Some of those questions make me wonder if you're asking about an AI model being trained to play ZK or the BrutalAI we currently have.
Strictly speaking about 1v1, BrutalAI is somewhere between Orange and BrightYellow (Subgiant and Giant). If the match is allowed to drag on a lot, I think AI makes the kinds of decisions the upper end of BrightYellow or low end of BlueWhite. It is very easy to forget to keep building overdrive when you're trying to micro a lot, but the AI will never forget. If you let BrutalAI live long enough, it will spread the grid all over its base and it will try to make striders and even a super weapon.
It would be trivially easy to disginguish a match that involves an AI with one that does not for really anyone, anyone at all who has mild experience with 1v1. The biggest sign is simply that all of its units are always doing something. Not necessarily something that is good or even makes sense, but they're always trying to go somewhere. Players, especially lower skilled ones, often have units sitting somewhere either guarding the front or just being left unattended after being produced.
For the most part, BrutalAI tries to play like a player. Other games increase difficulty of computer controlled opponents by giving them more resources or increasing the stats of the units, mainly because the developpers couldn't figure out how to make an AI that can beat players on equal footing. Another thing AI gets in other games is full map knowledge. If you picked a really large map with multiple start positions the human player would have to scout to find their opponent, but the AI would know immediately where everything is and would also craft its strategy around what the player is doing despite never having scouted the human.
BrutalAI does not "cheat" in any way. It makes decisions based on the information it discovers, either from radar dots or having sent a unit to actually see what the player does. BrutalAI does not get more resources or better unit stats. It simply tries to make smart decisions.
The biggest flaw of the AI is that it is very defensive, so it is trivially easy to contain it and expand without any sort of opposition. A good player will out eco the AI pretty much from the start of the match and will effortly crush BrutalAI with pretty much any combination of units.
The AI will attempt to raid, but it is so late at trying that it is trivially easy to deflect. The AI also cannot learn. It doesn't know what are common weaknesses for lower skilled player opponents, or habits of players it faces frequently. It will try to gather information and try to attack in the exact same way it would regardless of its opponent, meaning it will never try a quirky strategy that is very effective against a lower skilled opponent, against specific play style or on a specific map.
There are many things that work extremely well vs AI because it doesn't know how to react, and might not even be detecting certain threats at all. For instance the AI is notoriously bad at handling phantoms. Because they stay cloaked when shooting, the AI never sees them unless it runs into one by complete accident. A player who can get to a point where they can spam phantoms will just overwhlem any amount of AIs. The AI doesn't know spamming raiders and sending them all at once to the front to form a detection line. It will probably just never find the phantoms unless they make a mistake that causes them to take damage and decloak.
The AI also doesn't know how to deal with terraform. If you make a wall the only thing the AI can do is make spiders or flying units. If a map features narrow paths surrounded by water, making a small trench is basically making sure the AI can never use that path again. It doesn't know how to restore the terrain or make a bridge. All it knows is: there is water, therefore make ships/amphs/hovers. More often than not, if water is not prominent on the map and you dig all the paths to water, the AI will just get stuck and form massive balls of units standing at the edge of water, unable to do anything at all.
I have seen some AI models trained to play other games, like Starcraft 2. In those an AI model can become far superior to a player, but this has more to do with the game design than the kind of decisions the AI makes. For instance in Starcraft it is extremely difficult for the player to keep a group of units from bunching up. All orders tend to have the units try to stand on the point at the destination of the command and end up in a ball. In Starcraft this is terrible because units die quickly and AOE attacks tend to deal a lot of damage, so clumping usually means instant death. It is trivially easy for an AI with inifite inputs to keep units spread apart, while humans must spend time moving individual units apart.
Also there are units like stalkers which have a short range teleport ability. Any attack that was targeting them when they teleport will miss, which in Starcraft an attack missing is very rare. If a unit is in range and shoots, it will almost always deal damage. It is very hard for a player to tell their stalkers to blink (teleport) when they're getting targeted and are about to die when they have many to control, but for AI that is trivially easy. It can check thousands of times per second where all the attacks are going and if it detects a stalker is about to die, it will command it to blink to safety. I haven't checked in a long time, but last I saw, an AI model trained to use blink stalkers will effortlessly beat all the best players in the world. In that specific game, we've entered the realm of machine speed capabilities humans just can't rival, and we've done so years ago.
But in ZK there aren't many mechanics like that and the action is slower. Some attacks are instant, but most have a bullet/rocket that has travel time that is slow enough for a player to respond to. That said, if allowed to give an infinite number of commands, a well trained AI model could individually control all their units and perfectly skirmish them, avoiding every single shot. That would be the equivalent of the SC2 example. At that point a human's only chance is to find something that can kill the AI opponent before the match scales up to superhuman control. It would be the chess equivalent of opening with a weird set of moves that the AI can't handle. I haven't followed chess in a long time but I've seen Magnus Carlsen in interview. Stockfish is humanly unbeatable, so if given sufficient processing power and better models, there is no reason to think the same can't be done for RTS or any game really.
