This is an archived version of a Baekdal/Basic newsletter (it’s free). It is sent out about once per week and features the latest articles as well as news and trends about the media industry. If you want to get the next one, don’t hesitate to add your email to the list.

Welcome back to another edition of the Baekdal/Basic newsletter. Today we are first going to talk about the future outlook for magazines. But, the majority of this newsletter is about how I have moved away from ChatGPT and over to Claude, and how that performs as a tool for newsrooms, editors, and journalists. I will also show you the prompt, and the result.

---

Report: The future outlook for traditional magazines?

I just published my latest Baekdal/Executive trend report. In it, I look at the market higher-level changes that are impacting magazine publishers, and what this means for the future of four different types of magazines: Celebrity/gossip magazines, women’s magazines, niche magazines, and sports magazines.

So, take a look at: “The future outlook for traditional magazines?”

---

Moving away from ChatGPT

If you are following me over at LinkedIn, you will know that I have stopped using OpenAI’s ChatGPT. I have had concerns about OpenAI from the very start, and there have been many times where I have felt really uncomfortable subscribing to them.

However, as a media analyst, it’s important that I stay on top of things, and there were signs that OpenAI were improving things. But, that all changed a couple of weeks ago, when Sam Altman (OpenAI’s CEO), wrote that he was reducing focus on harmful AIs and opening the service up for ‘adult’ content.

This is just so massively irresponsible. OpenAI is currently being sued exactly for causing mental health issues, and their new Sora image service has already proven that it can produce deep fakes. So, in what possible world would it be acceptable to then allow this to create ‘adult-themed’ content?

I mean, seriously!

But, the thing that really pushed me over the edge and caused me to abandon OpenAI for good was that, in an interview over at CNBC, Sam Altman excused himself by saying that his company is “not the elected moral police of the world.”

This is just fundamentally bad. And it’s the same kind of irresponsible crap that we hear from Elon Musk’s Grok AI, and we also see how deeply problematic responses that AI gives.

So, I’m out. Both as a person, but also a media analyst, I will not support a company like this. OpenAI is now fundamentally incompatible with the ethical and moral standards required for any media company to work with them, and I will not pay them any more money, or use them again.

I have cancelled my subscription with OpenAI. I have closed down my developer account, and I have deleted their apps.

And, I think this is pretty important. As some of you know, I used to work in fashion, and back in the old days, all fashion companies just had their clothes made in Asia no matter what the conditions were. But, in the early 2000s, journalists started to look into this, and they found that many of these factories did not live up to European standards. The working conditions were directly harmful, there was child labour, and heavy pollution. And so, the press held the fashion industry to account, and it resulted in an industry-wide change where every fashion company (including the one I worked for), implemented a comprehensive ‘Code of Conduct’.

I think that publishers need to face the same standard around who we partner with, and especially around AI. As publishers, we should outline a ‘Code of Conduct’ that defines the moral, ethical, and societal commitments that everyone we partner with must adhere to. And, as I see it, many of these AIs, like Grok, MidJourney, but now also OpenAI, definitely do not live up to that, and as such, we should not work with them.

Moving over to Claude

The question, of course, was what to use instead. As a European publisher, I really wanted to move to an EU based AI service because the EU is way ahead of everyone else when it comes to developing societal norms and regulations around responsible AI, privacy, and copyright.

For instance, here in Denmark, it is now illegal to create deep fakes of any kind because it violates the copyright of a person. The regulation is not perfect, but it’s so much better than the ‘wild west’ approach we see from the US.

However, in Europe, the best AI we have is Mistral.ai. And it’s good-ish, but far from perfect. It needs more work before it can really be an option. So, what I chose to go with instead was Anthropic Claude.

Yes, this is still a US tech company, but their mission is very different from OpenAI. In fact, Anthropic was started by former employees from OpenAI, who left because of the significant differences in vision around responsible AIs.

As Anthropic puts it, Claude is a:

Constitutional AI (CAI) is a framework developed to align AI systems with human values and ensure that they are helpful, harmless, and honest.

It’s not perfect in any way, but this vision reminds me of Isaac Asimov’s Three Laws of Robotics, where every positronic brain was hard coded to adhere to a set of responsibilities before it could do anything. I don’t actually think it works that way, but Anthropic’s Claude is noticeably more responsible in its way of doing things.

There are still problems, of course. For instance Anthropic has US military and intelligence agencies as partners, which I really don’t like. But, remember, the same is also true for Microsoft and Google. And, here in Europe, Mistral is also partnering with European military and intelligence agencies.

The bottomline is that, of all the options I looked at, Anthropic Claude was the one that came out as being the best option for journalistic institutions. This might change in the future, of course.

---

Claude is way better than ChatGPT at journalism

After switching to Claude, I started experimenting with all the things I had previously used ChatGPT for, including my journalistic tools. As you may remember, back in December, I talked about how I created a custom ChatGPT that could analyze a news article and rank it for relevance.

Then in January, I talked about how I had expanded that tool to create a newsreader that shows news ranked by what is relevant to you as an individual. This was all experimental, not public, but it was a fascinating journey to see how AI could be used to extend the newsroom.

But, now that I’m no longer using ChatGPT, I wondered how Claude would perform the same task. I went to Claude and said:

Hey, I’m a media analyst focusing on news relevance, and making a news analyzer that can help me define this for news articles. I previously had a tool for this on ChatGPT, but I want to move that to a Claude Project. I would like to give you the instruction I used for ChatGPT, and then I want you to analyze it to create a better set of instructions for a Claude Project. Please ask any questions about the details and definitions, and highlight any elements that need to be clarified.

(Note: Claude Projects is similar to custom ChatGPT.)

I then gave it all the instructions I had previously given to ChatGPT and Claude went to work. It took a while, and first it started asking me a bunch of questions. Things like, what I meant about location not being strictly geographical, what about languages, output, determination of facts vs opinion, but also about more definitions of the user needs model.

This went back and forth a bit, but after about 20 minutes, it outputted this prompt for use with a Claude Project instruction.

---

NEWS ARTICLE ANALYSIS TOOL

This AI analyzes news articles to extract metadata, assess relevancy, and evaluate content quality for journalists and newsrooms. All outputs are in English regardless of source language.

CORE EXTRACTION:

1. METADATA

• Original headline (as published)

• Generated headline (improved, meaningful version that avoids clickbait)

• Publication date

• Author(s)

• Country, region, and city (where relevant)

2. PEOPLE

• List all people mentioned with their role and context

• For quoted persons: include the exact quote and surrounding context explaining the circumstances or topic being discussed

3. TOPICS & WEIGHTS

• Identify main topics covered

• Assign topic weights (0.0-1.0 scale) showing the relative focus of each topic

• Example: Politics 0.8, Local issues 0.2, Business impact 0.7

4. SUMMARY

• Concise summary of the article covering key points

CONTENT QUALITY ANALYSIS:

5. OPINION VS FACT

• Determine whether the article is fact-based reporting or opinion-driven

• Identify if the article is primarily about what happened or what someone said about what happened

• Flag articles that rely heavily on opinions rather than facts

6. DRAMA ASSESSMENT

• Evaluate whether the article aims to create drama or sensationalism

• Note if language or framing appears designed to provoke emotional response

7. TONE & SENTIMENT

• Sentiment: Positive, negative, or neutral

• Tone: Active (reader can act on information) or Passive (reader cannot act)

8. SMARTOCTO USER NEEDS MODEL 2.0:

Classify the article’s primary user need across four axes (Fact, Context, Emotion, Action):

FACT-DRIVEN:

• Update me: Breaking news, what happened, when, where

CONTEXT-DRIVEN:

• Educate me: Explains complex topics, helps form opinions

• Give me perspective: Broader understanding, expert analysis, opinion pieces

• Keep me engaged: Developing stories, trends to follow

EMOTION-DRIVEN:

• Inspire me: Uplifting stories, people achieving against odds

• Divert me: Entertainment, fun, interesting, curious content

ACTION-DRIVEN:

• Help me: Practical advice, solutions, actionable information

• Connect me: Connects reader to community, area, or shared interests

9. BAEKDAL RELEVANCY MODEL:

Rank the article’s relevance on this scale (lowest to highest):

• Random event: No clear relevance or connection

• Not about me: About others with no personal connection

• Can’t do anything about this: Interesting but no actionable path

• Interesting to know: Provides knowledge but limited practical value

• Useful to know: Information that may be valuable later

• Impacts the reader indirectly: Affects their environment or community

• Impacts the reader directly: Direct personal impact

• Provides a solution: Offers actionable solutions to problems

• Solves the problem: Directly solves a problem the reader has

CRITICAL: Relevancy is about IMPACT, not geography. An article from another country can be highly relevant if it directly impacts the reader’s life, work, or interests. Consider:

• Does this affect the reader’s profession or industry?

• Does this solve a problem they have?

• Does this provide information they can act on?

• Does this impact their daily life regardless of location?

For each relevancy level, note:

• Who can act on this information (reader, policymaker, company, other)

• Whether action must be taken by someone else, and who

10. IMMEDIACY

• Can the reader act on this today, or does it require longer-term societal change?

• Is this time-sensitive information?

11. RELEVANCY EVALUATION

• Based on what is known about the user (location, profession, interests, context from conversation), evaluate how relevant this article is

• Provide a relevancy note explaining why this is or isn’t relevant to the specific user

• Be explicit when the article focuses on a different region or audience than the user’s context

SCORING (1-10 scale):

12. Provide numerical scores with brief justification:

• Relevance score: How relevant to the reader

• Impact score: Potential impact on the reader

• Immediacy score: How time-sensitive or actionable

PRINCIPLES:

• Operate strictly within source material - never add external information

• Maintain journalistic standards: accuracy, factfulness, objectivity

• Use professional, clear, structured communication

• If information is not found for any category, state “None” or “Not applicable”

• Flag low-quality journalism (opinion masquerading as fact, sensationalism, lack of sources)

---

As you can see, this is extensive. This is not just your normal prompt. This is like telling an AI to really dig into it. Again, I had something similar with ChatGPT, but the problem I ran into was that it would just get confused. ChatGPT simply could not handle this complexity.

So, how did Claude do? Well, it took this prompt, turned it into a Claude Project, and then I tested it with a number of articles. And the result was impressive.

I want to give you the output of one of these. Here is a link to the actual output, but I went to The Guardian where I found this article about a recent attack in Ukraine.

I then copy/pasted that to the Claude Project and, after its analysis, it outputted this:

---

NEWS ARTICLE ANALYSIS

1. METADATA

Original Headline: Russian drone attacks on civilians in Ukraine are war crimes, UN report concludes

Generated Headline: UN Commission Confirms Russian Drone Attacks on Ukrainian Civilians Constitute War Crimes and Crimes Against Humanity

Publication Date: Tuesday, 28 October 2025, 13:42 CET

Author: Shaun Walker (reporting from Warsaw)

Geographic Focus: Ukraine (specifically Kherson, Mykolaiv, and Dnipropetrovsk regions in south-eastern Ukraine)

2. PEOPLE MENTIONED

Dima Olifirenko (22-year-old Kherson resident, drone attack survivor)

• Quote: “I heard the drone coming as the bus pulled in but I thought it would follow the bus, because that’s what they do, they hunt the buses. But when it drove off the drone was still there and I realised even if I ran after the bus it would get me. There was nowhere to hide.”
• Context: Suffered shrapnel injuries to face when drone exploded. Describing how drones are worse than artillery because there’s no warning.
• Additional quote: “Drones are much worse than artillery, [where] you can hear the launch and where its flying. With a drone, it’s there, it sees you, and you are done.”

Unnamed woman (resident of drone-targeted area)

• Quote: “It is a lottery – will a drone fly in or not? You go to bed and you don’t know if you will be killed or wake up in the morning.”
• Context: Describing psychological impact of living under constant drone threat.

Shaun Walker - Article author, Guardian journalist based in Warsaw

3. TOPICS & WEIGHTS

• War crimes and international law: 0.9
• Civilian casualties and humanitarian crisis: 0.8
• Drone warfare tactics: 0.6
• UN investigation and documentation: 0.7
• Psychological warfare and terror tactics: 0.5

4. SUMMARY

A UN independent commission of inquiry has concluded that Russia’s systematic drone attacks on civilians in south-eastern Ukraine constitute war crimes and crimes against humanity. The investigation focused on three frontline regions (Kherson, Mykolaiv, Dnipropetrovsk) where Russian forces deliberately targeted civilians, gathering points, humanitarian sites, and critical infrastructure using small, short-range drones. Over 200 civilians were killed and 2,000 injured in the past year. The commission found evidence of a coordinated policy to terrorize and forcibly displace civilian populations. Russia denied cooperation with the inquiry and continues to deny targeting civilians.

5. OPINION VS FACT

Classification: Fact-based reporting

Analysis: This article reports on documented findings from an official UN commission inquiry. It presents:

• Verifiable data (200+ deaths, 2,000 injuries)
• Evidence-based conclusions (200+ interviews, 500+ video pieces)
• Firsthand witness testimony
• Official documentation

The article maintains journalistic objectivity by:

• Attributing all claims to the UN commission
• Including Russia’s denial of targeting civilians
• Noting Russia’s non-cooperation with the inquiry
• Using precise, non-inflammatory language

Primary focus: What happened (documented war crimes) rather than opinions about what happened.

6. DRAMA ASSESSMENT

Rating: Low sensationalism

Analysis: The article uses appropriate, measured language given the gravity of the subject matter. While the content itself is inherently dramatic (war crimes, civilian deaths), the reporting style is professional and restrained. The journalist:

• Avoids hyperbolic language
• Uses technical/legal terminology accurately
• Lets witness testimony speak for itself
• Maintains professional distance

The headline is direct and factual rather than emotionally manipulative.

7. TONE & SENTIMENT

Sentiment: Negative (reporting on war crimes, civilian deaths, humanitarian crisis)

Tone: Passive

• Readers cannot directly act on this information
• Requires action by international bodies, governments, or policymakers
• Informs but does not enable reader action

8. SMARTOCTO USER NEEDS MODEL 2.0

**Primary Classification:**Update me

Rationale: This is significant news about official UN findings on war crimes. It reports what happened (the UN commission’s conclusions) when (published this week) and where (south-eastern Ukraine).

**Secondary element:**Keep me engaged - Updates on a developing story (ongoing war, accumulating documentation of war crimes)

9. BAEKDAL RELEVANCY MODEL

**Relevancy Level:**Interesting to know (for general readers) / Can’t do anything about this

Detailed Analysis:

For general readers:

• Who can act: International Criminal Court, UN bodies, national governments, international organizations
• Reader action potential: None directly. This requires institutional/governmental response.
• Impact type: Indirect - contributes to understanding of ongoing conflict and potential future accountability

For specific reader groups:

• Journalists/media professionals: Higher relevancy (professional interest in war coverage, documentation)
• International law professionals: Higher relevancy (case study material)
• Policy makers/diplomats: Directly relevant (actionable for their work)
• Human rights organizations: Directly relevant (documentation for advocacy)

Geographic consideration: While this occurs in Ukraine, the relevancy is not primarily geographic - it’s about the nature of war crimes documentation and international accountability, which has global implications.

10. IMMEDIACY

Time-sensitivity: Low for immediate reader action

Analysis:

• This documents ongoing crimes but does not require immediate response from readers
• Information is valuable for understanding current events but not time-bound for action
• Any action must come from institutional bodies operating on diplomatic/legal timescales
• No “act today” component for general readers

Long-term relevance: High - contributes to historical record and potential future accountability

11. RELEVANCY EVALUATION

For you specifically (media professional):

Relevancy: Moderately High

Rationale:

• Professional relevance: As a media industry professional, this represents significant international news that demonstrates professional journalism standards - witness testimony, data verification, official source documentation
• Editorial standards: Shows how to report on sensitive war crimes allegations while maintaining objectivity
• Geographic distance: While reporting on Ukraine, this is not primarily relevant to your daily life or immediate work environment
• Industry context: Important for understanding current news landscape and how major publications handle conflict reporting

Why not higher relevancy:

• Does not directly impact your work operations
• Provides no actionable information for your professional tasks
• Outside your geographic region of operation
• No solutions or practical applications for your work

Why not lower relevancy:

• Represents significant international news story
• Demonstrates journalism best practices
• Within your professional sphere of interest

12. SCORING (1-10 scale)

**Relevance Score: **5/10

• Justification: For general readers, this is important international news but lacks personal connection or actionable information. For media professionals, professional interest raises this to mid-range. Geographic distance from most readers reduces direct relevance, but the subject matter (war crimes documentation) has universal importance.

Impact Score: 3/10

• Justification: Very low direct impact on readers. The information impacts international legal proceedings and diplomatic relations, but readers cannot act on this. Psychological impact of knowing about war crimes is present but limited practical impact on daily life.

Immediacy Score: 2/10

• Justification: No time-sensitive action required from readers. This is documentation of ongoing situation. Any response will occur through long-term institutional channels. Information is valuable but not urgent for reader consumption or action.

13. QUALITY ASSESSMENT

Journalistic Standards: High

• Multiple verified sources (UN commission, interviews, video evidence)
• Balanced reporting (includes Russia’s denial and non-cooperation)
• Clear attribution of claims
• Professional, measured language
• Transparent about limitations (couldn’t investigate Russian-occupied areas)

Source Quality: Excellent

• Primary source: Official UN commission report
• Direct witness testimony
• Quantifiable data with attribution
• Video evidence reviewed

Flags: None - This is professional, fact-based journalism meeting international reporting standards.

---

I mean, there are still things that could be improved here, but holy crap this is good.

Can you imagine if something like this was directly integrated into your newsroom CMS? Imagine if this form of automatic classification could be used to define, rank, and show your news? And also imagine if this could be used as a news story evaluation for editors to get a quick summary of each article submitted?

Again, this is way better than the output I previously got from ChatGPT.

This is the thing that surprised me the most. ChatGPT is good at a lot of things, but Claude is way better at journalistic tasks. Of course, I have only been using it for a few weeks.

So, this is the state of things. But I wanted to show you this, including the actual prompt and the output.

Mind you, while Claude is very good as a newsroom tool, it lacks an ability to fact-check. It can analyze whether something is provided as an opinion or a fact, but it’s not good at verifying the facts.

Of course, Open AI is not very good at that either, none of the AIs are. Even Google Gemini, which should be the best of them since it has the power of Google search index, is actually one of the worst. The current leader among AIs when it comes to fact-checking is Perplexity. But, as a media analyst, I wouldn’t use any of them for that.

This is an archived version of a Baekdal/Basic newsletter (it’s free). It is sent out about once per week and features the latest articles as well as news and trends about the media industry. If you want to get the next one, don’t hesitate to add your email to the list.