Personal Assistant Use Case
Recommendations for models that excel at memory, context retention, and personalized interactions.
💡 Note: For personal assistant tasks-such as recalling preferences, maintaining conversation history, managing schedules, or adapting tone over time-instruct-tuned models with strong long-context handling are preferred over thinking-tuned variants. These models prioritize coherence, empathy, and user-specific adaptation over raw analytical power.
Prompting
To get the best results from a personal assistant model, structure your prompts around the user's preferences, the immediate task, and the exact form of the response you want.
- State the user context clearly. Include the preferences, constraints, and background details the assistant should remember for this task.
- Define the assistant's job. Be explicit about whether it should plan, draft, summarize, remind, or ask follow-up questions.
- Constrain the output. Specify the tone, format, length, and whether the assistant should make assumptions or request clarification.
Below are examples of well-structured prompts, followed by examples of what to avoid.
Effective Examples
Planning Around Preferences
This prompt gives the assistant the user's constraints and a clear deliverable.
You are my personal assistant. Help me plan meals for the next 3 workdays.
User preferences:
```yaml
diet: vegetarian
allergies:
- peanuts
weekday_cooking_time: 20 minutes max
dislikes:
- mushrooms
goal: high protein lunches
```
Output requirements:
- Give exactly 3 lunch ideas.
- Each idea must include a short ingredient list and a 1-sentence prep summary.
- Do not suggest anything with peanuts or mushrooms.
- If a recipe normally uses meat, adapt it.Why this works:
- Provides durable user context: The preferences and restrictions are explicit.
- Assigns a concrete task: The model knows it is producing lunch plans, not general advice.
- Constrains the response: The output format prevents rambling and reduces irrelevant suggestions.
Drafting a Message
This prompt defines the relationship, tone, and exact communication goal.
Draft a reply to my manager.
Context:
```txt
She asked whether I can present the Q2 metrics on Friday.
I can do it, but I need the latest dashboard export by Thursday afternoon.
Tone should be professional and concise.
```
Write a 4-6 sentence email that:
- confirms I can present,
- asks for the dashboard export by Thursday afternoon,
- and mentions that I will share a draft deck before the meeting.Why this works:
- Supplies the missing situational details: The model has the necessary facts to write the message.
- Specifies tone and medium: "professional and concise" plus "email" narrows the style.
- Defines success criteria: The required points make the draft easier to trust and reuse.
Poor Examples (What to Avoid)
The following prompts are less effective because they force the model to guess your preferences, priorities, or intended output.
Preference-Free Planning Request
This prompt asks for personalized help without any personalization.
What should I eat this week?Why it's ineffective:
- No user context: There are no dietary needs, schedule constraints, or goals.
- No output shape: The model does not know whether to give recipes, groceries, or a full plan.
- Too broad: "this week" can mean anything from one meal to a seven-day calendar.
Underspecified Writing Request
This prompt gives too little information to draft a useful message.
Write a message to my boss.Why it's ineffective:
- Missing purpose: The model does not know whether the message is about scheduling, status, or a problem.
- No tone guidance: The draft could easily come back too casual or too formal.
- No factual constraints: Without details, the model is likely to invent context you did not provide.