Play Chess Against the Computer: Levels, Engines and Practice
A computer opponent is the most patient sparring partner in chess. It is available at three in the morning, it never resigns in a huff, it will not abandon the game on move nine, and it does not sigh when you spend four minutes staring at a knight. For learning, that reliability matters more than most people expect. This guide explains what a chess engine is actually doing when it thinks, what difficulty levels change under the surface, where bots are unexpectedly weak, and how to use one to get better rather than just to pass the time. If you simply want an opponent right now, you can play chess against a bot in a browser tab and come back to the theory afterwards.
What a chess engine is doing when it "thinks"
A chess engine has two components: a search, which looks at possible move sequences, and an evaluation function, which scores the resulting positions. Everything else is refinement.
The search works in half-moves, called plies. One ply is a single move by one side. When an engine searches to depth four, it is considering your move, its reply, your reply to that and its answer — two full moves each. The branching is brutal: a typical middlegame position offers around thirty legal moves, so each extra ply multiplies the work by roughly thirty. That is why depth is the single biggest lever on both strength and how long the machine takes to answer.
The evaluation function scores the positions the search reaches. It counts material in centipawns — hundredths of a pawn — with a pawn at 100, a knight around 320, a bishop around 330, a rook 500 and a queen 900. Note that the knight and bishop are not identical, despite both being "worth three": most engines nudge the bishop slightly higher because the bishop pair tends to outperform in open positions. On top of material, evaluation adds positional terms: control of central squares, piece mobility, pawn structure, king safety.
What the engine then does is pick the move that leads to the best score assuming you also play your best. It has no plan in the human sense. It is not aiming at your weak square on f7 because it noticed a pattern; it is choosing the branch whose leaf positions score highest.
What difficulty levels actually change
"Easy, medium, hard" hides several different mechanisms, and they feel very different across the board.
Search depth. The cleanest approach. An easy setting might look two plies ahead, a medium one four, a hard one six or more. A two-ply engine sees that a capture is available and whether the immediate recapture is bad for it, and almost nothing else. A six-ply engine finds short forced mates and punishes any piece you leave undefended.
Deliberate randomness. Some easy levels do not simply search less; they pick at random among moves that are close to the best. This is what makes an easy bot hang pieces, and it is important for beginners — an engine that plays flawlessly but shallowly can still be crushing. A bot that genuinely blunders gives a new player something to win against.
Time budget. Rather than fixing the depth, the engine gets a fixed number of milliseconds and searches as deep as it can. This produces smoother difficulty scaling and is why an engine can seem stronger in simple endgames, where there is less to consider per ply.
Elo capping. Full-scale engines like Stockfish include a limit-strength mode that plays at a nominated rating. It works, but the resulting play often feels strange to a human — the engine makes a genuinely strong move for eight turns and then plays something inexplicable, because the weakening is applied artificially rather than reflecting a coherently weaker player.
Where computer opponents are surprisingly bad
Small browser engines have two structural gaps, and knowing them changes how you play against one.
No opening book. Serious engines consult a database of known opening theory before starting to calculate. A lightweight engine does not have one, so its early moves come from its evaluation function rewarding central control and piece activity. The moves usually look reasonable — developing knights, pushing central pawns — but they are reconstructed from principles rather than recalled. Against a player who knows a sharp line ten moves deep, this is a genuine weakness.
No endgame tablebase. Tablebases are precomputed solutions to positions with a small number of pieces, giving perfect play. Without one, an engine has to calculate endings like everyone else, and shallow searches handle them poorly. Endgames are where the branching factor drops but the depth needed rises — a king-and-pawn race can require twenty plies to assess, which is far past a casual bot's horizon.
There is a third, more general weakness: quiet closed positions where nothing is captured for fifteen moves and the game is decided by long-term planning. Search-based engines are excellent at calculation and mediocre at plans that pay off beyond their horizon. A strong club player can often out-manoeuvre a mid-strength bot by keeping the position closed and slowly improving pieces — which is, incidentally, terrible advice for a beginner, who should be doing the opposite.
How to actually improve using a bot
Playing endless casual games against a computer is pleasant and produces almost no improvement. Structured use does. Four approaches are worth your time.
- Play at the level where you win about half your games. Too easy and you learn nothing but bad habits; too hard and you lose without understanding why. Adjust until the games are genuinely close.
- Practise specific endings. Set up king and queen against a lone king, or king and rook against king, and force the mate. These are exact techniques, and the bot will defend correctly enough to test yours. Do the same with king and pawn against king to learn opposition.
- Use undo as a learning tool, not a rescue. When you lose a piece, take the move back, work out what you missed, and only then continue. The point is to identify the pattern, not to escape the consequence.
- Play the same opening ten times in a row. Against a bot you get consistent positions without wasting a human opponent's evening. You will start recognising the resulting middlegame structures, which is where most opening study actually pays off.
One habit worth building above all others: before every move, ask what the engine's last move now attacks. Most losses at beginner and intermediate level come from undefended pieces, not from deep tactics. Engines are relentless about taking free material — that is precisely what a shallow search is good at — so a bot is an excellent teacher of the habit of checking.
Choosing a bot to play against
The options range from browser boards with three fixed levels to full desktop installations of Stockfish. For most people, a lightweight browser opponent covers everything they need: three settings span the useful range from complete beginner to competent club player, and there is nothing to install. You can play chess against a computer opponent in a browser tab at several difficulty settings without registering, which is the shortest path from wanting a game to having one.
A locally-run engine like Stockfish becomes worthwhile when you want analysis rather than opposition — evaluation graphs, best-move suggestions and a numeric assessment of every position in a game you have already played. That is a different activity from playing, and it is worth keeping the two separate. Analyse after the game, not during it.
One practical advantage of engines that run inside your browser rather than on a server: the game keeps working when the connection drops. Nothing has to travel anywhere, because your own device is doing the thinking. That is not true of games against another person, as our guide to playing chess with friends online explains.
Bots versus humans: what each is for
An engine will punish every tactical mistake you make, immediately and without exception. That is enormously useful and slightly misleading, because human opponents do not. Against a person, an unsound attack sometimes works, a bluff sometimes lands, and time pressure changes decisions in ways an engine never experiences.
The sensible pattern is to use bots for drilling and humans for playing. Practise your endgame technique and your opening repertoire against the computer, where repetition is free; play people when you want a real game. A player who only ever faces engines tends to develop a strangely defensive style, having been punished for every speculative idea they ever tried.
Key point: difficulty settings mostly change search depth, and shallow engines blunder while deep ones do not. Neither has a plan. Play at the level where you win roughly half the time, and use the undo button to understand mistakes rather than to erase them. Further guides are listed on Bet365.
Common questions
What does "depth" mean on a chess engine?
The number of half-moves it looks ahead. Depth six means three full moves for each side. Each extra ply multiplies the calculation by roughly thirty, which is why depth costs so much time.
Why does an easy bot hang pieces?
Usually by design. Many easy settings add randomness so the engine sometimes chooses a move that is close to, but not, its best — which gives a new player something they can genuinely beat.
Can I beat a chess computer?
At low and medium settings, yes, with practice. At full strength, no — top engines have been beyond the best human players for well over two decades.
Is playing the computer good practice?
For tactics, endgame technique and opening repetition, very. For learning to handle time pressure and human unpredictability, less so. Use both.
Do I need to install anything?
No. Browser-based engines run on your device without installation, and are strong enough for anyone below serious club level. If you are entirely new, start with the rules of chess first, then look at the free online options or follow the beginner learning plan.