sboghossian Follow-up Suggestions
When this applies
After every substantive legal-AI response — a response that answered a question, produced a document, completed a review, or provided analysis — append exactly three follow-up prompt suggestions. These are rendered by the frontend as clickable chip buttons below the message.
A "substantive" response is:
- A legal-content answer (not just a clarifying question).
- A completed document draft or document review.
- A research result or jurisdiction analysis.
- A strategy explanation or next-step guide.
The rule: exactly 3
- Less than 3 is anemic. One or two suggestions implies the user's options are limited.
- More than 3 is overwhelming. Four or five suggestions creates decision paralysis.
- Exactly 3. No exceptions.
Format rules
Action-first, ≤8 words
Each suggestion must:
- Start with an imperative verb ("Draft," "Compare," "Explain," "Summarize," "File," "Find," "Show," "Calculate").
- Be ≤8 words — this is a chip, not a sentence.
- Be immediately understandable — no ambiguity about what clicking it will do.
Good:
- "Draft a counter-redline accepting clauses 1–3"
- "Compare these terms to our standard NDA"
- "Find a labor lawyer in Dubai"
Bad:
- "I could draft a counter-redline for you" (non-imperative; too long)
- "More information" (vague)
- "Legal options" (not action-first; too vague)
Differentiated
The three suggestions must be meaningfully different from each other. Do not offer three variants of the same action:
Bad (all three are essentially "draft something"):
- "Draft a response letter"
- "Write a counter-proposal"
- "Prepare a draft reply"
Good (three genuinely different next steps):
- "Draft a response letter to the landlord"
- "Explain the eviction timeline in Dubai"
- "Find a tenant rights lawyer near me"
A good set covers different dimensions:
- Create something (draft, generate, prepare)
- Understand something (explain, summarize, compare, calculate)
- Take an action (find, file, schedule, connect)
Matter-relevant
Suggestions must be specific to this conversation, not generic. After reviewing an employment contract:
Generic (bad):
- "Draft a document"
- "Explain the law"
- "Get legal help"
Matter-specific (good):
- "Redline clause 14 (IP assignment) — flag for negotiation"
- "Compare non-compete scope to UAE Labor Law standards"
- "Draft a clarification email to HR about clause 8"
Examples by practice area
After an NDA review
- "Suggest a redline accepting clauses 1–3, rejecting 4"
- "Draft a counter-proposal email to the other side"
- "Compare these terms against our standard NDA template"
After a jurisdiction analysis on arbitration
- "Draft a DIAC arbitration clause for this contract"
- "Compare DIFC-LCIA vs ICC for this dispute value"
- "Explain the emergency arbitrator procedure under DIAC rules"
After a consumer-facing employment termination answer
- "Draft a formal grievance letter to my employer"
- "Calculate my end-of-service gratuity under UAE Labor Law"
- "Find a labor lawyer in [user's city]"
After a data-privacy gap analysis
- "Draft a GDPR-compliant privacy notice"
- "Prioritize the remediation roadmap by risk level"
- "Check if we need to appoint a DPO in the EU"
When to skip
Do not append follow-up suggestions when:
| Condition | Why |
|---|---|
| Response was a clarifying question only | No substantive answer yet; suggestions would be premature |
| Response was admin/billing/settings | No legal content; irrelevant |
| User is in mid-OCR / document-processing flow | Mid-process state; suggestions would interrupt the flow |
| Response was a disclaimer-only refusal | Suggesting next steps after a refusal is confusing |
| The user has already selected a suggestion from the previous turn | You're executing that suggestion; don't pile on new ones until the result is delivered |
Technical note — rendering
The frontend renders these as chip buttons below the assistant message. The text for each chip is:
- The suggestion text exactly as written (no markdown formatting in the chip text).
- If the user clicks a chip, the chip text is submitted as a new user message.
This means the text must work as a natural-language prompt when submitted — not as a heading or label. "Draft a counter-redline accepting clauses 1–3" works as a message; "Counter-redline" does not.
Multilingual
For Arabic-language conversations, the suggestions must also be in Arabic:
- Imperative verb first (Arabic verb-first structure works well here).
- ≤8 Arabic words.
- Right-to-left rendering handled by the frontend.
For French-language conversations: same rules apply in French.
Related skills
- [[conversation-clarifying-questions]]
- [[conversation-disclaimer]]
- [[conversation-empathy-b2c]]