How should you evaluate voice capture and work samples?

Evaluate voice capture by tracing samples from executive input through research, drafting, review, and publication. Require the candidate to identify the work performed by its proposed team, the executive's substantive contribution, later edits, and the selection or payment mechanism behind publication. A polished byline or post supplies too little evidence on its own.

Establish provenance before judging the prose

For each sample, ask who originated the idea, interviewed the executive, researched and checked claims, drafted, edited, approved, submitted, and published it. Confirm whether those people will serve your account. Record material changes made by the executive, company reviewers, another agency, or the publisher.

Classify the result as company-controlled publishing, independent editorial selection, sponsored placement, or another paid opportunity. Ask what the provider controlled and what a publisher, platform, organizer, or internal team controlled. Do not credit a provider with an editorial outcome or audience result it cannot substantiate.

Client confidentiality can prevent disclosure of private drafts or names. Accept redacted artifacts, an anonymized process walkthrough, or references who can verify the candidate's role. Treat unexplained inability to show any attributed or verifiable work as a risk, not proof of discretion.

Strong voice capture preserves reasoning and uncertainty

Ask how the team learns an executive's vocabulary, reasoning, relevant experience, limits, and tolerance for personal disclosure. Inspect an annotated example or process record when confidentiality permits. Look for a traceable line from source interview and evidence to draft, changes, and final executive approval.

Voice fidelity means preserving the executive's actual point, level of certainty, and tradeoffs across formats. It does not mean copying verbal habits into every sentence. Ask how the writer handles a weak premise, missing evidence, conflicting feedback, and an executive who disagrees with the recommended edit.

The executive owns attributed judgment, personal boundaries, and final approval. The company owns official facts, required review, and company-controlled source material. The program manager routes evidence and feedback without silently resolving a substantive disagreement.

Separate sample quality from audience outcomes

A sample demonstrates process and craft. It does not establish that priority stakeholders saw, understood, trusted, or acted on the work. Ask the candidate to label output counts, audience-response evidence, outcomes, and organizational impact, including the source and time period for each figure.

AMEC's Integrated Evaluation Framework distinguishes outputs from audience response, outcomes, and impact. Apply that distinction to claims about reach, engagement, invitations, qualified conversations, or business contribution.

A paid trial can resolve ambiguous evidence

A finite, paid exercise can test the proposed delivery team and real review process. Use a representative but sanitized subject, define the input and decision being tested, and limit the output to what the buyer needs to assess. Avoid sharing material nonpublic information, personal security details, or unnecessary personal data.

Score the trial on:

• accuracy and traceability of claims;
• fidelity to the executive's meaning and uncertainty;
• quality of questions and editorial judgment;
• ability to distinguish personal view from company position;
• treatment of sensitive, regulated, or unsupported material;
• clarity and speed of the review workflow; and
• acceptance of corrections and disagreement.

Set rights and AI safeguards before the trial

The payer approves the trial fee and commercial terms. The executive approves use of their interview material and identity. The program manager controls source delivery, reviewers, deadlines, and return or deletion of trial materials. State who owns inputs, drafts, comments, recordings, and final output, and whether either party may display or cite the work.

If the provider uses AI, ask what data enters each tool, who can access it, how long it is retained, whether it trains a model, what subprocessors receive it, and how people verify claims. NIST's July 2024 Generative AI Profile identifies confabulation, data privacy, information security, and intellectual-property risks relevant to this review.

Carry the trial's authorship, confidentiality, rights, AI, and approval decisions into ghostwriting, authorship, AI, and approval and the final provider contract.