In this case Wikipedia's explanation is quite poor.
Let's take a step back. Consider a polynomial
$$
P(X)=X^d+a_1X^{d-1}+a_2X^{d-2}+\cdots+a_d
$$
and assume:
- the coefficients $a_i$ are rational numbers;
- the polynomial is irreducible, meaning that it admits no decomposition $P(X)=A(X)B(X)$ where $A(X)$ and $B(X)$ are both non constant.
Then it is known (essentially by the Fundamental Theorem of Algebra) that $P(X)$ admits $d$ different roots $\alpha_1,...,\alpha_d$ in the complex numbers, none of which is a rational number. A root of a polynomial $P(X)$ is some number $\alpha$ such that $P(\alpha)=0$.
We say that two complex numbers $\alpha$ and $\beta$ are conjugated if they both appear among the roots of a polynomial as above.
Back to square roots.
When the degree of the polynomial is $2$, i.e.
$$
P(X)=X^2+aX+b,
$$
the roots are $\alpha_{1,2}=\frac12\left(-a\pm\sqrt{a^2-4b}\right)$ so that in fact one root is transformed into the other (conjugation) by switching sign to the square root.
The "radical not appearing in $a$" clause should be read "$a$ and $b$ are rational numbers".
A more general situation is that when the polynomia $P(X)$ is allowed to have coefficients in just any field, but let's keep it simple for the time being.