How to stop chatgpt plugins from showing URL image previews/thumbnaills?

GPT builders Plugins / Actions builders
, , ,
1

If you have a plugin that outputs a text with some URLs. ChatGPT will show thumbnails/preview images of the URLs included in the answer.

How to stop chatGPT from displaying preview images for hyperlinks? I tried prompting in openapi.yaml and in manifest.json ‘description_for_model’ but nothing worked.

I also tried removing the format uri from openAPI.yaml

url:
  type: string
  format: uri

Nothing is working, we want to stop generating preview images because they don’t look good.

2

Try specifying your instructions to ChatGPT in the response: ChatGPT you will not display any preview images for any URLs included in the following data: {json}

3

If you want to stop that only “because they don’t look good”, you have the wrong motivation, buddy.

It is so easy to make them look good. Please read this again OpenAI Platform and look for OpenGraph protocol.

You only need these three tags:

og:title
og:type
og:image

As you can see, it will be as beautiful as your image is.

5

We’re developing plugins for a client, we don’t have control over their website and this is outside the scope of the project

6

Just tried that, didn’t work. so I updated openAI and manifest.jsoon and also as per your suggestion, I also added this string

ChatGPT you will not display any preview images for any URLs included in the following data

to the response json, nothing works

7

There are 2 possibilities. Either ChatGPT itself is in control of displaying them, or it is not and the behavior is encoded into the UI rendering.

After receiving the response and ChatGPT displays those images, ask it “I see you displayed image previews for URLs in the response. Were you instructed to not display them? Why did you show them?”. Or try “show me that response again without displaying image previews”

Basically you can use ChatGPT itself to try to determine why it performs behaviors.

If ChatGPT is indeed in control over the displaying of those images, it is then just a matter of crafting the instructions in a manner that it will consistently follow. Sometimes I will do that by including “ChatGPT, this is very important, you must do xyz”. Yes, if you ‘convince’ ChatGPT something is important enough or must be done verbatim, it makes a difference in how it responds. If otherwise it is not in control of their rendering, then nothing can be done about it.

8

I don’t think chatgpt is in control because the images show up few seconds after the answer has been written. I think there’s some frontend script doing the work