How to learn Spanish with AI and no tutor

Q: Isn't immersion — films, podcasts, reading — better than flashcards?

It is a far better source of material and a poor way to retain it: in natural input a new word recurs too rarely and too irregularly to stick. Do both — read, then pull ten words out into cards.

8/23/2026 5 min read

In Russian: Как учить английский с ИИ без репетитора

If you have tried learning Spanish with ChatGPT, this will sound familiar. The conversation flows, the explanations of ser versus estar are genuinely good, and every question about the subjunctive gets a patient answer. Two weeks later you still say estoy aburrido when you mean bored-as-a-person and freeze on por versus para.

That is not because the AI is a bad teacher. As an explaining machine it is excellent. The problem is that it does half the job while confidently looking like it does all of it.

What AI does better than a human tutor

Infinite patience. Asking a third time why it is me gusta and not yo gusto costs you nothing.

Conversation without embarrassment. The real barrier for adult learners is not grammar, it is the fear of making a fool of yourself in front of someone. A chat window removes it entirely.

Instant correction. You write a sentence, you get the analysis. With a weekly lesson the feedback loop is a week long.

Explanations shaped to you. "Explain the preterite versus the imperfect using film editing as an analogy" — a model will. A textbook will not.

Availability. Spanish practice at 6am on a Tuesday, in ten-minute pieces.

The one thing it does not do

AI is excellent at explaining material and does nothing at all about stopping you from forgetting it.

A chat is an event. It ends, the tab closes, and the ordinary forgetting curve takes over: without review, a week later maybe a fifth of a new word survives. The model does not know that you mixed up embarazada and "embarrassed" three days ago, and it has no intention of bringing it up today. It has neither a memory of you nor a schedule.

A human tutor does exactly this: returns to your mistakes, asks about old material, notices that you said más mejor again. That is the part you have to replace when there is no tutor — not the explanations.

What covers the other half

Spaced repetition covers it. You record what needs to stick, and the system shows it again right when you are about to forget it — not sooner, which wastes your time on what you already know, and not later, which means learning it from scratch.

The whole loop:

Study with the AI. Conversation, a text, a grammar point — whatever you like.
Harvest your own mistakes. Not "all the words from the lesson" — the specific places you tripped.
Turn mistakes into cards. One cause, one card.
Review on a schedule. Five to ten minutes a day; the system decides what comes up.

Steps three and four are the half a chatbot does not have.

Making the AI keep the record

The trick is to ask the model to teach and to keep a log.

Tell it plainly: run a lesson on this topic; record my mistakes by cause rather than by symptom; at the end, give me one block of what I need to drill. "By cause" carries the weight here. If you said estoy un profesor three times, that is not three mistakes — it is one: reaching for estar where Spanish wants ser for an identity. One cause gives one card, not three.

In RepeatMe this loop is already assembled: a deck can carry a lesson, the lesson page hands you a ready tutor prompt for your own chat, and the cards come back into a deck as a single paste. Details in AI in RepeatMe.

Spanish-specific things worth putting on cards

Spanish rewards flashcards more than most languages, because so much of it is systematic:

Gender with the article, never the bare noun. Learn el problema, not problema — the article is the information.
False cognates. Embarazada is pregnant, éxito is success, sensible is sensitive, actualmente is currently. English gives you thousands of free words in Spanish and about two hundred traps; the traps are worth cards, the free words are not.
Verbs in a sentence, not in a table. A conjugation table is recognisable and unusable. A gapped sentence — Ayer ___ (ir) al mercado — is what you actually need.
Preposition-verb pairs. Soñar con, pensar en, contar con. These have no logic to derive; they are pure memory, which is exactly what cards are for.

If you want the field-by-field reference, see the card authoring guide.

How much time this takes

Realistically: thirty to forty minutes with the AI two or three times a week, plus five to ten minutes of review daily. The review is not another study session — it fits in a queue at the supermarket.

The trap almost everyone falls into is adding fifty new words at once. A week later there are two hundred overdue cards, you open the app, see the number and close it. Ten to fifteen new cards a day is the ceiling before the avalanche starts.

Start today

Not by designing the perfect plan. Take a ready-made deck on something you need right now and do one session — it takes two minutes and needs no account.

FAQ

Can I get to conversational Spanish with no human at all?

To a confident B2, yes, provided you have both halves — the explaining and the retaining. A human stays irreplaceable for one thing: hearing your accent and noticing what you did not think to ask.

Which AI model is best for Spanish?

Any major model handles explanation and conversation well past the level you need. The gap between models is far smaller than the gap between "studied and reviewed" and "studied".

Isn't immersion — films, podcasts, reading — better than flashcards?

It is a far better source of material and a poor way to retain it: in natural input a new word recurs too rarely and too irregularly to stick. Do both — read, then pull ten words out into cards.

How many words do I need?

Around 3,000 of the most frequent words cover roughly 95% of everyday speech. At fifteen a day that is a little over a year.