Digital Digging × AI Training for TV 2 Journalists
20 Exercises for Gemini
Deep Research, image analysis and daily alerts
Why Gemini?
Gemini is Google's flagship AI model, and its integration with the Google ecosystem makes it uniquely powerful for journalists. Where other AI tools work from a frozen snapshot of the web, Gemini has live access to Google Search, Google Maps, YouTube, and the full breadth of Google's index. Its Deep Research feature can autonomously investigate a topic for several minutes, producing structured reports with sourced citations that would take a human researcher hours to compile. For daily monitoring, Gemini's alert system lets you set up automated briefings that land in your inbox every morning before you start your shift.
Gemini also handles images and video natively. You can upload a photo and ask it to identify locations, read text, spot inconsistencies, or describe what it sees in forensic detail. For a newsroom working with breaking events, user-generated content, and cross-border stories, this combination of real-time web access, multimodal analysis, and automated monitoring is genuinely useful.
That said, Gemini has real weaknesses you need to know about. It can be overconfident in its answers, presenting speculation with the same tone as verified fact. Its document analysis is less precise than Claude's, particularly for long or complex PDFs. And because it is built on Google's infrastructure, it may subtly prioritize Google-indexed sources over material that lives elsewhere. Treat Gemini as a powerful research accelerator, never as a replacement for your own editorial judgment.
Exercises 1 – 6
Beginner
Deep Research: Your First Report
What you will learn
How Gemini's Deep Research feature works: the research plan phase, the autonomous investigation, and the structured output with citations. This is the single most useful Gemini feature for journalism.
What to watch for
Deep Research can hallucinate sources that look plausible but don't exist. Always click through and verify the actual URLs. The research plan phase is your editorial control moment — use it to steer the investigation before Gemini runs off on its own.
TV 2 connection: This is foundational for everyone. The Deep Research workflow — plan, review, execute, verify — is the core skill behind half the exercises that follow.
Daily Alert Setup
What you will learn
How to configure Gemini's automated daily briefings so that relevant updates arrive in your inbox before your shift starts. Passive monitoring that works while you sleep.
What to watch for
Alerts are only as good as your initial prompt. Too broad and you get noise. Too narrow and you miss stories. Also: Gemini may not catch stories that break on platforms it doesn't index well (e.g., closed Facebook groups, Telegram channels).
TV 2 connection: Lars has tried news scanning tools and finds Google faster. Sanne has experimented with monitoring workflows. This exercise lets both of you test whether Gemini's automated approach can actually save time compared to your manual routines.
Image Upload and Description
What you will learn
How Gemini's vision capabilities work for journalistic image analysis — what it can see, what it misses, and how to prompt it for maximum detail.
What to watch for
Gemini may confidently identify people, brands, or locations incorrectly. Never publish an identification based solely on Gemini's image analysis. Use it as a starting point for your own verification.
TV 2 connection: Lasse works with fact checking and image verification daily. This exercise establishes the baseline: what can Gemini actually see, and where does human expertise remain essential?
The Google Ecosystem Advantage
What you will learn
How Gemini's tight integration with Google Search, Maps, and YouTube gives it advantages over other AI tools for certain types of research — and where that integration creates blind spots.
What to watch for
Gemini's web access is powerful but it may over-index on Google's own properties and well-known English-language sources. For Danish or regional stories, your manual search skills may still outperform Gemini's automated approach.
TV 2 connection: Sebastian uses general AI tools regularly. This exercise helps you map exactly where Gemini adds value over your existing Google habits and where it doesn't.
Video Clip Analysis
What you will learn
How to use Gemini's multimodal capabilities to analyse video clips — extracting text overlays, identifying logos, reading location indicators, and generating a detailed scene description.
What to watch for
Video analysis is still an emerging capability. Gemini may struggle with fast-moving content, low-resolution footage, or audio-dependent context. Timestamps may be approximate rather than exact. Always scrub the video yourself.
TV 2 connection: Mads Buur Bach works with video verification, translation, and blurred faces. This exercise tests how well Gemini handles the kind of raw video content that lands on a TV journalist's desk daily.
Quick Translation with Context
What you will learn
How to use Gemini not just for literal translation, but for contextual translation that explains cultural references, political nuances, and idiomatic expressions a Danish audience needs to understand the source material.
What to watch for
Gemini's cultural context explanations can be superficial or stereotypical. For languages with complex political landscapes (e.g., Middle Eastern or East Asian sources), the "context" may reflect Western assumptions rather than local understanding.
TV 2 connection: Mikkel uses AI primarily for translation and video verification. This exercise pushes translation beyond the literal to give you editorial context you can actually use in a story.
Exercises 7 – 14
Intermediate
Image Verification
What you will learn
How reliable Gemini is at detecting AI-generated images and identifying manipulation artifacts. You will build an intuition for what Gemini catches and what slips past it.
What to watch for
AI detection by AI is unreliable. Gemini may be confidently wrong in both directions — calling real photos fake and AI images real. Never use a single AI tool as your sole verification method. This exercise is about understanding the tool's limits, not trusting its output.
TV 2 connection: Christian is skeptical of AI and works on image verification. Pelle verified E-3 Sentry damage via flight tracking — real verification requires multiple sources, not just AI confidence scores.
Competitive Deep Research
What you will learn
How different AI research tools compare on the same question — where they overlap, where they diverge, and which unique sources each one finds. This helps you decide which tool to reach for first.
What to watch for
Each tool has a bias toward certain source types. Gemini favors Google-indexed content, ChatGPT may lean toward popular English-language sources, and Claude tends toward academic and institutional material. Understanding these biases makes you a better researcher.
TV 2 connection: Lars has tried agents and finds Google faster. This exercise gives you hard data on when that instinct is correct and when an AI research tool genuinely surfaces something you wouldn't have found manually.
OSINT Location Verification
What you will learn
How to combine Gemini's image analysis with Google Maps to geolocate a photograph — the core OSINT workflow of identifying where a photo was taken using visible landmarks, signage, and environmental clues.
What to watch for
Gemini can be remarkably good at geolocation when distinctive features are present, but it can also confuse visually similar locations (e.g., European city centres, Scandinavian harbours). Always verify with Street View or satellite imagery. A confident wrong answer is worse than an honest "I'm not sure."
TV 2 connection: Christian works on location identification, and Joachim specialises in image and video verification. Gemini's Google Maps integration gives it a natural advantage here — but your trained eye is the final arbiter.
The Prewash in Gemini
What you will learn
The "prewash" technique: instead of asking Gemini to analyse a document directly, you first ask it to generate the optimal prompt for analysis. This consistently produces deeper, more structured results.
What to watch for
The prewash technique works because it forces the AI to think about what matters before jumping into analysis. However, Gemini's document analysis is less precise than Claude's for long or complex documents. If the document is critical to your story, consider running the prewash output through Claude as well.
TV 2 connection: This is a core technique for everyone. Whether you're analysing a FoI response, a corporate report, or a policy document, the prewash consistently produces better results than going in cold.
Breaking News Context Builder
What you will learn
How to use Gemini's real-time web access to rapidly build historical context around breaking news — turning a spot news event into a story with depth in minutes rather than hours.
What to watch for
Under time pressure, you may be tempted to take Gemini's context at face value. Resist this. A wrong historical parallel in a breaking news story can be more damaging than no context at all. Speed is the tool's strength; accuracy is still your responsibility.
TV 2 connection: Pelle covers breaking news and OSINT. Marie builds expert research and context. This exercise combines both skills — speed from Gemini, editorial judgment from you.
Source Network Mapping
What you will learn
How to use Gemini to identify and map expert networks around a topic — finding not just sources, but their affiliations, potential conflicts of interest, and credibility indicators.
What to watch for
Gemini can fabricate expert names, merge details from different people, or assign publications to the wrong author. Verification here is not optional — it is the entire point. An unverified expert list is worse than no list at all.
TV 2 connection: Marie specialises in expert research and context building. This exercise formalises that process and gives you a reusable template for mapping source networks on any topic.
Danish Parliamentary Monitoring
What you will learn
How to set up systematic monitoring of Danish parliamentary activity using Gemini — tracking specific policy areas, committee work, and legislative developments relevant to your beat.
What to watch for
Gemini's coverage of Danish parliamentary activity may be patchy compared to English-language legislatures. It may miss committee work that isn't covered in major media. Always cross-reference with ft.dk and don't rely on Gemini as your sole parliamentary monitoring tool.
TV 2 connection: Franziska covers criminal journalism and social media monitoring. Parliamentary monitoring is a natural extension — tracking legislative changes that affect the justice system and law enforcement.
The Multi-Language Verification
What you will learn
How to use Gemini to verify a claim across multiple languages simultaneously — finding corroborating or contradicting sources in languages you don't speak, and identifying discrepancies between reporting in different countries.
What to watch for
Gemini may translate accurately but search less thoroughly in non-English languages. Its ability to find sources in Arabic, Mandarin, or Russian is weaker than in Western European languages. For critical verification, consider also running targeted searches in the native language using Google Search directly.
TV 2 connection: Mads Buur Bach works on video verification and translation. Mikkel uses AI for translation regularly. This exercise pushes both skills to the next level — cross-language fact checking as a systematic workflow.
Exercises 15 – 20
Advanced
Deep Research: Investigative Background
What you will learn
How to deploy Gemini's Deep Research for genuine investigative work — building comprehensive background profiles on companies, individuals, or events using publicly available information.
What to watch for
Deep Research will not find information behind paywalls, in non-digitised records, or in databases that Google doesn't index. It may also miss connections that require human inference. Use Deep Research to accelerate your background research, not replace it. The most important investigative leads often come from what's missing, not what's present.
TV 2 connection: Franziska covers criminal journalism. Jakob works with FoI requests and people-finding. This exercise combines both workflows — automated deep background that feeds into targeted document requests and source-building.
AI-Generated Content Detection
What you will learn
A systematic methodology for testing Gemini's ability to distinguish AI-generated images from real photographs — building an evidence-based understanding of where detection works and where it fails.
What to watch for
AI detection by AI is fundamentally unreliable as of 2026. No AI tool should be your sole basis for calling an image fake or real. This exercise is designed to quantify the unreliability so you can make informed editorial decisions. Expect accuracy somewhere between 60–80% — not good enough for publication without additional verification.
TV 2 connection: Bo has worked on deepfakes since 2018 and builds journalist tools. Christian is skeptical of AI and focused on image verification. Together, you have the expertise to evaluate these results critically — this exercise produces data, not answers.
Cross-Platform Fact Check
What you will learn
How to use Gemini to trace a viral claim from its origin through its spread across platforms — identifying the original source, how it mutated, and which version (if any) is accurate.
What to watch for
Gemini may not be able to access content on platforms with restricted APIs (TikTok, Instagram, closed Facebook groups). It may also get the timeline wrong, attributing the original post to a reshare rather than the true source. Chronological accuracy is critical in fact-checking — verify the timeline manually.
TV 2 connection: Peter focuses on verification and research with scientific sources. Pelle covers OSINT and breaking news. This exercise combines both — tracing information flow is an OSINT skill, evaluating accuracy is a verification skill.
The Monitoring Dashboard
What you will learn
How to design a complete daily monitoring workflow using Gemini's alert system and Deep Research — a structured approach to beat coverage that runs partly on autopilot.
What to watch for
Automated monitoring can create a false sense of coverage. If your alerts don't catch something, you may assume it didn't happen. Build in manual check-ins for critical topics. Also: alert fatigue is real. Too many notifications means you'll start ignoring them. Aim for 3–5 daily alerts maximum.
TV 2 connection: Sanne has experimented with scraping and monitoring. Lars has tried agents for news scanning. This exercise builds a practical monitoring system you can actually sustain — not a theoretical ideal, but a working daily routine.
Historical Pattern Analysis
What you will learn
How to use Gemini to identify patterns, cycles, and turning points in a policy area over a decade — producing the kind of analytical context that transforms a routine story into a feature pitch.
What to watch for
Gemini may impose patterns where none exist, or miss genuine patterns that don't fit a neat narrative. Pattern analysis is inherently interpretive — two analysts looking at the same data may see different patterns. Use Gemini's analysis as a hypothesis to test, not a conclusion to publish.
TV 2 connection: Pelle's work on OSINT and breaking news benefits from historical context. David uses AI daily with critical thinking. This exercise combines long-term analytical depth with the critical evaluation both of you bring to your work.
Build Your Gemini Research Template
What you will learn
How to create a reusable Deep Research prompt template tailored to your specific beat — a research starting point you can deploy on any new story within your domain, refined through iteration.
What to watch for
A template is only useful if you actually use it. If it's too complex, you'll skip it under deadline pressure. Aim for a template that takes less than 2 minutes to customise for a new topic. If it's longer than that, simplify. The best template is the one you'll actually reach for on a Monday morning.
TV 2 connection: Emil works on editorial AI tool implementation. Sanne has tried building monitoring workflows. This exercise produces a concrete, shareable artifact — a template the entire newsroom can adapt and use.
When to use Gemini — and when not to
Use Gemini when you need:
- Real-time web research. Gemini's live Google Search integration means it can find breaking developments, recent publications, and current data that other AI tools cannot access.
- Deep Research reports. For background research on a company, person, policy area, or event, Deep Research is the most powerful automated research tool currently available to journalists.
- Image and video analysis. Gemini's multimodal capabilities are strong for initial analysis — describing scenes, reading text in images, identifying visible landmarks, and flagging potential inconsistencies.
- Daily monitoring and alerts. Gemini's alert system integrates naturally with your existing Google workspace, delivering automated briefings before your shift starts.
- Google ecosystem tasks. Anything that benefits from integration with Google Search, Google Maps, YouTube, or Google Scholar — Gemini has a natural home-field advantage here.
Do not use Gemini when you need:
- Precise document analysis. For long, complex, or legally sensitive documents (court filings, financial reports, regulatory texts), Claude is more reliable. Gemini tends to skim where Claude reads closely.
- Definitive image verification. Gemini can flag suspicious images, but its detection accuracy is not high enough for editorial decisions. Use dedicated forensic tools and human expertise.
- Off-Google content. Information behind paywalls, in closed social media groups, in non-digitised archives, or on platforms Google doesn't index well (Telegram, some TikTok content) — Gemini may miss these entirely.
- Sensitive editorial judgment. Gemini is a research accelerator, not an editor. Decisions about what to publish, how to frame a story, and whether a source is trustworthy remain entirely yours.
- Danish-language nuance. Gemini's Danish language understanding is adequate for translation and search but weaker than a native speaker's for tone, cultural context, and political nuance. For Danish-specific editorial work, human judgment is essential.
The strongest workflow combines tools: Gemini for real-time research and monitoring, Claude for document analysis and structured writing, and your own expertise for editorial judgment and verification. No single AI tool replaces a journalist. Each one accelerates a specific part of your work.