How I Used Artificial Intelligence in Fundraising
Some time ago, I was collecting information on how AI can be applied in foreign economic activity. And I found several articles and even monographs. They all had roughly the same thoughts: in theory, artificial intelligence will take care of routine tasks, speed up work, simplify interaction, and so on. But at the same time, not a word about practice.
The same picture is seen in other areas.
![there are very few examples from practice so far there are very few examples from practice so far](https://habrastorage.org/getpro/habr/upload_files/a6a/940/15e/a6a94015e08488d1ea8579781efb9dd7.gif)
That's why I decided to share my own, albeit small and rather specific, but real experience of how I tried to use AI in fundraising. I'll tell you what worked well and what didn't work at all. I won't name the fund, but it's quite large. You've probably heard about it, and maybe even seen an ad.
I used ChatGPT, Claude and Kandinsky. These are the ones I will talk about.
First – Creating Text Content
For successful fundraising, there must be a lot of content. Social networks, media, the foundation's own website. And this content, on the one hand, is very monotonous, and on the other, it must be unique. Therefore, constantly rewriting and “packaging” the child's story for different formats is a large, complex and tedious part of the fundraiser's work.
This seems like the perfect field for ChatGPT!
AI is really good at shortening or increasing the text volume (actually, “pour water”). But it definitely can't increase the uniqueness. In addition, the quality is quite low. Repetitions, speech and even grammatical errors occur. And they need to be “caught”.
![you couldn't have said it better you couldn't have said it better](https://habrastorage.org/getpro/habr/upload_files/d1c/b81/af8/d1cb81af8f00032498020d4c3d4c6be9.png)
There are also funny moments when you ask the model to change the text, but the output is absolutely identical. Also, the AI corrects errors that were not there, which it later honestly admits.
But when it comes to writing from scratch, both ChatGPT and Claude are much better. For example, you need a short and concise text about some rare disease with a description of symptoms, prognosis, and treatment methods. We write a prompt, wait a couple of seconds, and get a structured (even excessively) informative article. Of course, it also needs to be checked. Because the model can lie very plausibly. Fortunately, knowledge accumulates with experience, so incorrect information is immediately visible.
Second – Editing and Proofreading
You can ask AI to “thicken the colors” and “amplify emotions” in the finished text. Yes, fundraising is like that, you need to evoke emotions, without them there are no donations. And here artificial intelligence can add drama. So if your presentations or reports lack something like this – use it.
![even in Tears in Rain you can add emotions, what can we say about my lyrics even in Tears in Rain you can add emotions, what can we say about my lyrics](https://habrastorage.org/getpro/habr/upload_files/6c3/b5b/d98/6c3b5bd98aa3d9494cc0c03c0af78afe.png)
Subjectively, as a proofreader, ChatGPT works better than services like text.ru or advego.com and the like. But if the article is large, it is not immediately clear what exactly the assistant has corrected. Therefore, it is better to compose the prompt as follows: “Correct and show what you have corrected.”
Third – Working with images
Almost any fundraising text needs its own creative. Usually these are photos of children with some slogan. I believe that it is better to entrust any task to a professional. That is why I always left the design work to the designer. Moreover, I did not even give a technical task – “you are a professional, you have experience, do as you feel.”
This approach has many disadvantages, because even the greatest professional without some, even illusory, control does the job a little carelessly. But the relationship with the colleague was always good.
![something clearly went wrong something clearly went wrong](https://habrastorage.org/getpro/habr/upload_files/ae8/ea8/20d/ae8ea820da2b55b205067ee864ba9662.jpeg)
![try to find a seal try to find a seal](https://habrastorage.org/getpro/habr/upload_files/620/094/4f0/6200944f0e4d33fe1c7c25386dbbad50.png)
This approach didn't work with AI at all. You can't do without a clear technical task. However, even with it, Kandinsky drew something incomprehensible to me. It was completely impossible to use the resulting creatives in any form. Even for posts on Zen. So this is definitely a miss.
There may be neural networks that handle visual content generation better than Kandinsky. But in this article I am talking exclusively about my experience, and that is what I have.
Fourth – Analysis of advertising campaigns
I tried uploading reports sent by internet marketers to ChatGPT. The neural network identified leaders and outsiders in CPM, CTR, CR. Actually, this could have been done without AI.
![good advice that no one will follow good advice that no one will follow](https://habrastorage.org/getpro/habr/upload_files/422/d54/0c2/422d540c214cee395ac61074b65645bb.png)
If you ask the model how to improve your ad campaigns, you'll get generally useful, but very obvious and general recommendations like “A/B testing can help you determine which ads attract more clicks” and “add a call to action to your ad copy.”
On the other hand, if you really make an effort and follow all the recommendations, the effectiveness of the advertising campaign will definitely increase.
Fifth – Correspondence
I communicate with media representatives from almost all over Russia. These are huge databases of contacts by region in a regular Excel file. I often do mass mailings from my work email. For this, I need a list of addresses in a suitable format. And here ChatGPT is very helpful.
![Thanks, ChatGPT! Thanks, ChatGPT!](https://habrastorage.org/getpro/habr/upload_files/486/b5b/22f/486b5b22fe38bb5f4f5b45671ecc8132.gif)
I throw in a copied column of addresses and ask to give me a comma-separated list with a space. In this form, addresses can be added to a hidden copy. This is really very convenient. I used to use autocorrect in Word, but AI does a much better job.
Conclusion
ChatGPT and other AIs really help me in my work. But the quality of texts and editing is still very low. I have to double-check and rewrite a lot.
Now I'm trying to learn Python in parallel and here, it seems to me, the same ChatGPT will be a great helper. Explain incomprehensible terms, give examples, find an error in the code (and write your code better than you do) – here the model copes brilliantly, without exaggeration.
P.S. When preparing the post, I, of course, asked ChatGPT how AI can help in fundraising. We disagreed on almost all points. The model offered approximately the same thing as the authors of scientific articles on the use of artificial intelligence in foreign economic activity – analyzes, performs routine tasks, and so on. In theory, all this is really possible, but in practice, such a “breakthrough” is still very far away.
So copywriters and fundraisers can sleep peacefully for now.