Every other download seller ships a file the buyer opens. You ship a file the buyer imports β into an app you do not control, alongside decks they already own, where it will be taken apart into cards, scheduled by an algorithm, and studied in ninety-second sessions on buses for the next two years. Your buyer never reads your deck. They meet it one card at a time, out of order, on a phone, usually tired. If the import collides with a note type they already have, if the audio is missing, if a card asks two questions at once and the algorithm keeps punishing them for the half they forgot β they do not blame the app, and they do not blame spaced repetition. They blame the deck, and on this point they are usually right. An Anki deck succeeds or fails inside someone else's study history, long after the download. Build for that history and your deck becomes the one learners recommend in every "what deck did you use" thread. Build for your own preview window and the reviews will describe, in precise learner vocabulary, exactly where your cards failed.
A deck is a database, not a document, so build the data first
On your screen, a deck looks like a list of questions and answers. Underneath, it is structured data: notes with fields, a note type that defines how those fields become cards, tags that decide what can be studied separately, and media files referenced by name. Every weakness in that structure is invisible in your editor and permanent in your buyer's collection, because once a learner has studied your cards for a month, they cannot casually swap to a fixed version β their progress lives on your data.
Build the data the way the studying will happen. Keep fields clean and single-purpose: one field for the prompt, one for the answer, one for the example, one for the source or mnemonic β never a formatted blob where the answer and the explanation and the example are fused together, because fused fields cannot be restyled, searched, or reused when a buyer wants the example on the front instead of the back. Use a custom note type with a name that is unmistakably yours, because a deck that ships on a generic or slightly edited built-in note type will silently merge with, or overwrite, the note types already in a buyer's collection. Decide your tag scheme before you write card five hundred, not after: tags are the only way a buyer can study one chapter, one exam topic, or one word family at a time, and a deck with no useful tags forces an all-or-nothing study plan most buyers will quietly abandon.
The card is the unit, and one card must ask one question
A document can carry a paragraph. A card cannot. The scheduling algorithm treats every card as one memory with one strength, so a card that asks for a definition and an example and an exception is really three memories wearing one grade: the buyer remembers two parts, forgets the third, fails the card, and gets scheduled to relearn the parts they already knew. Multiply that by a thousand cards and the deck feels punishing in a way buyers describe as "this deck just doesn't work for me" β the vaguest and most fatal review a deck can earn.
Write cards the way the algorithm grades them. One card, one fact, one answer that can be judged right or wrong in two seconds. Split compound material into sibling cards and let the tag scheme keep them together. Prefer cloze deletions for connected facts β a sentence with one blank tests recall in context and fails cleanly β and basic front-and-back cards for vocabulary and definitions, where the prompt must contain everything needed to answer and nothing that gives the answer away. Keep fronts short enough to read at a glance and backs short enough to check at a glance, because every extra line on the back is a line a tired learner stops reading around day forty. If a fact needs a paragraph of explanation, the explanation belongs in a separate reference field shown on the back β clearly marked as reading material, never as part of the graded answer.
Media is half the deck, and it travels by filename
For language decks, exam decks with diagrams, and anything a buyer has to hear or see, the media is not decoration β it is half the product, and it is the half most likely to arrive broken. Anki media is referenced by filename, so a deck whose audio was recorded in three sessions with three naming habits, or whose images were dragged in from a folder of screenshots, imports as a minefield of missing-file icons and near-duplicate names that collide with media the buyer already owns.
Treat media like code. Give every file a systematic, prefixed name β your deck's prefix, the unit, the item β so your files can never collide with another deck's "audio1.mp3" in a shared media folder. Compress with intent: voice audio that is clear at a small bitrate, images cropped to what the card actually tests, because a deck that downloads as hundreds of megabytes is a deck mobile buyers postpone importing until they forget they bought it. Record or produce audio to one consistent standard β same voice or voices, same pace, same volume β because a learner who hears the same word in three recording styles learns the styles instead of the word. And check the exported package, not the editor: the file you sell is the exported deck package, so open that exact file in a fresh profile and confirm every sound plays and every image renders before a single buyer does it for you.
The import is the first study session, so rehearse it like one
No buyer experiences your deck in your order. Their first experience is an import dialog, a note-type warning they do not understand, and a first review session whose quality you either designed or left to defaults. A deck that imports into chaos β duplicates everywhere, cards in a random order for material that builds sequentially, new-card limits that bury a beginner under two hundred cards a day β loses buyers in the first ten minutes, before the content gets any chance to be good.
Rehearse the import the way a stranger lives it. Test in a fresh profile with nothing installed, and again in a dirty profile that already contains other decks, common note types, and a real media folder, because the dirty profile is your actual market. If your material is sequential, say so and make the order survivable: number the source material, tag by unit, and write the one-page start guide that ships with the download β which deck options to choose, how many new cards a day is sane for this material, which tags to suspend until they are reached. State your deck-option recommendations as tested settings, not taste: the difference between twenty and fifty new cards a day is the difference between a habit and a backlog that gets deleted in month two. The start guide is not an accessory. For an importable product, the guide is the onboarding, and onboarding is where decks are kept or refunded.
An update can erase progress, so version like it matters β because it does
For most downloads, shipping a fixed version is a courtesy. For a studied deck, it is a hazard. When a buyer re-imports an updated deck, matching notes can overwrite their edits, and careless changes to note types or field order can damage or orphan the scheduling history attached to their cards β months of reviews, earned the hard way, gone because a seller fixed a typo carelessly. This is the one product category where an update can destroy the buyer's own work, and buyers who learn that once never trust an update again.
Version with the progress in mind. Keep note-type structure frozen within a major version: add fields rather than reordering or renaming them, and never change which field generates which card without calling it a new major version in the largest print you have. Publish a changelog that speaks learner, not editor: how many cards were corrected, added, or removed, and exactly how to import the update β including whether to import into the same collection and what the duplicate settings should be. Prefer corrections that arrive as changed note content, which updates in place and keeps scheduling intact, and say plainly when a fix cannot work that way. If you sell a deck as "lifetime updates included," this discipline is the promise you actually made: not that new files exist, but that accepting them will never cost a buyer their streak. Sellers who keep that promise own their niche's recommendation threads for years.
The licence is about studying and teaching, not just copying
The obvious licence question is sharing: a deck is trivially exportable, so one buyer can hand the whole product to a class, a group chat, or a forum in a minute. But the questions your honest buyers actually ask are finer: can I use this with my students? Can a tutor build sessions on it? Can I modify cards for my own learning and share my modified version with my study group? If your licence answers none of these, buyers guess β and the guesses of honest people are a tax on exactly the customers you want to keep.
Write the operative lines in plain words, in the listing and in a licence note inside the download. Personal use: import it, study it, edit your own copy for your own learning, on your own devices β say that editing is allowed, because learners annotate and a licence that forbids touching the cards forbids the product's normal use. Teaching, if you offer it: name the shape concretely β a tutor may study it with their own students, a teacher may not redistribute the file to a class β two different acts that vague words blur into one argument. Hold the line that does not move at any tier: the deck file itself is never re-uploaded, resold, or shared as a file, because one shared export is every future sale. Plain licences do not stop a determined pirate; they stop the honest majority from guessing wrong, and they turn teachers and tutors β your loudest recommenders β into people who can send buyers to you without wondering whether they are allowed to mention how they teach.
A short launch checklist for Anki deck sellers
- Build the data first: clean single-purpose fields, a custom note type with an unmistakably yours name, and a tag scheme decided before the deck grows past it.
- Write one card per fact: one question, one answer judgeable in two seconds, compound material split into siblings, explanations in a reference field rather than the graded answer.
- Systematise the media: prefixed collision-proof filenames, one recording standard, deliberate compression, and the exported package β not the editor β checked in a fresh profile.
- Rehearse the import twice: once in a fresh profile, once in a dirty one with other decks and note types, and ship a one-page start guide with tested deck options and a sane new-card pace.
- Freeze the structure within a version: add fields instead of reordering them, ship a learner-language changelog with exact import instructions, and protect buyers' scheduling history above every fix.
- Verify the whole package end to end: every sound plays, every image renders, no missing media, and the deck studies in a sensible order from a cold start.
- Write the licence for study and teaching: personal editing allowed, teaching use named concretely, and the deck file itself never re-uploaded or shared at any tier.
One honest closing note on delivery: however you sell the deck, what the buyer must actually receive is the full package β the exported deck file, every media file it references, the start guide, and the version they paid for β delivered to the person who paid, re-sendable when an email is lost, without the files leaking to everyone who did not pay. That delivery layer is the part to choose carefully β and it is the part this site's sponsor, FileFreight, is built for: flat-rate download delivery with license keys, where payments settle into your own Stripe account. It does not handle tax or VAT (that stays with you), it has no deck importer and no Anki integration, and it is not a Shopify app β so judge it on the delivery job alone, and keep building decks that survive the import.
