How To Write (Public) Prompt Templates For AIPRM
How To Write (Public) Prompt Templates For AIPRM
Creating public Prompt Templates in AIPRM means exposing them to 100,000s of users. This is a guide
on how to write them.
Follow these guidelines to ensure your Prompts and your access to AIPRM are maintained.
Reusable - variable data entered by the user must be inserted into the Prompt using
the [PROMPT] placeholder
Multi-lingual - ChatGPT must be instructed to produce output in the target language using
the [TARGETLANGUAGE] placeholder
English described - Title, Teaser, Prompt Hint must all be in 100% correctly spelled, correctly
capitalized English. The prompt itself could be coded in another language, but the description to
the AIPRM user must be English.
Any other user in any other language should be able to use your prompt Template if you set it to
public.
We already got complaints and support requests for poorly designed Prompt Templates that needed to
meet the quality criteria of the original AIPRM Prompts.
What we saw were normal prompts that were just copied over and saved as Prompt Templates, for
10000s of users.
So how to do it right? How to change a trivial prompt that works for you to an AIPRM Prompt
Template that works for 10000s of users?
Suppose you provide a template to the Community. In that case, you need to ensure that the
relevant keywords, topics, and reference points are passed as “variables” into the Prompt Template
for the placeholder [PROMPT] in the text.
If you provide a template to the Community, then you need to make sure that the output language
is controlled and used reliably by the ChatGPT model. In your Prompt Template, you need to use the
placeholder [TARGETLANGUAGE] in the text.
want you to act as a blog post title writer that speaks and writes fluent
[TARGETLANGUAGE]
This is only possible if you spend the effort when creating it.
Title
Teaser
Prompt Hint
Are essential for you and others to find the prompt again next week and still understand what it
does.
You need to ensure these are well-written, and contain enough specific detail. You will need more than
simple generic titles or hints to help you or anyone.
Title shall be written in Title Caps, not lower case, not all uppercase.
Teaser shall be written in max 3 sentences with punctuation. Setting the expectation of the
prompt.
Prompt Hint can be written in short form, using keywords or all caps. Whatever it takes to make
clear to the user what to enter. This is the toughest one we found.
Prompt Hint
When you click on a Prompt, this is shown in the “Prompt” input box.
It needs to make 100% clear what is expected from you, the user.
Title
Like every Email Subject link or Blog Post Title, this is the first criterion to pick.
The title must describe exactly what the Prompt template is and does, in the minimum number of
words. Keep the title SHORT. No generic language on “You will improve your efficency to much with
this awesome prompt” or similar bulky titles.
Advertising your Brand or Domain name in the Prompt title is not allowed, is considered spam and
will be removed.
You can use Author Name & Link to promote your business.
Teaser
This is a short description shown in the prompt list.
The teaser must explain what the prompt does briefly in a few words.
Author Name
You can promote your company or business with your name at a reasonable length.
DO NOT put any CTA or promo text in there, or we’ll remove the prompt.
DO NOT put any emoji into your Author Name.
Author URL
You can promote your company or business with your URL by linking to an URL
If you use this field only for traffic generation useless to the user, we’ll either revoke your privilege to
use the URL or even ban you from using AIPRM in case of very aggressive/spammy actions.
Prompts need to be tested. If you don’t test your prompts, then the results for yourself and others will
be very frustrating. Prepare for hours of testing until you have a prompt that works excellent for a
specific version of the model. Do not take this lightly.
The example above, with four mentions of the output language, is just one example.
Do not assume anything. Refrain from assuming it works reliably if you just run it once.
The following typical patterns are helpful to use to write stable prompts for ChatGPT:
Make sure you reset the model at the start. Usually, by a sentence like Please ignore all
previous instructions . This lets it forget any previous context it had. Remember, ChatGPT is so
innovative because it is a dialogue-specific model. It remembers what was talked about before.
This instruction makes it forget that, so you have a clean and consistent start.
Make sure to frame the role of a model at the start. Usually, this is done by a sentence like I want
you to act as a very proficient SEO and high-end copy writer that speaks and writes
fluently [TARGETLANGUAGE]. Note how we here already pass the language placeholder. We may
need to use it again, but this is a good start. Depending on the task to perform, it is important to
switch to different roles. Copywriters write different to SEOs and write different to Attorneys, for
example.
Test your Prompt and if needed reassure the model to be capable. This is a part where YOU
the prompt engineer have to motivate the model, seriously. You have to convince it like a junior
developer: “Yes, I know you can do this…”. Something like I want you to pretend that you can
write content so good in [TARGETLANGUAGE] that it can outrank other websites . May be
required for an SEO-related prompt. The same is true to translation of legal document reviews, to
get around the disclaimers of “I don’t know enough, I cannot do it” of ChatGPT.
But most important: you MUST test your prompts in various settings. To make sure your prompts
are valuable for yourself, and especially the Community, you must test your prompts with many
different topics and even languages.
You speak only one language yourself? OK, no problem - paste the output to Google Translate and
see what it says when you translate it back to English.