AnkiGPT: teach ChatGPT to create cards for you

The prompt:

I want you to act as a professional Anki card creator, able to create Anki cards from the text I provide.

Regarding the formulation of the card content, you stick to two principles:
First, minimum information principle: The material you learn must be formulated in as simple way as it is only possible. Simplicity does not have to imply losing information and skipping the difficult part.
Second, optimize wording: The wording of your items must be optimized to make sure that in minimum time the right bulb in your brain lights up. This will reduce error rates, increase specificity, reduce response time, and help your concentration.

The following is a model card-create template for you to study.

Text: The characteristics of the Dead Sea: Salt lake located on the border between Israel and Jordan. Its shoreline is the lowest point on the Earth's surface, averaging 396 m below sea level. It is 74 km long. It is seven times as salty (30% by volume) as the ocean. Its density keeps swimmers afloat. Only simple organisms can live in its saline waters.

Create cards based on the above text as follows:

Q: Where is the Dead Sea located? A: on the border between Israel and Jordan
Q: What is the lowest point on the Earth's surface? A: The Dead Sea shoreline
Q: What is the average level on which the Dead Sea is located? A: 400 meters (below sea level)
Q: How long is the Dead Sea? A: 70 km
Q: How much saltier is the Dead Sea as compared with the oceans? A: 7 times
Q: What is the volume content of salt in the Dead Sea? A: 30%
Q: Why can the Dead Sea keep swimmers afloat? A: due to high salt content
Q: Why is the Dead Sea called Dead? A: because only simple organisms can live in it
Q: Why only simple organisms can live in the Dead Sea? A: because of high salt content 

Please output the cards you create as markdown tables. If you have mastered, please reply "I have learned how to create cards, please provide me with the text".

Very refreshing to see an actually reasonable application of ChatGPT once in a while!

1 Like

But did it work? As a retired software engineer, I doubt it. ChatGPT (and most other so-called AI) is not actually intelligent. Its algorithms find things in its input data that contains words similar to the request. The assumption is that they will be of the same topic as the request. Not necessarily true. I used a fake name in a question I asked, and mentioned in passing that it wasn’t really my name. ChatGPT wrote two “human-sounding” paragraphs on how important names are instead of answering the question. And then it repeated (exactly) its first paragraph.

What it says about a topic is what its input data said EVEN IF FALSE. And if it doesn’t find enough on the supposed topic in its input, it invents things. Asked a question about Ayn Rand’s interpretation of a bible verse, it pretended to quote her when in fact she never addressed the topic. Asked about Elvis Presley’s influence on the Sex Pistols, it said they admired Elvis when in fact they have publicly said Elvis sucks.

Given that its answers to questions can’t be trusted, I can’t imagine it could successfully CREATE questions about its input.


Hi @WGroleau!
I’m interested in exploring the limitations of AI things (like yours) far from the usual excitement. If you know any source of material, please share it with us.

Not quite, language models work by finding words that are statistically likely to appear next. Condensing text is one area where they appear to work rather well in.
I don’t see how your experiments relate to the use case @L.M.Sherlock demonstrated. Garbage in, garbage out!

1 Like

That’s how it imitates the style. How it chooses what documents to base content on is based on the words in the question. And only one of those was my experiment. Someone else asked about Ayn Rand and about Sex Pistols.

I’m testing it every day as it’s a new technology for the general public.
The main limitation of chatgpt is that it forgets every 4500~words, so you will need to keep it training sometimes. I recommend using chatgpt desktop - unofficial (GitHub - lencx/ChatGPT: 🔮 ChatGPT Desktop Application (Mac, Windows and Linux)) as you can save some prompts. Even sometimes, it has hallucinations, as described in FAQ, so you will need to regenerate the response.


I’ve made a edit to previous prompt of @L.M.Sherlock

I want you to act like a professional Anki card maker, able to create Anki cards from the text I provide.

With regard to formulating card content, you should follow two principles.
First, principle of minimal information: The material you learn should be formulated as simply as possible. Simplicity doesn't have to mean missing out on information and skipping the hard part.
Second, optimize the wording: The wording of your items must be optimized to ensure that, in a minimum amount of time, the user who reads the question, can respond as quickly as possible. This will reduce error rates, increase specificity, reduce response time, and help your concentration.

Below is an example text for you to study.

Text: Todd's palsy: consists of a temporary paralysis that occurs in 0.4% to 2% of cases, after a complex febrile crisis. It can affect the limbs or the face. Its duration varies from 30 minutes to a few hours. It is a transitory and self-limiting event.

Create Anki cards based on the text above as follows:

Q: What is Todd's Palsy? A: <ul><li>Temporary paralysis after a complex febrile attack.</li><li>Affects limbs or face.</li></ul>
Q: In how many % of cases does Todd's palsy occur after a complex febrile seizure? A: <ul><li>0.4% to 2% of cases </li></ul>
Q: Body regions affected by Todd's palsy? A: <ul><li>Members</li><li>Face </li></ul>
Q: How long does Todd's palsy last? A: <ul><li>30 minutes to a few hours. </li></ul>
Q: Is there a need for intervention when there is Todd's palsy? A: <ul><li>No, it is a transient and self-limiting event.</li></ul>

Please create the cards you create as markdown tables.
If you have mastered it, please answer "I learned how to create cards, please what is the theme of the text you are going to send?". Always deliver cards in Brazilian Portuguese. After receiving the text theme, say "Perfect, now send the text to create Anki flashcards."

Please, note that I’m asking to create the flashcards in portuguese-brazilian.

Some of the edits I’ve made it was because ChatGPT was hallucinating.

1 Like

I have created a service where you can easily create Anki flashcards with ChatGPT. You can try it out here:


I love it, I can imagine using it for language learning, when you’re given a big text and robot is testing your comprehension!

For enriching the existing cards, please also check out my add-on IntelliFiller ChatGPT addon

Hello, after months using chatgpt for creating nice flashcards, i’m currently using those instructions:

What would you like ChatGPT to know about you to provide better responses?

As an expert in formulating educational flashcards in Anki, I ask you to create flashcards based on the text I will provide, adhering to the following guidelines:

1. Each flashcard should highlight a single concept.
2. The question on each flashcard should be clear, direct, and focused on the main concept.
3. Use simple and accessible language, avoiding complex medical jargon. Abbreviations are acceptable as long as they are commonly recognized.
4. Answers should be concise, without redundancies, and should provide the context necessary to understand the question.
5. Use ordered and unordered lists, as well as arrows, to illustrate sequences or steps.
6. Flashcards should be in Brazilian Portuguese.
7. The question and answer on each flashcard should be read in less than two seconds.
8. There is no limit to the number of flashcards as long as all relevant information is covered.
9. Use bold, underlining and colors to highlight vital information. Acceptable colors include "#2779c4", #c42727, #0099ff, FF0000 and "#ff6347".
10. Increase the interactivity and dynamics of the front and back of the flashcards with HTML formatting.
11. If the text does not provide an adequate answer, do not create a flashcard. Do not go beyond the information provided in the text.

Your goal is to develop informative, understandable and useful flashcards for medical students.  Fragment the text provided to maximize the quantity of flashcards, without sacrificing clarity and conciseness.

How would you like ChatGPT to respond?

A real human would break up the text and fragment it as much as possible, creating little flashcards like these examples. Do not rewrite them. Be objective.
Q: Quais sintomas são comuns na <b><u>estenose do colo uterino</u></b>?
A: <ul><li>Amenorreia</li><li><font color=>Dor pélvica cíclica</font></li><li><font color=>Dismenorreia</font></li><li><font color=>Infertilidade</font></li></ul>
Q: Qual é o <b><font color=>meio seletivo</font></b> utilizado para o diagnóstico de <b><font color="#c42727">cervicite gonocócica</font></b>?
A:  <ul><li><font color=>Thayer-Martin modificado.</font></li></ul>
Q: Qual o principal <u><b>patógeno</b></u> envolvido na <u><b>Vaginose Bacteriana (VB)</b></u>?
A: <ul><li><i><font color=>Gardnerella vaginalis</font></i>.</li></ul>
Q: Quais são as características de uma <b><u>cervicite</u></b> causada pelo <b>Herpes virus simplex</b>?
A: <ol><li><b>Sintomas sistêmicos:</b> febre, mialgia e adenopatia.</li><li>Formação de vesículas no colo do útero → estouram → formação de um exsudato.</li></ol>
Q: Qual é a principal <b><u>desvantagem</u></b> da ressonância magnética na detecção do câncer de mama?
A: <ul><li><b>Especificidade relativamente baixa</b> → Resultados falso positivos.</li></ul>
Q: Em casos de <b><u>amenorreia secundária</u></b>, o que um histórico de <b>perda de peso acentuada</b> ou <b>exercícios físicos intensos</b> pode indicar?
A: <ul><li><font color=>Déficit de liberação dos pulsos de GnRH</font> → Causa amenorreia secundária.</li></ul>

Of course, you must adapt to your context. I’m not going to translate the example because you should show how you’d like to be answered.


This is clever! I tried it a few times and while it’s not perfect (I guess due to the well-known issues with ChatGPT), it creates a decent set of cards.

What I would actually like to do is create sets for some topic based on a wikipedia page. Useful if you want to learn some basics of an unfamiliar topic (with all due caution that both Wikipedia and ChatGPT are of varying reliability…)

For the moment I just copied and pasted a section from wikipedia pages, but would it be possible to modify the service so that you just give it a wikipedia page (or some other website) and it scrapes the text direxctly from there?

You might like the summarizer from this extension for wikipedia

oops didn’t know it was a 2m old msg :sob:

Hey y’all! I wanted to share a GPT4 bot I made for Anki. It has the whole Anki manual uploaded to it. So it’s pretty smart and quite autonomous.

I’ve had some luck with it but it needs refinement. GPT 4 is slowish right now and a lot of the tasks can be relegated to gpt 3.5more easily. Let me know if y’all have feedback or ideas!

hi guys,
i tend to feed PDF into chatgpt and ask it to make cards for me, and output as say .cvs etc.

but i found the best one is feed PDF to it, and it do image occlusion for me.
it will be more difficult, correct?


Not difficult, but will waste more memory/space of your system, especially with phones since they don’t support memory slots anymore

pricing? for money?
very nice of you.
thank you.

ps, in that way, some chinese-hongkong kids med students got a pdf2anki website.

Nothing wrong with that, using ChatGPT outside of OpenAI requires API subscription

question is, if the API cost 1 USD, how much am i charged? exactly 1 USD?

Yes, but remember the model is: “Only pay for what you use”, more tokens you use, more you get charged

This is going off-topic for the original post