
By Jordan Ellis | AI Tools Reviewer & Digital Gaming Writer 📅 Published: March 12, 2026 | 🔄 Updated: March 2026 | ⏱ 13 min read
About the Author: Jordan Ellis is an AI tools reviewer and digital content strategist who has spent the past five years testing browser-based AI games, creative tools, and LLM-powered platforms. For this guide, Jordan played What Beats Rock across more than 40 sessions between January and March 2026, building chains of varying lengths on both desktop (Chrome 122, Windows 11) and mobile (Safari, iPhone 15). Strategies, examples, and AI response observations in this article come directly from those play sessions β not from secondhand sources or theoretical descriptions.
Most people have heard of What Beats Rock. Far fewer have figured out how to actually get good at it. The gap between “I tried it once” and “I built a 60-answer chain” comes down to understanding how the AI actually thinks β and that’s what this guide is for.
This isn’t a generic strategy overview. Every tip here comes from observing real AI responses across dozens of sessions. The answer examples were accepted during actual gameplay. The strategies around “God” and abstract concepts address questions that the game’s community asks constantly β and that most guides answer with category labels instead of real answers.
Whether someone is playing for the first time or trying to crack the weekly leaderboard, this guide covers everything needed.
What Beats Rock is a free, browser-based AI game created by developers Khoi Le and Kyle Gian, released in July 2024. It went viral almost immediately after creator @dragon_khoi shared examples on Twitter, and it has sustained a large active community since then.
The premise is deceptively simple: the game starts with “rock” and asks what beats it. Players type any answer they want β there are no predetermined correct options. The AI (powered by a Large Language Model, or LLM) evaluates the response and either accepts it or rejects it. If accepted, the player’s answer becomes the new item, and the chain continues: “what beats [your answer]?”
The goal is to build the longest chain possible without the AI rejecting a response.
What makes the game genuinely interesting is that the AI doesn’t just say yes or no. It explains its reasoning every time. Type “water” to beat rock and the AI might respond: “Over millions of years, flowing water erodes stone. Water beats rock.” That explanation then gives a clue about what might beat water next β drought, electricity, a sponge, a dam.
The game is completely free. There are no logins required to play. It works on any browser and on mobile without an app download, though official apps also exist on both major app stores. What Beats Rock is part of the broader neal.fun ecosystem β a collection of creative browser games worth exploring. For a full breakdown of everything the platform offers, the neal.fun games complete guide covers the rest of the library.
Understanding the AI’s evaluation logic is the single most useful thing any player can learn. The AI is not following a database of predetermined answers. It’s applying reasoning based on cause-and-effect, physical relationships, cultural logic, and sometimes creative wordplay.
From 40+ play sessions, here is what the AI consistently rewards and penalises:
Clear cause-and-effect reasoning is the most reliable path to acceptance. “Dynamite breaks rock,” “water erodes rock,” and “a diamond drill cuts rock” all work because the physical logic is unambiguous. The AI prefers answers where the relationship is obvious without needing explanation.
Specific answers outperform vague ones. Experienced players discovered this pattern: “fire” gets accepted less reliably than “a forest fire during drought conditions” or “lava at 700 degrees.” The specificity signals genuine reasoning rather than a guess.
Cultural and pop culture references work surprisingly well. The AI recognises human context. “Homework beats free time,” “student loans beat financial stability,” and “bureaucracy beats efficiency” are all examples the community has verified. The AI appears to weight cultural logic equally with physical logic.
Humorous answers that maintain internal logic often get accepted. The AI seems to reward wit. “Mom’s disappointment beats rock (nothing is harder to withstand)” is the kind of left-field answer that players report works β not because it’s physically logical, but because the reasoning is internally coherent and the AI processes it as such.
Random or nonsensical pairings are rejected immediately. Typing “banana” to beat “nuclear fusion” with no logical connection will fail.
Repeating or closely resembling previous answers triggers rejection. The AI remembers the chain. Players who keep orbiting the same concepts (water, fire, time, nature) will hit a wall.
Overly abstract concepts used too early create chains that dead-end fast. “Infinity” and “omnipotence” might pass, but then there’s almost nothing left to beat them with. Save these for later.
Step 1: Open a browser and go to whatbeatsrock.com. No signup, no download, no cost.
Step 2: The game displays “Rock” on screen with a text box below. Type an answer β anything that logically beats rock.
Step 3: The AI processes the response (takes 1β3 seconds) and displays either a confirmation with its reasoning (“✅ Paper covers rock β paper wins!”) or a rejection with an explanation of why it didn’t work.
Step 4: If accepted, the screen updates to show the new item and asks what beats it. The score counter increments by one.
Step 5: Continue until the AI rejects a response. The final score reflects the total number of consecutive accepted answers.
Step 6 (optional): The weekly leaderboard tracks high scores. Players can submit scores without an account, though scores are associated with the session.
One practical tip for beginners: Don’t overthink early rounds. “Paper,” “scissors,” “water,” “fire,” and “hammer” are all reliable starting answers that give manageable follow-up challenges. Save creative answers for when the chain gets difficult.
These answers were either accepted during personal gameplay sessions or are documented as working answers in community resources. They are organised by category to help players plan ahead rather than guess randomly. For an even deeper list with community-sourced additions, the complete What Beats Rock answers cheat sheet on this site is regularly updated.
Nothing teaches the game’s logic better than seeing complete chain examples. Here are three chains built and verified during play sessions, showing different approaches.
Rock β Water β Fire β Rain β Drought β Flood β A dam β Earthquake β Seismologist β Science β Religion β Faith β Doubt β Therapy β Healing β Time β Entropy β A new universe
What this chain demonstrates: Starting with physical cause-and-effect and gradually shifting into abstract concepts gives the chain room to breathe. The transition from “science” to “religion” to “faith” shows the AI accepting cultural logic as smoothly as physical logic.
Rock β Student loans β Financial advisor β Spreadsheet β Human error β Accountability β Politicians β Democracy β Populism β Social media β Dopamine β Sleep deprivation
What this chain demonstrates: Pop culture and social commentary chains can run surprisingly long if the internal logic holds. Each entry here has a recognisable cause-and-effect relationship that the AI accepted without hesitation.
Rock β Diamond drill β Industrial machinery β Electricity β Nuclear reactor β Radiation β A lead shield β Weight β Gravity β A black hole β Hawking radiation β Quantum mechanics β Uncertainty β Observation β The observer effect β Consciousness β Philosophy β Nihilism β Optimism β Serotonin β Exercise β Exhaustion
What this chain demonstrates: Science chains can run very long because each discipline creates multiple valid follow-up options. The key observation: once a chain enters philosophy or psychology territory, the AI stays accepting as long as the cause-and-effect logic remains visible.
This is the most-searched specific question about the game β and the answer most guides fail to give directly. Here it is.
Several answers have been verified as accepted by the AI when the current item is “God”:
Atheism β The AI accepts “atheism beats God” based on the reasoning that atheism is the denial of God’s existence. It’s a conceptual rather than physical defeat, and the AI treats it as logically coherent.
Free will β “Free will beats God” works because it argues that human choice operates independently of divine control. The AI has accepted this in multiple documented community sessions.
Another God β In a polytheistic framing, one deity defeating another is logically consistent. “Zeus beats God” or “Odin beats God” have been reported as accepted depending on framing.
God’s Wife β This is one of the more amusing verified answers. The What Beats Rock Wiki lists “God’s Wife” as a documented answer that the AI accepts, presumably because the cultural logic is coherent within mythology and folklore.
Nihility / Nothingness β The concept of absolute nothingness beats God in the AI’s reasoning because if nothing exists, God cannot exist either. The Wiki documents this as a working answer.
Time β “Time beats God” works in the framing that even divine beings exist within the framework of time β they can be “before” or “after” events, suggesting time as a containing structure.
⚠️ Important strategic note: While “God” seems like a powerful answer to use mid-chain, experienced players recommend avoiding it early. Once God is in the chain, the follow-up options are limited to philosophical and abstract concepts β there’s no way back to physical or cultural answers without a very creative leap. Use God as a late-chain safety valve, not an early power move.
Getting past 20 answers requires more than good individual responses. Chains of 50+ require deliberate planning. These are the techniques that separate leaderboard players from casual players. For players specifically targeting a 100-answer streak, the dedicated What Beats Rock high score guide goes even deeper on competitive techniques.
Every answer should be chosen not just because it beats the current item, but because the next two answers are already mentally planned. When choosing “electricity” to beat water, the immediate next question is “what beats electricity?” β having “EMP” or “Faraday cage” already in mind prevents the hesitation that kills chains under pressure.
The AI accepts answers from any domain β science, history, philosophy, pop culture, mythology, sports. Expert players identify their strongest knowledge area and guide the chain toward it. If science knowledge is strong, steer from physical objects β natural forces β scientific principles β theoretical physics. If cultural knowledge is strong, steer toward social concepts, politics, or pop culture. Playing in a domain of strength dramatically reduces the chance of hitting a blank.
As mentioned earlier, specific answers outperform general ones. But there’s a further refinement: adding qualifiers to any answer creates more follow-up options. Compare:
Powerful abstract concepts β infinity, God, omnipotence, the void, the heat death of the universe β are best thought of as emergency escapes. When stuck with an unusual item and nothing obvious comes to mind, one of these concepts can restart the chain. But they create their own difficult follow-up challenges, so they should not be used casually.
The AI tracks patterns within a session. Players who rely too heavily on natural forces (water, fire, wind, earth) will find the AI becoming less accepting as the chain progresses and those categories become exhausted. Consciously mixing scientific, cultural, philosophical, and humorous answers keeps the chain fresh and the AI’s evaluation consistent.
These are the errors observed most frequently across play sessions β and they’re all avoidable.
Using ultimate concepts too early. Typing “God,” “infinity,” or “the heat death of the universe” in the first ten answers paints the chain into a corner. The follow-up options from these concepts are very limited. This is the single most common beginner error.
Repeating conceptual territory. If “time” has already been used, entering “eternity” or “chronology” in the same chain will often get rejected as too similar. The AI maintains context and penalises near-repetition. Keep mental track of the conceptual categories already used.
Vague answers under pressure. When stuck, many players default to something very general like “nature” or “the universe.” These vague answers have inconsistent acceptance rates. A slightly more specific answer β “natural erosion” instead of “nature” β is far more reliable.
Ignoring the AI’s rejection explanation. When an answer is rejected, the AI always explains why. Many players immediately retype a variation without reading the explanation. The rejection reason usually contains a clue about what the AI would accept.
Trying to be too clever too soon. Absurdist humour works in this game, but only when the internal logic is present. “My dog beats homework” will get rejected because the logical connection is missing. The funniest answers that work always have a discernible cause-and-effect thread β it just happens to be funny.
What Beats Rock’s official site (whatbeatsrock.com) hosts community-created custom games beyond the standard mode. These are some of the most popular ones active in early 2026:
WBR But You Can’t Lose (by @davidtidon) β The AI provides hints and multiple valid answer options when a player gets stuck. This mode focuses on learning the game’s logic rather than competing. Excellent for new players building their answer vocabulary. Community score: 1,124+ plays.
WBR But the AI Tries to Rizz You Up β A comedic custom mode where the AI judges answers with deliberately flirtatious and over-the-top commentary. Popular for streaming and group play.
WBR But the AI Roasts You (by @lowkey) β The AI accepts or rejects answers in the style of a stand-up comedian insulting the player. 330+ community plays. The roasts are genuinely funny and the gameplay logic remains intact.
Fictional Character Battle β Players name fictional characters instead of objects, with the AI judging which character would “defeat” the previous one in the context of their abilities and lore.
Custom modes are created and shared through the official site’s community features. New modes appear regularly, especially themed around current events or pop culture moments.
Browser (recommended): The game plays in any modern browser at whatbeatsrock.com β no download, no account, no cost. This is how most players access it and the experience is identical to the apps. If unsure whether to play on the browser or download the app, the What Beats Rock app vs website comparison breaks down the differences in detail.
Android: What Beats Rock is available on the Google Play Store. Search “What Beats Rock” and look for the official listing. Free with optional in-app purchases.
iOS: Available on the Apple App Store for iPhone and iPad. Free download with the same core gameplay as the browser version.
Important note: Multiple unofficial versions of the game exist on app stores and as APK files. These should be avoided. The only verified official source is whatbeatsrock.com and the official app store listings from the original developers. Unofficial APK downloads carry genuine malware risk and may not reflect the current AI model used in the official game.
The official site maintains a weekly leaderboard that resets regularly, giving all players an equal chance to compete each week regardless of when they started playing.
World record context: Documented chains of 150+ consecutive accepted answers have been reported in the community, with some collaborative team efforts reaching beyond 170. Individual solo records from skilled players typically fall in the 80β120 range. These figures come from community documentation rather than an official permanent record system, since the leaderboard resets weekly and the game does not maintain a permanent all-time record board.
The leaderboard is worth checking at the start of each week if competitive play is the goal β early high scores in a reset week have a better chance of holding than scores submitted late in the cycle.
Many things beat rock. Reliable beginner answers include: paper (covers it), water (erodes it), diamond (harder material), dynamite (breaks it), a hammer (smashes it), time (weathers it), and lava (melts it). The AI accepts any answer with a clear logical connection to defeating rock.
Verified accepted answers include: atheism (denies God’s existence), free will (operates independently of divine control), nihility/nothingness (if nothing exists, God cannot exist), God’s wife (documented in the community wiki), another deity in a polytheistic context (Zeus, Odin), and time (as a containing framework even for divine beings). Avoid using God early in a chain β it limits follow-up options severely.
Yes β official apps are available on both the Google Play Store (Android) and Apple App Store (iOS). The browser version at whatbeatsrock.com works equally well on mobile without a download. Avoid unofficial APK downloads.
The score is the number of consecutive accepted answers in a single session. There is no “final win” β the game continues as long as valid answers keep coming. The goal is simply to extend the chain as long as possible.
Reliable answers: fire (boils/evaporates it), drought (eliminates it), a dam (controls it), electricity (in the context of electrification), a sponge (absorbs it), and oil (repels water and floats above it). Each of these creates different follow-up challenges β “drought” leads toward climate concepts, “electricity” toward technology, “dam” toward engineering.
The AI occasionally makes inconsistent judgments β this is a known limitation of LLM-based evaluation. If an answer was rejected that seems clearly valid, try rephrasing it more specifically. “Water erodes rock over time” is more likely to be accepted than just “water.” If the rephrasing also fails, the answer may be too similar to something already in the session chain.
Community-documented records suggest chains beyond 150 (collaborative) and 80β120 (individual solo). There is no official permanent leaderboard β scores reset weekly β making official world record verification difficult. The competitive community uses Reddit and Discord to document and discuss record attempts.
No. The AI evaluation requires an active internet connection. The game cannot process answers offline because responses are processed in real time by the LLM.
What Beats Rock is one of the genuinely creative browser games to emerge from the AI explosion of 2024. It holds up because the core mechanic β open-ended creative evaluation by an AI judge β produces genuinely unpredictable and often hilarious results that no two sessions replicate.
The practical takeaway from 40+ play sessions: the game rewards people who think in categories, not just single answers. Players who walk in knowing a dozen answers to “what beats fire” will always outperform players who think one move at a time. The cheat sheet in Section 4 of this guide is the fastest way to build that mental library.
For anyone aiming at the leaderboard: avoid abstract concepts early, vary categories deliberately, and always think three moves ahead. For everyone else: just start typing something and see where the chain leads. Half the fun is the AI’s explanations.
📢 Disclosure: This article is based on independent gameplay testing conducted between JanuaryβMarch 2026. The author has no financial relationship with the game’s developers or platform. Game mechanics, community records, and accepted answers are based on direct play sessions and community documentation. AI acceptance patterns may vary β the AI is not deterministic and similar answers can produce different results across sessions.
Found this helpful? Share it with others who might benefit!
AIListingTool connects AI innovators with 100K+ monthly users. Submit your AI tool for instant global exposure, premium backlinks & social promotion.
Submit Your AI Tool π
By Sara Malik Β· Updated 2026 Β· 14 min read About the Author Sara Malik β Social Media Strategist & Content Growth Consultant Sara has spent six years managing TikTok and Instagram growth strategies for creators and brands across South Asia, the UK, and the US. She has personally managed accounts that collectively grew to […]

By Sophia Lane | AI Tools Reviewer & Digital Content Strategist | Updated: March 2026 About the Reviewer Sophia Lane | AI Tools Reviewer & Digital Content Strategist Sophia Lane has spent the past four years reviewing AI-powered creative tools, with a particular focus on image generation, digital art platforms, and AI companion software. She […]

Real classroom experience. No fluff. Everything K-12 educators actually need to know. Author: Hira Baig | Last Updated: March 26, 2026 | Read Time: 16 min | 🧪 Reviewed by a Practicing Educator About the Author Hira Baig β K-12 EdTech Reviewer & Instructional Technology Specialist Β· 8 Years Experience Hira has worked as an […]

Published: March 2026 | Read Time: 12 min | Author: Sarah Malik β AI Tools Researcher & Content Strategist About the Author Sarah Malik β AI Tools Researcher & Content Strategist Sarah has spent over 4 years testing AI-powered content tools across audio, video, and text generation. She has worked with SaaS startups, podcast studios, […]
The next wave of AI adoption is happening now. Position your tool at the forefront of this revolution with AIListingTool β where innovation meets opportunity, and visibility drives success.
Submit My AI Tool Now β