The hardest part of ghostwriting isn't the writing. It's making the writing sound like someone else.
There's one test that determines whether a piece of ghostwritten content is working. When the executive's peers read it, the people who've spoken with them at conferences, sat across from them in deals, traded emails over years, do they recognize the executive's voice? Or do they read it and notice nothing, because nothing about it sounds like a specific person?
That recognition is voice fidelity. It's the test most ghostwriting fails.
What voice actually is
Voice is built from things that mostly happen below conscious awareness. The vocabulary choices a person makes when they're trying to sound precise. The sentence rhythm they fall into when they're explaining something they care about. The positions they tend to take. The references they reach for. The kind of humor they use, or don't. The way they open and close.
Most of this is invisible to the person whose voice it is. An executive describing their own writing will tell you it's "clear and direct" or "professional and conversational." That's a description that fits ten thousand other executives. The actual signal is in the patterns they can't see.
Which means ghostwriting that captures voice can't be built from what the executive tells you about their writing. It has to be built from the writing itself, and from extended conversation, and from a writer trained to hear what the executive doesn't know they're doing.
Why most agencies fail the test
Three reasons.
The first is source material. Most content agencies start engagements with a 30-minute intake call. Thirty minutes is enough to extract a list of topics and a vague sense of tone. It's not enough to identify the patterns that constitute a voice. The result: writers operate from a thin sketch and fill the rest with generic professional register. The output is competent. It sounds like an agency wrote it.
The second is the writer bench. Senior ghostwriting is a craft developed over years. Most content agencies run on a model that requires distributing work across many writers, many of them junior, to keep margins healthy. A junior writer can produce a passable blog post. They can't reliably recognize and reproduce the specific patterns of a senior executive's voice. That requires pattern recognition they haven't built yet.
The third is AI. When AI is used as a content generator rather than a research tool, it averages everything. AI is excellent at producing prose that sounds like reasonable B2B writing. That's the problem. Reasonable B2B writing is exactly what voice fidelity is supposed to escape. A piece written with AI in the loop trends toward the mean. Voice doesn't live at the mean.
What passing the test actually requires
The components are unsurprising once you list them.
Longer interviews. Ninety minutes minimum, recorded, with an interviewer trained to push past the executive's surface answers. The depth difference matters more than any other variable. Most voice failures trace back to insufficient source material.
Multiple voice samples. Past articles, internal memos, transcripts of talks, even long-form emails. The writer needs material the executive has actually written so the patterns are visible.
Dedicated calibration. The first few pieces of content go through more rounds than later ones. The writer learns the voice through iteration, and the editor's job is to flag voice failures specifically. Not "this could be tighter" but "this doesn't sound like them."
An editorial layer. Every piece reviewed by a senior editor before it reaches the client, with voice fidelity as one of the things being checked. Writers can't reliably catch voice failures in their own work. Editors can.
The test, in practice
There's a simple version of the test that works in the real world. After the first few weeks of writing in an executive's voice, the executive should have trouble telling which posts they wrote themselves and which were ghostwritten. If they can tell easily, voice fidelity isn't there yet.
When it's there, the work disappears. The executive's name is on the content. Their peers read it and recognize them. The ghostwriting becomes invisible, which is the entire point.