A Dirichlet series is Σ a_n / n^s. Dirichlet L-functions L(s, χ) = Σ χ(n) / n^s for Dirichlet characters χ factor over primes and have analytic properties tied to prime distribution in arithmetic progressions.
The Riemann zeta function ζ(s) = Σ 1/n^s, which you already know, is the simplest Dirichlet series — the one where every coefficient a_n equals 1. A Dirichlet series Σ a_n / n^s is just the natural generalization: replace those constant coefficients with an arbitrary sequence. The series converges in some right half-plane Re(s) > σ_c and defines an analytic function there. The parameter s plays the same role as in ζ(s): it controls convergence and connects the series to analytic tools from complex analysis.
The deep structure comes from multiplicativity. You learned that an arithmetic function f is multiplicative if f(mn) = f(m)f(n) whenever gcd(m,n) = 1. When a_n is multiplicative, its Dirichlet series factors into an Euler product: Σ a_n / n^s = ∏_p (1 + a_p/p^s + a_{p²}/p^{2s} + ...). This is the same miracle you saw with ζ(s) = ∏_p 1/(1-p^{-s}), and it is what connects Dirichlet series to primes. The product form shows that each prime p contributes independently, and the series encodes arithmetic information prime-by-prime.
Dirichlet characters χ mod q are completely multiplicative, periodic functions taking values on the unit circle (or zero). They are designed to detect arithmetic progressions: the character χ acts as a kind of indicator that "weights" integers according to their residue class mod q. The associated L-function L(s, χ) = Σ χ(n)/n^s is multiplicative (since χ is completely multiplicative), so it has an Euler product L(s, χ) = ∏_p 1/(1 - χ(p)/p^s). This product converges and has no zeros for Re(s) > 1, and for non-principal characters χ, L(s, χ) extends analytically to Re(s) > 0 — crucially, L(1, χ) ≠ 0.
That non-vanishing at s = 1 is the key analytic fact. Dirichlet used it to prove his theorem: there are infinitely many primes in any arithmetic progression a, a+q, a+2q, ... as long as gcd(a,q) = 1. The proof mimics the elementary proof that ζ(s) → ∞ as s → 1⁺ implies infinitely many primes, but now the characters isolate specific residue classes. When you take a product over all characters mod q and extract the character that detects residue a, the divergence of L(s, χ) as s → 1⁺ forces a sum over primes ≡ a (mod q) to diverge — hence infinitely many such primes. The L-functions are the analytic engine that makes this algebraic decomposition work.