Play Chess Against the Computer: Bots, Levels and Engines Explained
Playing against a computer is how most people in the UK get their first hundred games of chess. There is no queue, no rating on the line, no opponent watching the clock, and the machine will happily replay the same opening thirty times while you work out what you are doing wrong. This guide explains how chess engines actually work, what "level 4" or "level 10" really means, how to pick a strength that improves your play rather than crushing it, and where computer practice stops being useful.
On this page
Why practise against an engine at all
Computer opponents solve four problems that human opponents create. They are available instantly at three in the morning. They never resign, sulk or take twelve hours to reply. They can be set to an exact strength, so you can play someone slightly better than you every single game — which is the condition under which skill develops fastest. And crucially, they let you take a move back and try a different idea, turning a game into an experiment.
For a player who has just learned the rules, an engine set low is the ideal sparring partner: it punishes hanging pieces without being so sharp that every game ends by move fifteen. If you are still consolidating the basics, the guide on how to play chess covers movement, castling and checkmate patterns before you start counting engine levels.
Most modern boards run the engine in your browser rather than on a server, meaning no download and no installation. You can open a page and play chess against an adjustable bot in a few seconds, then step the difficulty up or down between games until the games feel genuinely competitive.
How chess engines work
A chess engine has two jobs: generate candidate moves and judge the resulting positions. Traditional engines pair a fast search — examining millions of positions per second, pruning branches that cannot matter — with a handcrafted evaluation function scoring material, king safety, pawn structure and piece activity. The engine plays the move at the root of the best line it can see.
The last decade changed the second half of that equation. Neural network evaluation, popularised by AlphaZero and adopted into open-source engines through NNUE, replaced human-written scoring rules with a network trained on hundreds of millions of positions. The result is engines that evaluate long-term positional factors — a knight outpost, a locked pawn chain, compensation for sacrificed material — far more accurately than the previous generation.
The practical upshot: the strongest freely available engine today, Stockfish, plays at a level far beyond any human world champion. Estimates put top engine strength well above 3500 Elo, against roughly 2800 for the strongest human players. Nobody beats it in a fair game. This is not a discouraging fact — it is the reason engine levels exist.
What engine levels actually mean
When a board offers "level 1" through "level 10", it is not running ten different programs. It is handicapping one engine in some combination of three ways:
- Search depth limits. The engine is only allowed to look a few moves ahead, so it misses tactics beyond that horizon exactly as a human would.
- Time or node limits. The engine gets a fraction of a second per move instead of several seconds.
- Deliberate imperfection. The engine picks from among its top-rated moves rather than always the best, or injects occasional errors to simulate a human rating band.
The third method produces the most realistic opponents. A depth-limited engine plays strangely — inhumanly solid positionally, then abruptly blind to a five-move tactic. A rating-calibrated bot makes the kind of mistakes an actual 1200-rated player makes, which is far better preparation for facing real people.
| Level band | Rough playing strength | Behaves like |
|---|---|---|
| Levels 1–2 | Under 800 | Hangs pieces regularly, misses one-move threats |
| Levels 3–4 | 800–1200 | Sees simple tactics, poor at planning |
| Levels 5–6 | 1200–1600 | Solid club standard, punishes loose pieces |
| Levels 7–8 | 1600–2000 | Strong club player, exploits every inaccuracy |
| Levels 9–10 | 2000+ | Effectively unbeatable without preparation |
These bands are indicative rather than standardised — one platform's level 5 is another's level 7. Treat the number as a dial to calibrate by results, not as a rating claim.
Choosing the right difficulty
The useful rule is to sit at the level where you win roughly a third to a half of your games. Winning nine out of ten teaches you nothing except that you can beat that setting. Losing nine out of ten teaches you nothing either, because you cannot tell which of your twenty mistakes actually cost the game.
A sensible progression looks like this. Start two levels below where you think you belong and win comfortably three games in a row. Move up one level. When you stop winning, stay there until you are winning half your games again, then move up. Most beginners can climb three or four levels within a couple of months, which is a far more motivating feedback loop than a rating graph that drifts sideways.
One caveat worth stating plainly: beating a bot on level 6 does not mean you would beat a human rated 1400. Bots make different mistakes from people, they never get flustered, and they never fall for the psychological tricks that decide club games. Use engine levels to measure progress against yourself, and use online games against real opponents to measure it against the world.
Training methods that work against a bot
Simply playing an engine repeatedly produces slow improvement. Structured practice produces fast improvement. Four methods worth using:
- Opening repetition. Play the same first five moves in twenty consecutive games. You will start recognising the resulting middlegame structures instead of feeling lost by move ten.
- Endgame drilling. Set up king and rook against a bare king, or king and pawn against king, and force the engine to defend it. These are the positions that decide close games and they can be learned to a mechanical standard.
- Handicap games. Remove one of your own pieces and play a level or two below your usual. Being down material forces you to play actively rather than drifting.
- The undo experiment. When a game turns, take the move back and try the alternative you rejected. Seeing both continuations from the same position teaches more than two separate games would.
These drills pair well with a broader study plan. Our page on learning to play chess sets out how to combine engine practice with tactics puzzles and game review across a typical week.
Using the engine as an analyst, not an opponent
The engine's most valuable function is not playing you — it is explaining you. Run a finished game through analysis and you get an evaluation graph, a list of blunders and inaccuracies, and the move the engine would have played instead. That is a personalised coaching report for free.
Two rules make it useful rather than overwhelming. Look at your own moves first, not the opponent's; and look only at the two or three biggest evaluation swings, ignoring every 0.3-point inaccuracy. A beginner does not lose games because of subtle positional drift, they lose because of one hanging knight on move twenty-two. Find that move, understand why you missed it, and move on.
The other rule is temporal: analysis belongs after the game, never during it. Consulting an engine mid-game on any online platform is cheating, is detected statistically, and results in account closure.
The limits of computer practice
Engine training has three blind spots. It removes clock pressure psychology, because a bot moves instantly and never makes you nervous. It removes the reading of an opponent — the hesitation that signals uncertainty, the confident tempo that suggests preparation. And it can encourage passive play, since bots below full strength rarely mount the kind of sustained practical pressure a determined human applies.
Set against those limits, the engine remains the most convenient sparring partner ever invented: no scheduling, no waiting, and the option to play chess at precisely the strength that stretches you today rather than the strength of whoever happens to be online.
The answer is not to abandon engines but to mix formats. A reasonable balance is roughly half engine practice for drilling and half human games for the competitive element, whether that is online play against strangers or a regular game with friends. The engine builds your technique; people test whether it holds up.