# Arival AI Lab Prompt Sheet

Use this with `arival_review_intelligence_demo_data.csv`.

Upload the CSV to Claude, ChatGPT, Gemini, or any LLM that accepts file uploads. The file contains three datasets separated by the `dataset` column:

- `operator_review`
- `competitor_review`
- `guide_activity`

## Step 1: baseline analysis

Use this first to show what a model can and cannot learn from operator reviews alone.

```text
First, analyse only the operator_review rows.

Tell me what useful insights you can find, but also what you cannot know yet because competitor_review and guide_activity rows are not being used.

Keep the answer practical for a tour operator. Separate:
- useful insights from reviews alone
- missing context
- questions you would need another dataset to answer
```

## Step 2: combined dataset reveal

Use this next so the room feels the before and after: summaries become business intelligence when datasets connect.

```text
Now combine all three datasets:
- operator_review
- competitor_review
- guide_activity

Explain what new insights become possible only because these files are combined.

Focus on:
- competitor gaps
- guide signals
- stronger sales proof
- better product decisions

Be specific about which dataset supports each new insight.
```

## Main lab prompt: build a review intelligence dashboard

Use this after the baseline and reveal prompts to create the full working dashboard.

```text
Your job is to show how combining different business datasets creates deeper insight than analysing reviews alone.

You are helping a tour operator turn guest reviews into business intelligence.

The uploaded CSV contains three datasets separated by the dataset column:
- operator_review: fictional reviews for one tour operator
- competitor_review: fictional reviews for competing tour companies
- guide_activity: fictional guide activity totals

Build a concise review intelligence dashboard.

Use this schema when analysing operator and competitor reviews:
- traveller_type: who the review appears to be from, using evidence or the traveller_type_hint
- emotional_driver: the underlying reason the guest cared, such as safety, confidence, belonging, discovery, status, ease, family harmony, or depth
- anxiety_resolved: any worry the review suggests was solved, such as cycling safety, pace, children staying engaged, crowds, value, or choosing the right tour
- memorable_moment: the specific detail or moment the guest remembered
- guide_or_staff_mention: named guide/staff praise or issue, if present
- product_theme: the operational or product theme, such as storytelling, pace, safety, hidden gems, family suitability, hospitality, food, history, logistics, or premium feel
- guest_language_quote: a short exact phrase from the review that could be reused as proof
- sales_use_case: where this review would be most useful, such as landing page, luxury advisor email, family-tour page, ad copy, FAQ, proposal, or post-tour follow-up
- confidence: high, medium, or low, based on how clearly the review supports the extraction

Return:
1. Executive summary: 5 bullets a busy operator would care about.
2. Top review themes from operator reviews.
3. Objection map: common anxieties and review quotes that resolve them.
4. Testimonial extraction: 5 strong proof quotes and where to use them.
5. Guide review velocity: calculate reviews_mentioning_guide / tours_led, named_positive_mentions / tours_led, and reviews_mentioning_guide / guests_hosted. Explain cautiously. Do not rank guides as employees.
6. Competitor benchmarking: analyse competitor reviews through the same schema and compare against the operator. Identify gaps and opportunities.
7. Marketing copy ideas: 5 lines or sections the operator could use based on guest language.
8. Data caveats: what not to over-trust.

Rules:
- Use only evidence in the CSV.
- Do not invent guest details.
- If evidence is unclear, say so.
- Keep it practical and concise.
```

## Optional dashboard prompt

If your tool can build artifacts, canvases, or simple dashboards, paste this after the first analysis:

```text
Turn the review intelligence analysis into a simple one-page dashboard for a tour operator.

Include:
- 4 KPI cards
- top themes
- objection map
- guide velocity table with caution notes
- competitor gap table
- best testimonial quotes by buyer type

Keep it readable for a non-technical operator. Use the uploaded CSV as the data source. Do not over-claim beyond the data.
```

## Focused prompt 1: structure raw reviews

```text
Using only rows where dataset = operator_review, convert 10 representative reviews into structured data using this schema:
review_id, traveller_type, emotional_driver, anxiety_resolved, memorable_moment, guide_or_staff_mention, product_theme, guest_language_quote, sales_use_case, confidence.

Use exact quotes where possible. If a field is not supported, write "unknown".
```

## Focused prompt 2: guide review velocity

```text
Using only rows where dataset = guide_activity, calculate:
- reviews_mentioning_guide / tours_led
- named_positive_mentions / tours_led
- reviews_mentioning_guide / guests_hosted

Then identify patterns cautiously. Do not treat this as a guide ranking. Explain what extra context would be needed before making management decisions.
```

## Focused prompt 3: competitor gap analysis

```text
Analyse operator_review and competitor_review rows through the same schema.

Find:
- themes guests praise in each company/category
- repeated complaints
- traveller types that appear underserved
- emotional outcomes each competitor seems to own
- positioning gaps a premium operator could use

Keep this practical for a tour operator deciding how to improve marketing and product positioning.
```

## Focused prompt 4: psychographic proof retrieval

```text
Create 5 buyer personas from the operator_review rows. For each persona, select the 3 best review quotes and explain:
- why this quote fits the persona
- what anxiety or aspiration it speaks to
- where to use it: landing page, advisor email, ad, FAQ, proposal, or post-tour follow-up
```

## Final prompt: evidence check

Use this at the end to teach evidence, confidence, and schema discipline.

```text
Check your own analysis.

List 5 claims you made, the exact evidence from the CSV that supports each claim, and whether confidence is high, medium, or low.

If any claim is weakly supported, rewrite it more cautiously.
```