After my first iteration the Googlesheet add row, or the Excel add row fails with no error. But the Make module cell placeholder is not dark, it is outlined and very light in color. When it is like that it does not update the sheet. The json parse module has the data and it is good.
If I rerun, the first one is good. Then it stops adding the row.
Hi @Rich_Buckley and welcome to the Make Community!
Usually the outlined module is an indication that the value passed is no longer available. This happens sometimes when you clone modules, make copies of modules, link/unlink modules, etc. It can also happen if you add a new module between to modules that were already configured to work together. If the âin-betweenâ module consumes the same values as (in your case) the Excel module, the data may no longer be available.
Try reinserting the values in your module configuration. You might have to delete and recreate your module.
If none of that works, please share a screenshot of your modules or even, possibly, your blueprint.
L
Hi @Rich_Buckley
If an aggregator module is placed between these modules, you wonât be able to map variables from any modules that come after the source module specified in the aggregator.
Best regards,
Msquare Automation
Platinum Partner of Make
@Msquare_Automation
Hello @Rich_Buckley,
@L_Duperval & @Msquare_Automation have already given specific answers to this topic.
My replay is just about the suggestion to use the proper category for your post. It will help other users to properly identify content.
Youâve used
Showcase
Show off the cool stuff youâve built with Make - hacks, best practices & more.
This needs to go to
How To
Ask questions related to building, expanding, and troubleshooting your scenarios.
P.S.: Always search first, Check Make Academy. If this is helpful, mark it as a solution 
and 
Need expert help or have questions? Contact or comment below! 
I recreated the scenario from scratch. Same problem. It looks like the blueprint saves personal info? I am not sure.
this is what I have. it doesnât even run once. If this means the data is no longer available when it gets to excel, it has to be the ChatGPT?
This is my prompt:
Please process this message {{1.body.content}}. Find the following relevant information:
Date/Time
Invoice Description
invoice Amount
Invoice Payment Method
Customer Id
Lastname
please create a random 4 digit number and call it theID.
Please output as json. Do not include any pretext and do not include ```json.
I ended up changing the scenario to this one. It seems to work every time. The prompt is this: Please process this message {{1.body.content}}. Find the following relevant information from the keys below:
âDate/Timeâ
âInvoice Descriptionâ
âInvoice Amountâ
âInvoice Payment Methodâ
âCompanyâ
âLastnameâ
Please replace commas and replace with space . Please remove text (USD).
Please create a random 4 digit number and call it theID. Just output the relevant information. Please do not include âââ .
Please output as text and place a comma after each of the items above.
Hello @Rich_Buckley,
In the prompt youâve given what output you needed. Itâs only JSON without ```json. Thatâs OK
But the important point is that youâve asked multiple field values to return inside that JSON. Youâve also given the list of fields. But as per JSON syntax, AI changes the field names according to JSON output, it also changes the given field names to specific syntax-related field names.
Because JSON is a set of Key: Value. The most common use case the key must be used without space.
As per the current setup, the AI output is within JSON but there are no rules for field(key) names.
In your prompt also predefine keys for all the fields then tell them that the output key is always these particular keys and then these values are associated with it.
In your prompt also include that output also matches the naming of keys based on value.
{"dateTime": {{Date/Time}}, "invoiceDescription":{{Invoice Description}}, "invoiceAmount:"invoice Amount"}, {"other fields"}
Using this approach your keys always come identical from the AI and you can use it on the next module.
I hope you can understand the things that Iâve explained to you.
P.S.: Always search first, Check Make Academy. If this is helpful, mark it as a solution and
Need expert help or have questions? Contact or comment below!
Aha! Thatâs the problem, Iâm sure. You can check @Rich_Buckley by looking at the history of your runs.
Click on one of the runs that didnât give the results you wanted and look at the JSON output bundle from ChatGPT. Most likely, that run will not have the right key names.
This is one of the key parts of prompt engineering. If you want a specific output, you need to tell ChatGPT (or any other tool) to adhere to a specific structure and it will always give you the correct results.
Test out your prompt in ChatGPT (or the playground) until it always gives the expected result. Then, you can use it in your scenario.
L
