Same prompt, inconsistent results on GPT4

Prompting
, ,
1

Hello everyone,

I have been trying to setup GPT4 as a copywriter following very strict guidelines.

These guidelines include:
Persona: How I want ChatGPT to work for me (i.e: A seasoned copywriter)
Purpose of the text (i.e: informative article)
Tone of voice: SEO Optimized
Mininum size: Using 2500 chars.
Text structure: Title, Outline, Text, Conslusion.
Tag usage: H1, H2, and so on.
Complying to Googles EEAT Rules
Narrative: 3rd person
Included on text: External links referencing directly the subjetc (here is where the main problem lies)

Subject information:
Client’s website: URL
Main Keyword (the subject of the article)
Target demographic: age, location, etc.
Pain points: X, Y, Z.
Article Objective: Funnel Stage

The issues I have been confronting are:

  • ChatGP* T not always include external links to articles merged with the text
  • ChatGPT sometimes include “quotes” but says it cannot access the source or it just quotes on the text.
  • ChatGPT sometimes simply ignores the request for adding external links in the context of the article
  • When it adds links to the text they are ALL 404s.
  • It sometimes produces text that are NOT informative at all (Quality issues)
  • Content produced sometimes is very shallow

I have been trying to design prompts that balance requirements and information as balanced as possible, but using the same prompt over and over is turning into VERY inconsistent results.

ALSO: Does someone knows if prompting via API would turn more consistent results than over Chat?

1 Like
2

Yes, prompting via API’s gives you more control. Importantly, you should try setting Temperature to 0 and trying this out via API’s.