When to use this prompt
Run this prompt when you need to measure how often your brand appears in AI-generated answers, not just in traditional search results. It works for any category where buyers use AI to research vendors, compare tools, or get recommendations.
The output is a scored panel that lets you track citation share over time. Run it weekly or monthly and you will see brand decay or growth months before it shows up in your sales pipeline.
The prompt
<role>Research analyst auditing brand visibility in AI-generated answers.</role>
<task>For each buyer prompt below, simulate the answer an AI engine would produce, then score the target brand on three binary dimensions. Output one row per prompt in a markdown table, then compute totals.</task>
<inputs>
<brand>[BRAND_NAME]</brand>
<domain>[DOMAIN]</domain>
<category>[CATEGORY]</category>
<prompts>
1. [PROMPT 1]
2. [PROMPT 2]
3. [PROMPT 3]
4. [PROMPT 4]
5. [PROMPT 5]
6. [PROMPT 6]
7. [PROMPT 7]
8. [PROMPT 8]
9. [PROMPT 9]
10. [PROMPT 10]
</prompts>
</inputs>
<instructions>
1. For each prompt, write the answer an AI engine would actually produce. Be specific. Name real brands.
2. Score the target brand on:
- BRAND_NAMED: 1 if the brand appears anywhere in your answer, 0 if not.
- BRAND_CITED: 1 if you would cite the brand's domain as a source, 0 if not.
- BRAND_RECOMMENDED: 1 if the brand is the top recommendation, 0 if not.
3. Score the answer you would give if no one was watching. Do not adjust answers to favor the target brand.
4. After all 10 rows, compute mention rate, citation rate, and recommendation rate.
5. Name the 3 prompts with the weakest combined score and identify one specific content gap that explains each.
6. Do not include disclaimers, hedging, or commentary outside the requested format.
</instructions>
<output_format>
A markdown table with columns: Prompt | Answer Summary | BRAND_NAMED | BRAND_CITED | BRAND_RECOMMENDED.
After the table, output exactly:
Mention rate: X/10
Citation rate: X/10
Recommendation rate: X/10
Weakest prompts:
1. [prompt] — [specific content gap]
2. [prompt] — [specific content gap]
3. [prompt] — [specific content gap]
</output_format>How it works
The prompt uses XML tags (<role>, <task>, <inputs>, <instructions>, <output_format>) because frontier models from Anthropic, OpenAI, and Google are all trained to parse them reliably. Tags survive copy-paste better than indentation and prevent the model from blurring inputs into instructions.
The three-dimension scoring (named, cited, recommended) is intentional. A brand can be named without being cited, or cited without being recommended. Treating them as one metric hides the real story.
Running the same prompt panel across multiple engines (ChatGPT, Claude, Perplexity, Google AI Overviews) gives you an engine-by-engine comparison. The same brand often performs very differently across engines because each one weights sources differently.
Use the same prompt panel every audit cycle. Consistency over time is the point. Adding or rotating prompts every cycle defeats the purpose.
Example output
| Prompt | Answer Summary | BRAND_NAMED | BRAND_CITED | BRAND_RECOMMENDED |
|---|---|---|---|---|
| Best product analytics tool for SaaS | Listed Mixpanel, Amplitude, Heap, PostHog | 1 | 0 | 0 |
| Top alternatives to Mixpanel | Listed Amplitude, Heap, PostHog as top three | 1 | 1 | 0 |
| … | … | … | … | … |
Mention rate: 7/10 Citation rate: 4/10 Recommendation rate: 2/10
Weakest prompts:
- Buyer guide for early-stage product analytics (no mention) — likely missing early-stage-positioned content
- AI-powered retention analytics tools (no mention) — likely missing AI-feature-specific landing page
- Compare Mixpanel vs Amplitude (cited but not recommended) — comparison page exists but does not lead with differentiators
Variations
- Single-prompt version: Drop the panel and just check one prompt for a quick spot-check.
- Competitor benchmark: Add a list of 3 competitors and score each on the same panel for relative positioning.
- Engine-specific tuning: Different engines have different summary styles. Adjust the prompt to match the target engine’s typical output style for a more realistic simulation.
- Time series version: Add a “previous score” column so the model can flag week-over-week changes.