I tried to write a blog post with chatt, Claude and Gemini – here is how you stacked yourself

AI-generating blog post ideas

Before I go on, I would like to be hyper-clear in one thing: AI cannot read her mind.

It doesn’t know much about you – your tone, your style or How you think. But give him the right context, and it could quickly start to live for funny headlines and like to appear as helpful and apologetically directly.

Therefore, the first step in my process is to set the expectation stage. Just like you would do it if you have a new team member on board.

Before I put these tools to the test, I gave them the same background information and then fed them with the same input requests.

Would you like to learn more about my approach? I recently wrote about writing AI content in the brand and shared my approach (and recorded tips from other professionals).

Example requests

Here is the example with which I introduced myself to the new team members, um … tools.

“Hi (tool name). I am a marketing strategist and content expert. I write for an audience of business owners, marketers and creative experts who improve their visibility, connect them more closely with their audience and want to use messaging as a lever for business growth.

I have specialist knowledge in e -mail marketing, messaging strategy and in the use of AI tools such as chatt and claude to improve workflows and copies. My style is direct, talkative and strategic.

My content destinations:

• Show my POV and build trust through implementable, lived knowledge.
• Help my audience to write and communicate more effectively with tools like AI.
• Spark discussions with people who appreciate clear thinking and human content.

About my audience:

• Trainer, consultant, service provider and content manufacturer.
• Many test or curious about AI, but are not sure how to use it well.
• They appreciate voice, connection and strategic depth over cheap hacks.

Please let me know if you have any questions about me before I guide you through the next entries. “

After I had clarified everything, I gave every tool the same task:

“I’m looking for blog post ideas for AI-driven letter from workflows and how to train tools to sound how they sound. Please generate 10 different, specific and click-worthy ideas that avoid clichés. Contain a brief explanation for everyone.”

Remarkable – all three tools delivered 10 great ideas, but my screen could only record so many. So I share screenshots of the smaller inquiries and a text list of the 10 post ideas.

How Chatgpt was executed

Oh chat, how do I love you? Let me count the paths.

Chatgpt has one leg up here because it knows me (and openly probably also where I live.) I would say that all of these tools are solid.

Here are Chatgpts Top 10 Ideas:

1. The first 5 input requests that I give a AI tool before I let it touch my content
2. How to create a Ki description flow that does not flatten out your voice
3. Training from Chatgpt to write like you? Start with these 3 voice anchors
4. What Ai cannot hear (unless you teach it): the hidden signals in your writing style
5. From an empty side to the final draft: my end-to-end workflow with chatgpt
6. Why your AI letter sounds – and how you can fix it quickly
7. How I use AI as a creative partner (not as a crutch)
8. Teach Chatgpt how to think: a framework for strategic content requests
9. I feed the exact language instructions (and how you can make your own)
10. Your voice, if necessary: ​​How to create a reusable fast library that sounds like it

I decided with “the first 5 input requests that I give every AI tool before I had my content touched.”

How Claude fell off

Claude is fast. How attractive. And I love that I didn’t have to make a lot of request to understand what I wanted.

I hadn’t planned because I would like to test the raw editions of every tool, but it was an absolute pleasure that I didn’t have to go down this way.

Here are Claude’s top 10 ideas:

1. The fingerprint method: How I trained Claude to write in my voice (with examples)
2. Beyond templates: Creating Dynamic AI writing systems that develop with your brand
3. The conversation-to-content pipeline: recording client calls to feed your AI writing system
4. Ethical language clones: Where can you draw the border when you train AI to sound like you or your team
5. The 80/20 rule of the KI letter: identify your “signature phrases”, which have to do content unmistakable
6. Language stratification: The 3-pass method for adding personality to ai-generated content
7. From friction to the river: Create custom AI tools that eliminate your specific writing pain points
8. The authenticity paradox: How AI can actually make its content more human, no less
9. Content ecosystem assignment: Design of KI workflows that connect over several platforms
10. Voice -Dna: Creating a practical style manual that can follow both humans and AI

I decided with “The Authenticity Paradox: How Ki can actually make its content more human, not less.”

How Gemini worked

Gemini was a bit like a dark horse because I had less experience with it. But here too it is incredibly fast and developed some really decent ideas. I would argue that they were some of the most creative, even if they are a little formula.

1. Beyond the bot: The “Ghost in the Machine” method for training Ki to embody your branded voice.
2. Stop asking AI, “how to sound”: the strategic shift in language enamel for authentic connection.
3. From generic to genius: the feedback loops that transform AI writing tools into their eerie content.
4. The advantage of “human-in-the-loop”: AI provider content that does not sound like a robot wrote it.
5. Switch on your inner author: Use AI to bypass the empty side and strengthen your clear perspective.
6. The AI ​​whisperer’s tool kit: essential input requests and techniques for the creation of language information.
7. More than keywords: training of AI in their core messages for consistent brand communication.
8. Beyond processing: the strategic art of rewriting the content of AI generated in order to really own their voice.
9. The ethical AI author: integration of efficiency with authenticity in your content workflow.
10. The future of “Sie” in content: How AI will develop personal voice and brand identity.

I decided to continue with the advantage “The ‘Human in the Loop”: Content with AI-Appointed content that does not sound like a robot, wrote it. “

The winner?

Hard call. I think everyone did great here.

Here are a few things that I noticed that I emphasized for myself and felt more:

• Gemini used words such as “manufactured”, which is often a giveaway for me.
• Both Gemini and Claude created repeating structures for the titles with a powerful idea, followed by a large intestine and a more detailed explanation.

In order to be more official, I used the following questions as a section to see how the individual came about:

Did the tool reflect an exact understanding of me and my audience?

• Chatt: Yes, but it also has an unfair advantage. Did that come into play? We will never know.
• Claude: Absolutely. All of these topics sound like something that I would interest in writing or reading about it.
• Twins: I think so. They are pretty interesting and I love some of the ideas that have come up.

Has it hit my tone, pov and priorities?

• Chatgpt: Yes. But while it came to things that I would write #rightnow, it felt a bit more generally, which feels an interesting observation for me.
• Claude: A solid yes here.
• Twins: Also yes.

Were the ideas fresh, not clichéd? Do I want to click?

• Chatgpt: Yes, but although I liked the structure here more, I like some of the ideas better in the others.
• Claude: Absolutely. Have to work on the structures, but that can be fixed while we roll.
• Twins: Same answer as for Claude.

So who is the winner?

To be honest, too early to tell it. I will say I was incredibly impressed by the idea of ​​thought and the details of the detail in Gemini’s ideas.

Outline of the contribution

As soon as I had a list of strong mail ideas, I chose one to advance with every tool

Now the next test came: How well could every tool transform this idea into a structured, human-friendly outline?

Well, let us find out! I gave every tool the same entry request.

“Please create a structure for a blog post of approximately 800 words with the title” (Insert the title) “. Use the H2 and H3 structure and give an individual summary of what every section should cover. Keep it human and logical.”

So it worked (note: For reasons of brevity, I left out the individual summary and concentrated on the sections and key points for everyone):

How Chatgpt was executed

Here is Chatgpt’s outline:

The first 5 input requests that I give a AI tool before I let it touch my content

• AI can be a powerful writing partner – but only if you train him from the start. So I set the stage with five basic requests that save me time, protect my voice and make sure that I have control.
• Why AI needs context – not just commands

• The myth of the “perfect input request”
• Remember how to take a new team member on board

• Entry request No. 1: “Here is who I am and for whom I write …”

• Why is that important
• Example request

• Entry request No. 2: “So I want my letter to feel …”

• It’s not just a voice – it’s mood
• Example language to include

• Entry request No. 3: “Here is what I think and how I talk about it …”

• Consistency strikes prudence
• Voice anchor and phrasing pattern

• Entry request No. 4: “So I want the reader to feel and act …”

• The sound follows the purpose
• Example of calls

• Entry request No. 5: “Here is the format that I should follow …”

• The guardrails that keep things up to date
• Reusable input template

• Last thoughts: better inputs = better outputs

• A simple checklist for your next input prompt session
• Start here, then develop

To be honest, I think this is a pretty solid start. It feels pretty close to the mood I want, and although I have to share more of my opinions and my writing style, I think that we (I and GPT) could get into the finish line pretty quickly.

I also love that it called the conclusion, while other tools only described it as a boring old “conclusion”.

How Claude fell off

Here is Claudee’s outline:

The authenticity paradox: How AI can actually make its content more human, no less

• Request the common assumption that the content of ai-generated content is missing from humanity
• Set up the paradox: How the right AI approach can actually improve authenticity
• Thesis: If you are used strategically, AI can free you, to be more human in your content, no less

• The real problem is not ai – it is how we use it

• The copy paste trap
• The strategic alternative

• The liberation effect of the bandwidth

• Automate the everyday
• Invest in deep work

• Practical techniques for more human-supported content

• The language-first workflow
• The sample reinforcement method
• The editorial process of man in the loop

• Measurement of real content authenticity

• Beyond commitment metrics
• The most conversation

I won’t lie, I love that. It feels Hochbrow and well thought out. This is definitely something that I can write myself. Could take a little more time, but this is a solid article that I could really get behind.

How Gemini worked

Here is Gemini’s outline:

The advantage of “human-in-the-loop”: Creating AI-affirming content that does not sound like a robot wrote it

• Introduction: Briefly set the Allure and the pitfalls of AI-generated content and emphasize the decisive role of human supervision in maintaining authenticity.
• The siren song of pure AI generation (and why it falls flat)

• The promise of the effortless creation of content
• The unmistakable taste of the algorithm
• Why “good enough” is not good for your brand

• The frame “human-in-the-loop”: your secret weapon

• Defining the “human-in-the-loop” approach
• The critical phases for human intervention
• Why this hybrid approach provides superior results

• Practical strategies for the introduction of their humanity into AI-generated content

• Training from AI with her voice: Beyond basic input requests
• The art of strategic rewriting: Add nuance and perspective
• Fact examination and ethical supervision: maintain trust and accuracy
• Iterative refinement: The power of feedback loops

• The future is hybrid: Inclusion of cooperation with AI

• AI as a powerful assistant, no substitute
• Building a sustainable and authentic content workflow

Short answer – I don’t love it. It has some great ideas, but it feels less tactical and content. It is even more important, I think there is a pretty large elevator to bring it into my voice and share my thoughts.

The winner?

Before I go on, there was no clear winner, but there was one that I didn’t like so well and you probably guessed it. Sorry, Gemini. Better luck next time.

This means that here are the questions I have considered:

Does the outline flow logically?

• Chatgpt: Yes!
• Claude: Yes.
• Twins: Yes.

Were the sections helpful and clear?

• Chatgpt: Yes.
• Claude: Yes.
• Twins: Yes.

Was there enough depth/details for each section?

• Chatgpt: Absolutely.
• Claude: Absolutely.
• Twins: This is a no. Instead of offering specific advice, it offered a generic observation at a higher level that I would need more help to think through thinking.

Did it support the kind of blog post that I want to write?

• Chatgpt: Secure.
• Claude: Absolutely. 1 million percent.
• Twins: To be honest, not really.

Based on my answers to these questions, I take my statement back in order not to have a clear winner. There was one – and it was Claude.

Write the post

It gets interesting here.

Developing ideas and outlines is fun, but writing the thing? This is the real game card here. And I had an idea of ​​what I would expect – a hypothesis if you like.

Each platform already had an overview of who I am and who I serve but not how I write. So I shared this prompt with each platform:

“Here are three patterns of my writing style so that you can make an idea of ​​how I think and what I think. Please internalize this tone and keep it consistently while helping with the generation of ideas, surfing and writing.

Please inform me of questions before I give you the next request. “

Everyone came back with a careful description of my writing style, which was switched on with what I shared in the very first prompt.

Next I shared this prompt:

“Based on the outlines that you have just made for a blog post about writing with AI, write a contribution. Keep it directly, helpful, conversal -and please to convey the love for clarity when you share examples -if you support yourself in thoughts that you have identified in my examples and avoid all cliché -Ki constructions or phrasing.”

Here are the first paragraphs that every platform has developed. For reasons of brevity, I will not pass on the respective article. However, this should give you a feeling of content, rapprochement and voice.

How Chatgpt was executed

First thoughts?

That sounds like me. But there are definitely some AI Sentence Structures that I try to stay on a minimum-namely:

• Ask? Followed by a disarming sentence and one of the answer.
• Parallel construction. Don’t just do it … do that instead.
• Three short sentences start with the same word.

Now it is known that I do all of this. But I always pay attention to such content because it just feels Well, average.

How Claude fell off

Okay-one kind of fainting here, because although this feels more than the most I would write, it is also a bit close to something I would write.

It is too short here. I would never:

• Open a post with “Look, I understand it.” I would save this for later in the content.
• Talk like “every time someone raves …”
• Make strange analogies such as “a committee of robots who desperately try to pass the Turing test”.

But I don’t think it would take so much work to polish it and bring it into something I would say.

How Gemini worked

I will stay on my weapons here. I still love the idea of ​​”human-the-loop” absolutely. It has an A of me.

Execution is a flat C minus. I’m sorry, Gemini, I realize that this is a first date and everything, but that definitely doesn’t work for me.

Although the title promise not to be written like a robot, wrote it, this definitely screams AI. Here are the stories for me:

• The word “craft”
• In the middle of the conversion “We all saw …”
• Language that sounds good, but realistically feels empty.
• Analogies that are catchy, but again unfounded “The Siren Song”.

However, this is not good again, twins. We are still being edited and after that I am ready to give you another chance.

The winner?

It is a scarce, but Claude is at the top, followed by chatt. Gemini is not near the finish line.

Here are the metrics that I considered:

Did the design feel like me?

• Chatgpt: Yes!
• Claude: Yes!
• Twins: So. Much. NO.

Was it well structured, with a logical flow?

• Chatgpt: Yes, I give that.
• Claude: Yes!
• Twins: I will give it a yes. River was not the problem that was substance.

Was it to be robot, or did it show personality?

• Chatgpt: A little robot side, but close and a simple solution.
• Claude: Same answer as chat.
• Twins: It was so robot that it almost circled and reinvented robots. (That means it was bad.)

How much processing would I have to do to publish it?

• Chatgpt: Pretty light. I just have to remind you of what to do and to layer in some personal examples.
• Claude: Same answer as for chatt.
• Twins: It is not even worth being persecuted for me.

If I really dealt with a serious contribution, I would probably have written down notes or divided a voice recording or an absence script with my two cent between ideas and outline levels.

But without this context and only with what I fed with the tools about myself, Chatgpt and Claude both did a solid job.

Editing the contribution

Fairyly, Ai First Drafts almost never hit the nail on the head the first time. Even if an AI-generated blog post for structure or sound is exactly correct, human touch is required. I found that the difference between “Meh, that’s okay, I think” and “Cue the confetti, that’s great!”

Something you hear a lot is that mediocre is no longer good enough. It was never really, but many mediocre content passed the sniff test in front of the AI.

Nowadays your things have to be good – quality about quantity and all of this. (Ai, however, helps you create more content, faster if you use it correctly)

If I had worked on each of these posts, I would probably do a lot to become more specific, including:

• High level feedback.
• Take the section according to section.
• Checking the entire piece for consistency and clarity.
• Take a final correction spin.

In order to see how these tools reacted to the same input requests, I gave them the same input requests again. I focused on my personal brand red red flags. I shared partial lists in feedback for every tool and put them together here.

In this sense, my goal was not to clean up grammar. AI usually performs a decent task. I wanted to see which my changes would accept and run with them to ai-generated blog posts that like me sound-or at least a little more.

Here is the request:

“Please revise this draft to tighten the river, to improve the rhythm and sound more naturally. Priority of clarity and sound consistency. Do not add any fluff.

Avoid the following:

• The word tinker
• Starting with mid-converse phrases like “We have all seen”
• Language that sounds good but says nothing
• Analogies that are clever but empty, like the siren song
• Opener like look, I understand it
• Sentences as every time someone raved about
• Strange analogies like a committee of robots who try to pass the Turing test
• The question> softening> response structure
• Parallel constructions like not just X. Do Y
• Three short sentences that start the same way
• Fluffy generalizations – make it specific and grounded. “

Before and after screenshots that help you to see the respective changes.

How Chatgpt was executed

It is clear to me that these screenshots are quite small, but to recognize the differences in the set structure, I think it is helpful to see the side by side.

Here is the screenshot of the after version:

I would give chatt aa to put on and remove the constructions that I don’t like. Sounds like me. It is also tight and to the point … and it could make a great LinkedIn carousel or a contribution.

How Claude fell off

I have the chills! Claude was able to take my feedback and do great work to implement it.

Here is a larger version so that you can read word for word:

Does it cheat if I say #obses? Because Claude really nailed it. Is it perfect? No, but is it very close and sounds a lot more after me? Yes.

How Gemini worked

Here is the side-by-side view:

And the individual view:

I admit that after the fluff from the last round, I didn’t have any great hopes here.

And I was pleasantly surprised. Does it sound like me? Not quite, but Gemini Is come closer. It is even more important that the feedback I existed and ran. Needs even more substance. But a huge part of the river I hated is gone.

Based on this, Gemini is running again as a tool that I would like to play with a little more. I think we could probably get there.

Overall best blog post generator?

Oh boy. That is difficult.

I am still a head-over heels for chatt. It is my point of contact for everything, from child-friendly recipes to the planning of my garden (zone 5b, in the house!) And of course AI-capable workflows.

However, I don’t think Chatgpt played the best in the restrictions of this test. I wonder if we are too close. Chatgpt knows which direction I will probably go in and it often goes there and focuses on short and puddish about specific and descriptive.

Can Claude “learn me” too? Perhaps. Will Claude win every test? Of course not.

But Claude did the best when it came to the limited information I provided:

• Develop creative blog post ideas.
• Development of a solid, descriptive (also known as non -generic) outline.
• Execute a well -structured and specific article.
• Processing on My Content of the heart.

I also know that if I did more editing rounds, we could absolutely arrive. And while Gemini did not win this round, I would not count it from my next round of the AI-generated blog posts or tests. I think it has solid potential!

Where these tools cannot replace people

Simply put, by nature, AI cannot do what people are universal – people are.

However, AI is a phenomenal assistant and becomes more impressive every day. The biggest disadvantage (which also takes good over the world in a “Will robot?” Is that it cannot think or think for myself.

I was able to use AI platforms to sting holes in my ideas, but as inherent human lovers, they have to find that they have to request specifically constructive criticism.

Otherwise, they are great to reflect their cadence and emulate their structure. You just miss the you factor of your experience and perspective.

The true magic is in the middle

I went into this thinking that Chatgpt would be the winner. After all, it is the tool that I use every day that has already seen how I write what I think and how I work.

But Claude surprised me best. It took the direction, offered creative ideas that I hadn’t considered, and made changes that actually sounded like me. Gemini even redeemed himself in the last round, which also makes me curious to play more with it.

At the end of the day I believe that the best tool for blog posts for A-generated blog posts is not the one with the most functions. It is the one who helps you to write faster, think better and sound more like You. And that means that it can change from time to time or even from article to article.

And if this article has convinced you to finally give a real shot – or to refine how you are already using it – then great. Because it is not about replacing yourself. It’s about expanding your voice.

Note from the publisher: This post was originally published in October 2023 and updated for completeness.