Why AI Writing Turns Generic Before the First Sentence

5 min read

The draft sounds fine. That is the problem. It says to understand customers, communicate clearly, and build trust. Nothing is wrong, but nothing sounds like...

Share:
A person follows a solid blue rock path toward an open sea horizon.

Trust and quality notes

Last updated
November 21, 2026

The draft sounds fine. That is the problem. It says to understand customers, communicate clearly, and build trust. Nothing is wrong, but nothing sounds like the person who asked for it.

Before trying another writing prompt, check what the model had to work with. A topic such as “better customer follow-up” tells it the subject. It does not tell it what happened, which decision was difficult, or what someone learned.

A useful first repair is to supply one concrete moment before asking for another draft. This guide offers a source-note format, interview questions, and a writing brief for that test. It does not require a content system or a new agent.

Separate the instruction from the material

An instruction defines the job: write a short post for a sales manager, use plain language, and explain one lesson. Source material supplies the substance: the question a buyer asked, the answer that failed, and the reason the team tried something else.

Adding “be specific” to an instruction does not provide the missing example. Asking for a confident voice does not establish that a claim is true.

Alex Lieberman's approach in Claire Vo's How I AI episode makes that distinction useful. His workflow interviews him for stories and opinions before drafting. The interview supplies the substance; separate guides shape the voice. This is his described process, not evidence that better inputs solve every writing problem.

For a team staring at a bland draft, the practical question is narrower: does the source contain anything the draft could not get from a general explanation of the topic?

Turn a broad topic into a usable source note

Consider this fictional example, not a customer case study.

The starting instruction is: “Write a post about following up with buyers.” It leaves the model to choose the problem, example, and lesson.

A stronger source note might read:

Situation: A buyer asked whether setup would disrupt the team's normal work.

First response: The seller sent the standard product overview. It described features but did not answer the setup question.

Decision: The seller replaced the overview with a short outline of the setup steps and the person responsible for each one.

Observation: The next conversation focused on those steps rather than the feature list. No sales or revenue result has been measured.

Point of view: A follow-up should answer the unresolved question before introducing another reason to buy.

Limit: This is one fictional illustration, not proof that the approach improves sales.

The second input gives the writer a particular argument. It also sets a limit. The resulting post should not invent a conversion gain, a delighted customer quote, or a claim that feature overviews never work.

For an actual note, replace the fictional material with an experience the team can document and is allowed to share. Remove names and sensitive details before uploading it. An anonymized example can still identify someone through its circumstances.

Ask questions that uncover a moment

A vague answer needs a narrower question, not immediate expansion into more paragraphs. These prompts are an editorial adaptation of the source's interview-first approach:

  • What exact question was the person trying to answer?
  • What did the team do first?
  • Where did that response fall short?
  • What choice did someone make instead, and why?
  • What was directly observed afterward?
  • Which conclusion remains uncertain?
  • What detail must stay private or needs permission?

A person can answer in rough notes or a transcript. The format matters less than whether the answers contain a real event and a defensible interpretation. “We care about customers” is a value statement. “We answered a setup question with a feature list” gives the writer a specific mistake to explain.

The companion workflow describes using the interview transcript as the drafting source. Its broader review and repurposing system is not necessary for this small test.

Give the writer an evidence-only brief

Keep the source note separate from the style instructions. The following brief can be copied into a writing conversation with an approved note attached:

Write a short post for the audience named below, using only the attached source note for factual substance.

Audience: Sales managers reviewing a buyer follow-up.

Reader job: Decide whether the follow-up answers the buyer's unresolved question.

Argument: Use the point of view recorded in the note. If the note does not support it, explain the gap before drafting.

Boundaries: Do not invent people, quotes, numbers, outcomes, or research. Keep observations separate from interpretations. Preserve uncertainty. Label fictional examples as fictional.

Voice: Plain, direct language. Prefer the concrete event over general advice.

Return: A draft, followed separately by the sentences that need evidence or permission. Do not publish, send, or update another system.

For teams collecting many possible assignments, the client research brief guide covers choosing which sources deserve a draft. This test starts after that choice: it improves the material for one piece.

Check the next draft against the note

Highlight every sentence that describes an event, result, or person's belief. Find its support in the note. Delete invented detail or replace it with a question for the person who supplied the material.

Then ask whether the draft retained the useful disagreement or decision. A draft can repeat every fact yet flatten the argument into “communication matters.” If that happens, clarify the point of view before polishing the wording.

The content review packet guide addresses the wider release check. Here, compare just one draft made from a broad topic with one made from a concrete note, keeping the audience and length the same. Record invented claims, missing details, and edits needed before either version is usable. This is a small diagnostic, not a promise of better reach.

Agentic Workers features include chat with file attachments and agents with supplied knowledge and instructions. That supports trying this brief with approved material. It does not mean a ready-made interview workflow is installed, or that an agent automatically verifies every claim. Start in a conversation; consider a recurring agent only if the same bounded job keeps returning.

The useful result is not more words. It is a draft whose specific argument can be traced to something the team actually knows.

Explore agents for a repeatable writing workflow.

Found this article helpful? Share it with others:

Share:

Written by

Agentic Workers Team