AI SEARCH (GEO)

Write commentary an assistant can attribute

Use when a page needs a named opinion rather than another neutral summary.

expert-commentary-block.md

Download .md

You are drafting attributable commentary from what a named expert actually said. You are not inventing an opinion on their behalf.

Expert name and credentials: {{EXPERT_CREDENTIALS}}
Raw notes, interview transcript or written answers: {{RAW_NOTES}}
Page topic: {{PAGE_TOPIC}}
Who reads the page: {{AUDIENCE}}

Produce:
1. Three to five commentary blocks. Each is 40 to 70 words, first person, takes one clear position, and includes one specific detail from {{RAW_NOTES}}.
2. An attribution line for each block, drawn from {{EXPERT_CREDENTIALS}}, naming only the credential that makes this opinion relevant to {{PAGE_TOPIC}}.
3. A table: | Block | Position it takes | Supporting words from {{RAW_NOTES}}, verbatim | Needs approval before publishing (yes/no) |
4. Positions the notes point at but never state clearly enough to publish, written as questions to send back to the expert.

Constraints:
- Every sentence must trace to {{RAW_NOTES}}. Do not carry a position further than the notes carry it.
- Cut any block that could have been written without the expert, including generic advice for {{AUDIENCE}}, and list what you cut.
- No hedging stacks, no "in my experience" openers, no compliments to the reader.

Fill in before running

Replace each placeholder with your own detail. The more specific you are, the less the model invents.

{{EXPERT_CREDENTIALS}}
{{RAW_NOTES}}
{{PAGE_TOPIC}}
{{AUDIENCE}}

Getting a better result

Send section four back before publishing; the unstated positions are usually the quotable ones.
A credential line naming the relevant experience carries more than a job title does.
Get written sign-off on the exact wording, since an attributed opinion is far harder to retract than prose.

Questions about this prompt

When is this the right prompt?

When a page needs a position rather than another balanced summary, and you have actually spoken to someone qualified. It is not a way to manufacture an expert voice, since with no {{RAW_NOTES}} it has nothing to work from. If the interview has not happened, the interview is the task.

What do I need in front of me?

Raw notes, a transcript or written answers from the expert, plus their credentials in full. Rough notes are fine, and verbatim half sentences beat your tidied summary of what they meant, because the table has to quote supporting words from {{RAW_NOTES}} for every block the prompt produces.

What comes back, and which section is the useful one?

Three to five first person blocks of 40 to 70 words with an attribution line each, a table tracing every block to the expert's own words with an approval flag, and questions to send back. Section four is the useful part. The positions the notes gesture at are usually the publishable ones.

What is the mistake that costs me?

Publishing before the expert has approved the exact wording. An attributed opinion is far harder to retract than prose, and the risk here is a block carrying a position slightly further than the notes did. Keep the list of blocks cut for being generic too, so nobody quietly reinstates them.