The AI Word Game Index

77 games tracked and scored on one question: what does the AI actually do while you play? Verified August 13, 2026.

Quick answer: Of 59 live word games we track, only 32 run a model while you play. Of those, 25 just score your guess by meaning, 9 produce language you actually read, and 3 change anything based on how you specifically played. Six more market AI that never runs at all. Every score below comes from a published six-criterion rubric, and the raw dataset is downloadable so you can recompute it.

We own one of these games. WordGameAI is operated by MidnightDev, which makes Yapword. Yapword is scored here by the same rubric as everything else. It ties for the top score and does not win outright, and it loses to Converge once the model-disclosure point is removed. Our other pages rank games editorially in Yapoleon's voice; this page is the data those pages are built on, which is why it reads like a spreadsheet instead of an emperor.
59live word games tracked
32run AI while you play
9have AI that writes back to you
17of those 32, name their model
6market AI that never runs while you play

The category is thinner than it looks

The first gate is whether a model runs at all while you play — roughly half the field clears it. The indented rows count what those games do with it. They are not nested: judging meaning and writing language overlap by 2 games and together account for all 32.

Live word games tracked59
AI runs while you play32
— judges meaning25
— writes language to you9
— adapts to you3

Live word games only (59 of 77 tracked records). Excludes 5 adjacent language games, 5 non-word games in the Wordle format, 2 puzzle-maker tools, and 6 dead or unreachable entries. Each criterion is verified per game; unverified counts as absent.

How the score works

The AI Involvement Score runs 0–10 and measures how much the AI does — not how good the game is. Wordle scores 0 and is still an excellent daily word game.

Anything unverified counts as absent, which pushes scores down for games that publish little about themselves. That is deliberate. Post-game AI — such as the NYT Connections Bot — is tracked but scores zero, because it runs after you finish.

Leaderboard

RankGameScoreExcl. disclosure
1=Converge88
1=Word.Studio Games *87
1=Yapword87
4=Linxicon76
4=Semantle76
4=Semantle (legacy build)76
4=WordRival: Guess The Word77
4=סמנטעל (Semantle Hebrew)76

* Word.Studio Games takes the disclosure point on a technicality we are not comfortable with: it names OpenAI, Anthropic and Grok site-wide but attributes no model to any individual game. Its record is flagged in the dataset. Without that point it scores 7.

Search the index

Filter by what the AI actually does. Every entry links to the game and carries the date we last confirmed it was live. 48 of them also have a full profile page with a criterion-by-criterion score breakdown.

Download the data

Every figure on this page is a function of one file. If you disagree with a score, open the record and tell us which field is wrong.

games.json games.csv

Free to reuse with attribution (CC BY 4.0). Please link this page so readers can see the verification date.

Want the one that talks back?

Yapword is one of the 9 games in this index where the AI produces language during play, and one of 3 that respond to how you personally played. Free daily, no account.

Play Yapword Download Yapword on the App Store

Frequently Asked Questions

What is the WordGameAI Index?
It is a tracked database of 77 word games, scored 0–10 on how much their AI actually does while you play. Unlike a listicle, every entry carries a source URL, a last-verified date and a published rubric, and the raw dataset is downloadable so anyone can recompute the scores.
How is the AI Involvement Score calculated?
Six criteria, worth ten points total: the AI runs while you play (3), it generates language you read (2), it evaluates your guess by meaning (2), it derives hints from the game state (1), it adapts to how you specifically played (1), and the operator publicly names the model it uses (1). Anything we could not verify counts as absent, which pushes scores down for games that publish little about themselves.
Does Yapword rank first in the Index?
No. Yapword ties for the top score of 8 with Converge and Word.Studio Games, and Converge leads outright once the model-disclosure point is removed. We make Yapword and we score it with the same rubric as everything else, which is the only way a ranking published by an owner is worth reading.
Why does the Index list games that have no AI at all?
Because the absence is the finding. 6 live entries market AI that never runs while you play, and 8 more use AI only to build the puzzle before you arrive. Listing them at a score of zero lets you check the claim instead of taking our word for it. Wordle also scores zero and is still an excellent daily word game.
How often is the Index verified?
Every record carries its own last-verified date; the current pass is August 13, 2026. Records older than 90 days are shown as stale rather than quietly presented as current, which is the standard failure of roundup articles in this category.
Can I use this data?
Yes. The dataset is published as games.json and games.csv, and the scoring rubric is public. If you cite a figure, please link the Index so readers can check the verification date. If you operate a game and a field is wrong, tell us and we will fix it and re-date the record.

Keep exploring

The Index is the data layer. For the ranked, opinionated version, see the best AI word games, which AI word games actually interact, and games like Wordle. For the semantic corner of the field specifically: games like Semantle and games like Contexto.