit its first draft by comparing every sentence back to what you actually heard and learned
- Create a simple rubric for source credibility that you write yourself, then reference it when an AI summary treats all sources as equally valid
- Set a rule: no story ships without you having read the three most important primary sources and conducted at least one follow-up interview based on your own questions