TV 2

AI Model Comparison

Six models, one newsroom — which tool fits which task?

Work smarter, not harder: AI for journalists
12 April 2026 6 models COMPARISON

Not all AI is the same. Mathias knows PimEyes and finds it hard to trust ChatGPT. Lars tried agents and finds Google faster. Bo has been doing deepfakes since 2018. This overview compares six AI models on what matters for you as journalists — and helps you choose the right tool for the right task.

What is "temperature"?

Every AI model has a temperature setting that determines how predictable or creative the output is. Think of it as the dial between a news bulletin and a literary essay.

0.0 — News bulletin
Always the same
1.0 — Column
Variation allowed
2.0 — Creative writing
Unpredictable

This is why the same prompt sometimes gives different results: the temperature introduces randomness. At temperature 0 you always get the same answer. At temperature 1+ you get variation.

Comparison Table

Criterion Claude Gemini NotebookLM Perplexity ChatGPT Copilot
Source citation●●●●○●●●○○●●●●●●●●●●●○○○○●●●○○
Information recency●●○○○●●●●●●●○○○●●●●●●●○○○●●●●○
Document analysis●●●●●●●●●○●●●●●●●●○○●●●●○●●○○○
Danish language●●●○○●●●●○●●●○○●●●○○●●●●○●●●○○
Reliability●●●●●●●●●○●●●●○●●●○○●●●●○●●●○○
Free usability●●●○○●●●●○●●●●●●●●○○●●●○○●●●●○
Privacy●●●●●●●●○○●●●○○●●●○○●●●○○●●●○○
Work tool integration●●●○○●●●●●●●●●○●●○○○●●●●○●●●●●

Per model

Claude

Anthropic · Opus 4.6 / Sonnet 4.5

"Most reliable for journalism work"

Strengths:
  • + Best document analysis, follows complex instructions
  • + Honestly says when it doesn't know
  • + Strongest privacy stance among major models
Weaknesses:
  • - No built-in web search
  • - Free version has limited daily use

Lars is considering switching to Claude — with the Prewash Method he'd get more out of his source research.

Gemini

Google · Gemini 2.5 Pro

"Strongest search integration"

Strengths:
  • + Direct Google integration and Deep Research
  • + Very current information
  • + Large context window for long documents
Weaknesses:
  • - Sometimes shares sources only after pushing
  • - Privacy: Google ecosystem

Mads Buur Bach already uses Gemini for Q&A — the Deep Research function makes it even more powerful for verification.

NotebookLM

Google · Free

"Best document processor"

Strengths:
  • + Upload documents and ask targeted questions
  • + Generates audio summaries (podcast-style)
  • + Completely free, no limits
Weaknesses:
  • - Only works with uploaded sources
  • - Danish sometimes unnatural in audio

Perfect for Mads Oxlund Petersen and Mathias, who want to extract information from large documents — upload and ask questions.

Perplexity

Perplexity AI · Pro needed for reliability

"Best source citation in search results"

Strengths:
  • + Real-time web search with direct source citation
  • + Every claim is clickable to source
  • + Pro: access to multiple AI models
Weaknesses:
  • - Free version makes many errors (Mathias's experience with double-checking)
  • - Document analysis weaker than Claude/NotebookLM

Peter Møller can use Perplexity Pro to research journalistic and scientific sources — it gives direct links.

ChatGPT

OpenAI · GPT-4.5 / o3

"Most versatile, most familiar"

Strengths:
  • + Broadest functionality (text, image, code, file analysis)
  • + Most familiar interface — low barrier to entry
  • + Good plugin ecosystem
Weaknesses:
  • - Source citation less reliable than Perplexity
  • - Tends toward "helpfulness" that can contain framing

Most of you already know ChatGPT — David uses it daily. With the Prewash Method it becomes a research assistant instead of a search engine.

Copilot

Microsoft · Microsoft 365 integration

"Handy within Microsoft, limited outside"

Strengths:
  • + Integrated in Word, Outlook, Teams, Excel
  • + Handy for daily office work
  • + Bing search integration
Weaknesses:
  • - Limited outside Microsoft 365
  • - Less suited for depth analysis

Lasse tried setting up a CoPilot Agent for Power BI — the integration potential is there, but requires structured prompting.

Why Claude ranks #1

For journalism work — where reliability, nuance and instruction-following weigh more than speed — Claude is the strongest model. It follows complex instructions precisely, honestly indicates when it doesn't know something (instead of making things up), and delivers the best document analysis.

Honesty clause: this ranking is subjective and context-dependent. For real-time search Perplexity is better. For Google workflows Gemini is better. For documents NotebookLM is sometimes handier. The "best" tool depends on the task.

And yes — this report was itself made with Claude. That is relevant to know.

Which tool for which task?

TaskRecommendedAlternativeWhy
Finding sources for a storyPerplexity ProGeminiDirect source citation per claim
Searching a long reportNotebookLMClaudeUpload and ask targeted questions
Transcribing an interviewSonixChatGPT (Whisper)Clickable transcript with timecodes
Fact-checking after an articlePerplexityGeminiReal-time web search with sources
Verifying images/videoGeminiInVID@synthid + image recognition
Researching Danish legislationPerplexity ProGeminiCurrent sources with links to legislation (use Claude for analysing uploaded legislation)
OSINT people researchPerplexity + PimEyesGeminiCombined web + facial search

Final Ranking

1
Claude — Most reliable for journalism work
2
Perplexity Pro — Best source citation in online search
3
Gemini Pro — Strongest search integration and Deep Research
4
NotebookLM — Best free document processor
5
ChatGPT — Most versatile, most familiar
6
Copilot — Handy within Microsoft, limited outside

Conclusion

No single tool is perfect. The art is choosing the right tool for the right task. Start with the model closest to your current work, learn it well, and then expand.

For this group: Bo and Pelle can deepen immediately with Claude and Gemini. Mads Oxlund Petersen and Mathias benefit most from NotebookLM for their documents. Peter and Franziska are served by Perplexity Pro for their research questions. Jakob can combine Perplexity with OSINT techniques for people finding. David and most of the group can improve their existing ChatGPT use with the Prewash Method. Sanne, who tried Python monitoring, will find Gemini Daily Alerts a working alternative. And Lars, who finds Google faster, should try Claude's document analysis to see the difference.