AI SEARCH (GEO)

Stress test an answer across follow-ups

Use when the first answer about you reads well but the conversation turns against you later.

followup-question-stress-test.md

Download .md

You are running a buyer conversation and then auditing your own answers. Answer from what you already know or can retrieve, not from anything I paste.

Brand: {{BRAND_NAME}}
Category: {{CATEGORY}}
Buyer situation: {{BUYER_CONTEXT}}
Objections our sales team hears: {{KNOWN_OBJECTIONS}}

Write a five turn conversation, both sides, labelled by turn:
1. An opening question, naming no product, that a buyer in {{BUYER_CONTEXT}} would type.
2. A narrowing follow-up based on your own first answer.
3. A direct question about {{BRAND_NAME}}.
4. A challenge taken from {{KNOWN_OBJECTIONS}}.
5. "So which one should I go with."

Answer each turn in full before writing the next question.

Then an audit table: | Turn | Were we named | How were we described | What you drew on | Confidence (high/medium/low) |

Finish with the turn where our position weakened, the single statement that caused it, and what a page would need to say to survive that turn.

Constraints:
- Do not soften answers because we are the subject. Answer as you would for a stranger.
- Where you cannot name a source, write "recall, unsourced" rather than attributing it to a site.
- Name only competitors you genuinely associate with {{CATEGORY}}.

Fill in before running

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

{{BRAND_NAME}}
{{CATEGORY}}
{{BUYER_CONTEXT}}
{{KNOWN_OBJECTIONS}}

Getting a better result

Turn four is where most brands lose the recommendation, so read that answer before the others.
Run it in a clean session; anything earlier in the chat contaminates the result.
Keep the transcript with a date, because the turn where things break moves as sources change.

Questions about this prompt

When do I run this rather than the brand description test?

When the first answer about you reads well and you suspect the recommendation is lost later. The description test asks one question in isolation. Real buyers ask four or five in sequence, and the turn that decides the recommendation is usually the objection rather than the opening question.

What do I need in front of me?

The buyer situation in enough detail to generate a plausible opening question, and the objections your sales team actually hears rather than the comfortable ones. Run it in a clean session, because the prompt tells the model to answer from what it knows and earlier chat contaminates every turn.

What comes back, and which turn should I read first?

A five turn transcript with both sides, an audit table showing whether you were named at each turn and what the model drew on, then the turn where your position weakened and the statement that caused it. Read turn four first. That is where the objection lands and shortlists are lost.

What is the mistake here?

Rerunning until a flattering transcript appears. Variation between runs is normal, so a good run proves as little as a bad one. The output that matters is the single statement identified as the turning point and what a page would need to say to survive it, which is a content task.