Full blog post with OpenAI - is it possible with one pompt?


I’ve been testing OpenAI for some time and I was able to generate text with a Python program via the API without any problem.

I would like to do a new test, a little more advanced by generating a complete blog post including

  • an H1 title,
  • an introduction,
  • Some H2/H3 sub-headings that I will declare (sentence containing Why / How / For whom / Advantages / Our opinion…)
  • Several paragraphs under each subtitle.

Is it possible to do this with OpenAI?
I tried several times with this method in the editor:

"Generate a blog post in html with the following instructions:
- title h1
- introduction
- subtitle starting with Why
- paragrpahe
- paragraph continued
- subtitle beginning with How

But the result is not the best…

Should I generate several small texts via different queries on OPEN AI or is it possible to integrate everything in one Pompt?

If I have to create several small texts, I don’t know if the AI will optimize the text to not have too many repetitions of some keywords…

How would you do it in my place? :slight_smile:

Translated with DeepL Translate: The world's most accurate translator (free version)

1 Like

Only if you fine-tune to custom model.

1 Like

@jhsmith12345 which structure do you suggest?
A cell per each post (= copy paste of the whole post inside the file to upload for fine tuning), or should we split the posts into sections such as introduction, problem, etc…?

1 Like

What I would do… To fine tune, do not include any instructions. The prompt should include the info you want to provide about the article only, nothing more. Completion should be a perfect example of an article that you want to produce (based on the data you provided in the prompt).

Find or generate 200 to 400 ideal articles with their appropriate data points extracted.

Fine tune on that. Should work


So help me understand.
The structure may be something like:

prompt + completion

and I repeat for all of the 200-400 articles.

Is it?
I’m sorry, but I’m just a beginner learning how to do these basic things.


Yes, the prompt contains info on what is in the article, the completion is the article.

Giving details on what exactly you want it to do is only helpful when working with the instruct engine.

For fine tuning, keep it as simple as possible. Do not include any extra instructions in the prompt or the completion.


This is also what I did. Had to fine-tune to generate long articles. Basically, I created a well-formed Davinci prompt that would generate longer articles maybe 30% of the time. I ran it a bunch until I got a few hundred articles with my desired length and would save these to use in my fine-tune dataset.

Then when fine-tuning, my prompt was simply the article topic and the completion was the entire article. It works really well.


That’s interesting, you used prompt engineering to generate data which you then used as training examples.


This is called creating synthetic data. It’s a thing.


This got me confused. Shall I use the prompt in every row in the fine tune or not?


Fine-tune data does not need instructions. The prompt should only include the input data, and the completion the output data. If you show me an example of what you have been doing, all tell you if it is what I mean or not.



I thank you for your different interventions :slight_smile:
I must be an idiot but I have a hard time understanding when I don’t have an example in front of me.
Do you have an example of a project using fine-tune or the address of an explanatory Youtube video?

Thanks again. David

Hi. Did you use the entire article as a Completion? How did you structure your data?