You could test the different implementation yourself to see what happens:

const UnmountingChild = ({ unmount }) => {
  const { isMounted, isMountedNC } = useIsMounted();

  useEffect(() => {
    console.log('IS MOUNTED BEFORE FETCH', isMounted()); // true
    console.log('IS MOUNTED NO CALLBACK BEFORE FETCH', isMountedNC.current); //true
    fetch('https://jsonplaceholder.typicode.com/todos/1').then((d) => {
      console.log('IS MOUNTED', isMounted()); //false
      console.log('IS MOUNTED NO CALLBACK', isMountedNC.current); //false
    });
    unmount(); // <-- This triggers the unmount of the component
  }, []);

  return <>Child</>;
};

function useIsMounted() {
  const isMounted = useRef(false);

  useEffect(() => {
    isMounted.current = true;
    return () => (isMounted.current = false);
  }, []);

  return {
    isMounted: useCallback(() => isMounted.current, []),
    isMountedNC: isMounted,
  };
}

Check the demo HERE

The usage of a callback is useful because it lets you to read directly the value of the ref with ref.current when needed , simply by calling isMounted(), if you want to directly return the ref you just have to make sure to return all the ref and not just ref.current because if you pass ref.current you will pass just the actual value and not the mutable ref object hence its value will always be stuck to false. Returning the ref forces you to then read every time the .current property, and that's not very nice, and it could mislead people using the hook without knowing the internal implementation, while calling isMounted() is nicer to read, and to use and avoids usage troubles.