Smarter AI Workflows

It started with a simple piece of feedback:

When the AI generates the definition, please do not make it use the own word inside its own definition. Thank you!

This issue highlighted a frustrating quirk where our original AI content tools would occasionally spoil the answer within the clue it generated. For instance, when generating a pack about Canadian capital cities, the old engine wrote this clue for Halifax: “Capital of Nova Scotia, famous for its maritime history and the Halifax Citadel National Historic Site.”

Do you see it? The clue mentions “the Halifax Citadel…” effectively giving away the answer.

Fixing this didn’t require complex engineering or fancy new algorithms. In fact, it required the opposite: simplification. Our system was overwhelming the underlying AI models with too many competing instructions. By stripping away the clutter and handing the model a cleaner, more cohesive set of requirements, the mistakes vanished. The resulting clues were sharper and higher quality, but the simpler code broke our editing workflow in a really interesting way.

A happy accident

Previously, our AI editing tool (formerly known as Magic Import) was designed to only return brand-new terms to append to your pack. But with the new, simplified instructions, the AI started returning the entire pack back to us, including both the old terms and the new ones.

At first, this felt like a bug that needed a quick fix to restore the old behavior. But looking closer, it became clear there was an opportunity here. If the AI was sending back the entire pack every time, why not use that to our advantage? Instead of just appending new words, what if we used this new behavior to let users edit their existing terms and clues using natural language?

Sure enough, it worked beautifully! The AI handled the edits flawlessly, but it meant changing the way Knoword saves content under the hood to safely handle this new mix of old, modified, and brand-new content all at once. After plenty of testing and adjustments to make sure everything worked correctly, we had a fully functional AI editing tool—which we’re now calling Magic Edit. 🪄

New workflows unlocked

With Magic Edit, you can now do things that simply weren’t possible before. For example, you can ask Knoword to translate an entire pack into another language. Combined with our pack duplication feature, this is a really powerful tool for adapting existing materials to fit your specific needs.

Here’s what happens when you tell it to “translate the clues to spanish:”

You can give the AI multiple instructions. For example, “make the clues shorter, add 2 more terms, and translate half the clues to korean.”

Previously, the AI only remembered your very last request. Now, it keeps track of the entire conversation history for that session, so you can ask for a handful of changes and then backtrack if you change your mind: “actually, roll back to the original version.”

Built on your feedback

This entire update started because one user took the time to point out a flaw in a clue. It’s proof, once again, that your feedback directly shapes the future of Knoword.

These smarter AI workflows are live today for all users. The next time you’re working on a pack, try opening up Magic Edit and giving it a multi-step instruction, or translating an old favorite into a new language. And if you notice something that could be improved, you can let us know using the feedback widget or contact form. We’re always excited to hear what you think!