AI Content Writer Showdown

AI Content Writer Showdown

It's been a minute since I last posted about what prompts and AI platforms I am using to generate content, so I thought I would share my latest process with you.

Key takeaways

  • custom prompt
  • multiple platforms
  • combine into human-edited article
  • fact check sources and URLs

Custom prompt

I have spent a lot of time writing and rewriting and customising a prompt that gives consistent results without sounding like it's been written by an AI.

Here is a redacted sample for you:

Redacted prompt


What I then do is post the same prompt into different LLMs and compare the output. I take the best answers from each and combine them into a human written article which I then fact-check before publishing on my website.

Lately I am using ChatGPT less and less for writing. As a LLM, it follows my instructions pretty well, but the fact that 3.5 is not connected to the internet means it can't give me references and this is a disadvantage. ChatGPT includes a lot of fluff in its content and includes words and phases that are a dead giveaway that it is an AI written article.

So, I most often am using Perplexity and Claude (accessed via Poe.com) to generate the initial draft of my article based on my prompt.

Perplexity often fails to follow my instructions, but its great at giving real URLs as sources, and it is possible to search for images and videos within the user interface. This is great when I want to check the source material is relevant. Many times Perplexity will link to an article about a study, instead of the paper itself, but I can usually prompt it to find the correct source for me.

However, the best tool for content is Claude (owned by Anthropic). It follows my instructions, gives URL sources, and writes in a more natural style without the fluff or including a biased opinion.

In terms of research, Microsoft Copilot (previously Bing Chat) is also pretty good, and its answers in browser links to its sources and feels very much like a featured snippet without pushing organic content too far down the page. The great thing is that I can also talk to Copilot while I am on any webpage, or access it on a new tab. While I detect more bias in Copilot than Perplexity or Claude, it does a pretty good job of summarising a topic.

How does Google Gemini (previously Bard) stack up?

Test #1

I asked a simple question: "Why don't airplanes fly over the Antarctic?"

Perplexity- straight and to the point ??

Claude- acceptable answer ??

Copilot - acceptable answer ??

Mentions de-icing which no other LLM did. ??

Gemini - nothing original ??

It also annoyingly gives its opinion at the start and end of its response which I did not ask for.

Overall, in terms of research and fact-checking I will avoid Gemini.

Test #2

An article prompt using my custom prompt, with the following topic:

"Write an article about the controversy around climate change"

  • Perplexity failed to follow my instructions ??
  • Claude wrote a decent length article from a first person POV ??
  • Copilot wrote a more formal and longer article. It also posted a point of view from climate denialists, which was very interesting and more balanced. Its result was actually pretty good??
  • Gemini failed 50% of my instructions, and instead of sticking to the topic added its own opinion about renewable energy ??

So, in conclusion, don't rely on only one AI to write your content, and be aware that they are programmed by teams who have their own biases.


+++++++++++++++++

Follow Janet Brandon on LinkedIn.

Struggling to create quality content that ranks on Google? Get your copy of Mastering ChatGPT: for Marketers, Content Writers and SEO Professionals and start dominating search engine results with ease!

Grab a copy of my new ebooks: Framework Secrets for Effective Copy, and ChatGPT for Business.

Author of GreenFingers.info

Hi there! I am a seasoned digital marketer with a wealth of experience across a range of marketing activities, from SEO, digital marketing, GPT prompting, social media to email automation and ecommerce, with a love for oil painting, reading and mountain biking.


Ronald Mutasa

AI for good |Project development lawyer specialising in Renewables (Solar, Wind) & PPPs (Roads, Ports)| Special interest in International Arbitration| ProBono Lawyer |GenAI & Legal Prompt Engineering Expert

9 个月

Fair assessment. Of late, I find myself actually using Gemini Advanced (2 month trial from google) more than ChatGPT 4.0. I like the fact that Gemini seems actually to be more of a co-pilot than Co-pilot. For instance, when attempting to draft an agreement, Gemini will ask questions and guve you suggestions etc whilst other LLMs simply dive into drafting. Chatgpt 4 does also prompt at the end (sometimes) but not in an interactive a manner as with Gemini. So my take is when you want a straightforward draft, no questions asked, just go with chatgpt or other LLMs but when you want something more interactive then Gemini seems to do the job better. *In drafting articles, I also like the idea of doing 3 separate drafts using the same prompt across chatgpt, claude and gemini. I then copy and paste the other 2 articles into chatgpt and ask it to combine the 3 articles into one coherent one, makig sure it captures all the points raised.

Exciting comparison! It's fascinating to see how different AI models stack up in article generation. Understanding their nuances can definitely help tailor content strategies effectively.

要查看或添加评论,请登录

社区洞察

其他会员也浏览了