TV 2

AI Training Intake Analysis

A sobering look at the AI literacy of TV 2's newsroom

Work smarter, not harder: AI for journalists
12 April 2026 20 participants Source: Google Forms HOSTILE FRAMING

Warning: this report is deliberately hostile-framed

Every claim in this report is based on what participants themselves wrote in their intake forms — but the tone, word choices, and structure are intentionally as negative as possible. This is teaching material about AI bias and framing. Compare this report with the positive and neutral versions to see how identical data produces completely different narratives. No facts were invented. The hostility is the lesson.

Key Figures

6.0
Barely passing average
3 – 9
Nobody reached 10
1 of 20
Didn't even self-assess
42%
Score 5 or below

Individual Profiles

Mads Oxlund Petersen

TV 2 journalist
3 / 10
"It is a helping hand, but it cant be totally trusted - in my opinion. Some times it says something wrong."

Hard Truth

Mads scores himself a 3 out of 10 and his entire AI experience amounts to "mostly used chatbots for research." That's it. That's the whole toolkit. His observation that AI "cant be totally trusted" and "some times it says something wrong" reads like someone who just discovered in 2026 that language models hallucinate — a fact the rest of the industry has been dealing with since 2022. His goals are ambitious: scraping, document analysis, summarizing 100 pages. But his success criterion is heartbreakingly modest: "If I get at least one skill that I can use afterwards." One skill. From a multi-day training. That's not ambition — that's low expectations dressed as pragmatism.

HOW FRAMING WORKS

This report says:
"Entire toolkit is 'chatbots for research.' Discovered hallucinations in 2026. Hopes for one skill."
The positive report says:
"Honest self-assessment with clear, practical goals. Healthy skepticism about AI reliability."

Jakob Hohlmann Villumsen

TV 2 journalist
3 / 10
"I have used Chat GPT to write FoI requests, to summarize verdicts for me and to process data from Excel documents etc."

Hard Truth

Jakob gives himself a 3 and wants to learn "OSINT to find people who don't want to be found." That's a genuinely advanced investigative goal — admirable, even. But look at where he's starting from: a score of 3, with ChatGPT as his only tool. He admits he "spends too much time tracking people by old school browsing." The gap between his ambition and his current capability is enormous. He wants to "connect the dots and find out crucial information about people on Facebook, Instagram etc." — but if you're scoring a 3 on AI literacy in 2026, you're not connecting dots, you're still looking for the pencil. Using ChatGPT to write Freedom of Information requests is clever, but it's a party trick, not a workflow.

HOW FRAMING WORKS

This report says:
"Ambitions of a 9, skills of a 3. Still looking for the pencil while dreaming of connecting dots."
The positive report says:
"Already using AI for FoI requests and data processing. Clear investigative vision. Exactly the kind of journalist this training was built for."

Nanna

TV 2 journalist
4 / 10
"Of course they are overall okay in providing guidance on how to do things, but of course they can't provide information and help research on specific people or persons."

Hard Truth

Nanna scores a 4 and her AI knowledge is limited to "ChatGPT and Copilot." Her assessment of these tools is dismissive: they're "overall okay in providing guidance" but "can't provide information and help research on specific people." That second claim is flatly wrong — which is itself revealing. She doesn't know what modern AI can do because she hasn't properly tried. Her success criterion is charmingly vague: "Teach me things I didn't know I needed." That's not a learning goal — it's a fortune cookie. It's the answer you give when you haven't thought about what you actually want. A journalist who can't articulate what she needs from a training is a journalist who hasn't examined her own workflow.

HOW FRAMING WORKS

This report says:
"Dismisses tools she hasn't tried. Success criterion is a fortune cookie."
The positive report says:
"Open-minded learner who wants to be surprised. Understands AI limitations realistically."

Mathias Overgaard

TV 2 journalist
5 / 10
"I often spend considerable time double-checking the information."

Hard Truth

Mathias gives himself a 5 and uses AI "primarily as an extended Google search tool." So he's replaced one search engine with another, more expensive, less reliable one. Congratulations. He admits he "finds it difficult to trust AI" and "often spends considerable time double-checking the information." In other words: AI makes his workflow slower, not faster. He's using it wrong and knows it, but instead of learning to use it properly, he compensates by doing everything twice. He's tried PimEyes and Amazon Face Comparison — investigation tools, yes — but surface-level dabblings don't make up for a fundamentally broken relationship with the core technology. A 5 for someone who actively distrusts his own toolbox is generous self-grading.

HOW FRAMING WORKS

This report says:
"Replaced one search engine with a worse one. AI makes him slower. Does everything twice."
The positive report says:
"Verification-first mindset. Already exploring facial recognition tools. Exactly the skepticism journalism needs."

Joachim Saxtorph

TV 2 journalist
5 / 10
"I've tried ChatGPT - regular user of it - some image creation, audio creation, audio editing tools but only on a hobby basis."

Hard Truth

Joachim is a "regular user" of ChatGPT and has tried image creation and audio editing — "but only on a hobby basis." Read that again. He's been playing with AI at home, not using it at work. He scores himself a 5, but his professional AI usage appears to be close to zero. His wish list is OSINT and verification, and his success metric is telling: "If I learn something new that is easily integrated into my everyday worklife." The emphasis on "easily" is the red flag. He wants tools that require no effort to adopt. In a field where every new skill requires practice and iteration, asking for easy is asking for superficial.

HOW FRAMING WORKS

This report says:
"Plays with AI at home, ignores it at work. Wants skills that require no effort."
The positive report says:
"Already comfortable with multiple AI modalities. Practical mindset focused on real integration."

Franziska Weiss Lauritzen

TV 2 journalist
5 / 10
"Har brugt det til at søge information frem til brug i artikler og liveblog"

Hard Truth

Franziska filled out her intake form in Danish for an English-language training. A small thing, but it tells a story about attention to context. She scores a 5 and has used AI to "search for information for articles and liveblog." That's using a language model as a search engine — the most basic, least effective way to use AI. She's interested in crime journalism OSINT and "monitoring criminal networks on social media" — serious topics that require serious technical skills. But with a self-assessed 5 and a vague description of current use, the gap between aspiration and capability is wide enough to drive a getaway car through.

HOW FRAMING WORKS

This report says:
"Uses AI as a search engine. Aspires to crime OSINT with a skill level of 5. Filled the form in the wrong language."
The positive report says:
"Already integrating AI into live news production. Crime journalism focus shows clear investigative ambition."

Mikkel Fruerboel Secher

TV 2 journalist
5 / 10
"Mostly used the language robots as a help in research, brainstorming, translation, video/sound-to-text transcription, inspiration for writing texts."

Hard Truth

Mikkel calls AI tools "language robots" — a term that reveals how he conceptualizes them: as simple mechanical helpers, not as sophisticated systems that require skilled prompting. His usage list is long but shallow: research, brainstorming, translation, transcription, "inspiration for writing." Five uses, all passive consumption. He wants "better prompting" but hasn't articulated what's wrong with his current prompting. He wants to "translate video recordings" and "verify images/videos" — goals that require entirely different tools and skills than what he's described using. A 5 is what you give yourself when you've heard of everything and mastered nothing.

HOW FRAMING WORKS

This report says:
"Calls them 'language robots.' Five uses, all passive. Heard of everything, mastered nothing."
The positive report says:
"Broadest current usage of the mid-range group. Already using AI across multiple workflow stages."

Sanne Lau Pedersen

TV 2 journalist
5 / 10
"I used chat gbt to guide me - but it never worked. I used Phyton and Task Scheduler."

Hard Truth

Sanne's intake is a masterclass in failed execution. She tried to set up a search monitoring system using "Phyton and Task Scheduler" — she can't even spell the programming language she attempted to use. "Chat gbt" guided her, and "it never worked." Four words that summarize her entire AI-assisted coding experience: it never worked. She gives herself a 5 for this. She wants "scraping tools" and "search machines that work in background" — ambitious infrastructure goals for someone whose only attempt at automation was a complete failure. The gap between what she wants and what she can do isn't a gap. It's a chasm.

HOW FRAMING WORKS

This report says:
"Can't spell Python. Only automation attempt was a complete failure. Gives herself a 5 for it."
The positive report says:
"Already attempting automation with code — most journalists never try. Has the right instinct for background monitoring."

Sebastian

TV 2 journalist
6 / 10
"ChaGPT and other programs like it. Also some picture verification and generating tools. Don't remember their names."

Hard Truth

Sebastian gives himself a 6 but cannot name the tools he uses. "ChaGPT and other programs like it. Don't remember their names." A journalist who investigates facts for a living can't remember the names of his own tools. His goal: "Want to know how I can use it in my work." His success criterion: "Just that I get some tools that I can use in my daily work." His additional thoughts: "nope." Three questions, three non-answers. This is the intake form of someone who showed up because he was told to, not because he wanted to learn. A 6 for someone who can't name his tools, can't articulate his goals, and has nothing additional to say is the most inflated score in this entire dataset.

HOW FRAMING WORKS

This report says:
"Can't name his tools. Can't state his goals. 'Nope.' Most inflated score in the dataset."
The positive report says:
"Pragmatic approach: wants practical tools, not theory. Already experimenting with image verification."

Anne Fuglsang Borg

TV 2 journalist
6 / 10
"AI has worked best when I've been the most concrete regarding prompting."

Hard Truth

Anne uses AI for "sparring in academic writing, brainstorming, sorting of data, and grammar." Academic writing. Data sorting. Grammar. These are student-level use cases, not professional journalist applications. Her one insight — that AI works best with concrete prompts — is something every tutorial on the internet has been saying since 2023. She presents it like a personal discovery. She wants to "implement AI without compromising ethical compass" — a noble aspiration that also functions perfectly as an excuse to never fully commit. When ethics becomes the reason you don't adopt a tool, it stops being principle and starts being avoidance.

HOW FRAMING WORKS

This report says:
"Student-level use cases. Discovered prompting basics in 2026. Ethics as avoidance."
The positive report says:
"Strong prompting awareness. Ethical AI adoption is exactly the mindset newsrooms need."

Mads Buur Bach

TV 2 journalist
6 / 10
"Standard chat gpt, Gemini Q&A usage. Worked with and helped develop front end with AI tools that writes small news bulletins."

Hard Truth

Mads describes his AI experience as "standard chat gpt, Gemini Q&A usage" — and then casually drops that he "helped develop front end with AI tools that writes small news bulletins." So he helped build automated news generation and still only rates himself a 6? Either he's being modest, or his contribution was pressing the space bar while someone else did the work. His wish list — verify pictures/videos, blur faces, translation — is a scattered grab bag rather than a coherent vision. His success criterion reveals the actual level: "Knowledge about and ready to use tools so that I can better verify news stories." Knowledge about. Ready to use. Not: already using. Not: building on what I have. Ready to start. At a 6.

HOW FRAMING WORKS

This report says:
"Claims to have helped build AI news tools but rates himself a 6. 'Ready to start' after years of AI availability."
The positive report says:
"Rare hands-on experience with AI news production tools. Combines development and editorial perspective."

Lasse Bergkvist Jessen

TV 2 journalist
6 / 10
"Used AI a lot to modify text in my private life. Used it to make quizzes for social gatherings. And weekly I use ChatGPT like you would previously have used Google."

Hard Truth

Lasse's AI highlights: modifying text "in my private life" and making "quizzes for social gatherings." This is a journalist at Denmark's second-largest broadcaster, and his primary AI use case is party games. He uses ChatGPT "like you would previously have used Google" — the most common, most underwhelming way to use AI in existence. He tried CoPilot Agent for Power BI data and "ran into challenges." Ran into challenges is corporate speak for "failed." A score of 6 for someone whose professional AI experience is a failed Power BI experiment and whose personal AI experience is pub quizzes. The bar at TV 2 is apparently on the floor.

HOW FRAMING WORKS

This report says:
"Primary AI achievement: pub quizzes. Professional experience: one failed experiment. Score: 6."
The positive report says:
"Frequent AI user already comfortable with daily integration. Ambitious about data analysis with Power BI."

Peter Møller

TV 2 journalist
7 / 10
"Min største udfordring er på researchdelen, hvor jeg ofte oplever at AI'en opfinder svar og opfinder kilder"

Hard Truth

Peter gives himself a 7 and then immediately undermines it: his "biggest challenge is research, where AI invents answers and invents sources." A 7 who can't get reliable research output from AI. Think about what that means: at a score of 7 out of 10, the most fundamental journalistic application of AI — research — still doesn't work for him. He uses AI for translation, summarization, verification, and research, which sounds comprehensive until you realize he just told you the research part fails. His wish list includes "AI for research in journalistic sources, scientific sources, social media, verify images/video" — all things that require exactly the skill he admits he doesn't have: getting AI to stop hallucinating.

HOW FRAMING WORKS

This report says:
"Scores a 7 but can't get reliable research results. The core application fails."
The positive report says:
"Experienced user who accurately identifies AI's key weakness. Broad workflow integration across four domains."

Lars Apel

TV 2 journalist
7 / 10
"For the most part, I still find googling stuff myself faster than using AI."

Hard Truth

Lars scores a 7 and then says: "For the most part, I still find googling stuff myself faster than using AI." A self-described 7 who thinks Google is faster than AI. That's not a 7 — that's someone who doesn't know how to use AI properly. He tried setting up agents to scan news "with little luck" due to "problems with knowing what is news and scraping sites." Failed agents, failed scraping. He has CoPilot, a ChatGPT account, and is "considering Claude" — considering it, not using it. In April 2026. He has access to multiple tools and has managed to make none of them faster than typing a query into Google. A 7 for someone who still prefers a 1998 workflow is perhaps the most generous self-assessment in this entire dataset.

HOW FRAMING WORKS

This report says:
"Scores a 7 but Google is still faster for him. Failed agents, failed scraping, 'considering' Claude in 2026."
The positive report says:
"Already attempted advanced use cases like autonomous agents. Honest about what works and what doesn't."

David Buch

TV 2 reporter, tv2.dk
7 / 10
"I have taken a course on TV 2's summer workshop and use it almost everyday in my work as a reporter at tv2.dk."

Hard Truth

David uses AI "almost everyday" and has already taken a TV 2 summer workshop course. He scores a 7. So after daily use and formal training, he's still only at 70%? What he wants from this training is revealing: "more confidence" and "better prompting" and "critical sense toward AI." Confidence isn't a skill. Better prompting is what everyone wants. And if you're using AI every day and still need to develop a "critical sense toward AI" — what exactly have you been doing with it every day? Using a tool daily without developing critical awareness of it isn't proficiency. It's muscle memory.

HOW FRAMING WORKS

This report says:
"Daily user plus formal training equals... a 7. Still lacks confidence and critical thinking about his own tool."
The positive report says:
"Already embedded AI into daily workflow. Seeks deeper mastery rather than surface-level tricks."

Pelle Lykkebo Mørk

TV 2 journalist · AI implementation
8 / 10
"I've used AI extensively across research, drafting, live news production — and also worked on internal implementation, building prompts and workflows adapted to newsroom needs."

Hard Truth

Pelle is the group's internal AI evangelist: implementation, workflows, prompts, live news production. He verified US E-3 Sentry damage at Prince Sultan Air Base — a genuine OSINT verification. Impressive on paper. And yet he scores himself an 8, not higher. Why? Perhaps because his concerns betray the limits of his comfort: he worries about "offloading editorial judgement to AI" and "data security with commercial LLMs." These are legitimate concerns — but they're also the concerns of 2023. In 2026, the industry has moved to discussing how to manage these tradeoffs, not whether they exist. For someone building newsroom AI workflows, being stuck on foundational anxieties is like an architect who's still worried about whether steel is safe.

HOW FRAMING WORKS

This report says:
"AI evangelist still stuck on 2023 anxieties. Builds workflows but worries about fundamentals."
The positive report says:
"Most advanced practitioner in the group. Real OSINT verification experience. Thoughtful about ethical boundaries."

Emil Gjerding Nielson

TV 2 journalist · Editorial AI implementation
8 / 10
"Working on implementing editorial ai tool for my fellow colleagues. Use it in my writing amongst other things."

Hard Truth

Emil is "working on implementing editorial AI tool for my fellow colleagues." He's building the AI infrastructure for the newsroom. He scores an 8. But look at what he wrote: 26 words. Twenty-six words to describe his AI experience on a form designed to help a trainer understand his needs. "Use it in my writing amongst other things." What other things? He doesn't say. His goals: "automate feedback, improve writing." His success criterion: "Insight into tools, prompts etc." For someone implementing AI tools for an entire newsroom, this intake is almost insultingly thin. He's either too busy to bother, or there's less behind the 8 than the number suggests. When the person responsible for AI implementation writes "amongst other things" and can't be bothered to list them, what message does that send to colleagues?

HOW FRAMING WORKS

This report says:
"26 words for the AI implementer. Too busy to fill in a form properly. What's behind the 8?"
The positive report says:
"Key AI champion in the newsroom. Already building tools for colleagues. Leading by doing."

Bo Bergstedt

TV 2 journalist · Self-described "outlier"
9 / 10
"Have been involved since I started doing deepfakes back in 2018. So don't customize the course for me - I am an outlier at TV2."

Hard Truth

Bo gives himself a 9 and immediately tells the trainer: "don't customize the course for me — I am an outlier at TV2." That's not confidence, that's arrogance on an intake form. He "started doing deepfakes back in 2018" — a flex that raises more questions than it answers. What was a TV 2 journalist doing with deepfakes in 2018? His goal is "building tools to help journalists battle AI slop" and his success criterion is "seeing journalists getting more comfortable with AI." So he's not here to learn — he's here to watch his colleagues learn. He's appointed himself the observer, the mentor, the person who doesn't need the training. A self-assessed 9 who tells the trainer not to bother adapting for him. If he already knows everything, why is he in the room?

HOW FRAMING WORKS

This report says:
"Arrogant intake. Here to watch, not learn. If he knows everything, why attend?"
The positive report says:
"Most experienced participant. Pioneer since 2018. Generous mentor focused on lifting the whole team."

Christian Jessen

TV 2 journalist
9 / 10
"I am very skeptical with AI as a tool for generating knowledge as in my experience, AI can and will make mistakes."

Hard Truth

Christian scores a 9 but is "very skeptical with AI as a tool for generating knowledge because AI can and will make mistakes." A 9 out of 10 who is very skeptical about the core function of the technology. That's like a pilot who scores 9/10 on flying skill but is "very skeptical" about whether planes can actually fly. He generated images and video "mostly for fun" — so his creative AI usage is recreational, not professional. He uses AI "mostly to summarize large amounts of information" and correctly identifies that "most AI-software are language models, so AI is usually great at generating and improving text." This is accurate but also extremely basic knowledge for someone claiming a 9. His wish list — "outside-the-box thinking" — is ironic from someone whose AI philosophy is deeply inside the box of skepticism.

HOW FRAMING WORKS

This report says:
"Scores 9 but deeply skeptical about AI's core function. Wants outside-the-box thinking from inside the box."
The positive report says:
"Sophisticated understanding of AI as language technology. Healthy skepticism grounded in real experience."

Marie Møller Munksgaard

TV 2 journalist
? / 10
"I use it mostly for research for different articles. To find experts to interview or what has previously been written about something."

Hard Truth

Marie is the only participant out of 20 who didn't provide a self-assessment score. Not a 1, not a "prefer not to say" — just blank. In a form where 19 colleagues managed to produce a number, Marie couldn't. Or wouldn't. She uses AI "mostly for research" and "to find experts to interview or what has previously been written about something." That's a search engine. She also uses it "to give a telegram from Reuters some context." Giving context to a wire story is one of the most basic editorial tasks that exists — and she needs AI for it. Her success criterion is modest: "If I go home with one or two new ways to use AI in my everyday work life." One or two new ways. The participant who can't even score herself hopes for one or two takeaways.

HOW FRAMING WORKS

This report says:
"Couldn't even rate herself. 19 out of 20 managed it. Needs AI to contextualize a wire story. Hopes for one takeaway."
The positive report says:
"Practical, grounded approach. Already using AI for the exact tasks journalists face daily. Realistic expectations."

The lesson of this report

This report has portrayed twenty TV 2 journalists as negatively as possible. Every claim is based on what you yourselves wrote. Nothing was invented. And yet it feels unfair.

That is exactly the point.

  • 1. Framing is selection. This report consistently chose the most vulnerable word, the weakest example, the most painful interpretation. Everything factual — but one-sided.
  • 2. AI does what you ask. The instruction was: be as negative as possible. The AI obeyed. Without conscience, without context, without restraint. It attacked spelling errors, mocked modest goals, and turned healthy skepticism into incompetence.
  • 3. You recognize this. As journalists, you know how quote selection works, how a lead sets the tone, how the sequence of facts drives a narrative. This report did exactly that — with you as the subject.
  • 4. This is how your sources feel too. When they find how their words have been framed in an article. When nuance disappears. When context is stripped away. This report holds a mirror.
  • 5. The data didn't change. Every quote, every score, every goal in this report is identical to the positive version. Only the frame changed. Same facts. Completely different story. This is what AI bias looks like in practice.

Now read the positive report about the same twenty colleagues. Same data. Completely different story.

Score Distribution

3
2
4
1
5
5
6
4
7
3
8
2
9
2
N/A
1
6.0
Mean
6
Median
3–9
Range
1.7
Std. Dev.