Writing & chat · guidance updated 4 Oct 2026
Perplexity prompt guide
Default is Perplexity's own search-tuned model; Pro users can pick frontier GPT, Claude and Gemini models. Research mode picks its model automatically. Your wording seeds the first web search, so query vocabulary matters most.
How to prompt Perplexity
- Write a complete question using the words that would appear on the pages you want found.
- Put the core question first; it seeds the search query.
- Add scope: time range, region, industry, source type (e.g. 'peer-reviewed', 'official statistics').
- Cap list lengths explicitly ('top 5'); otherwise length is arbitrary.
- Request the output format (table, bullets, short report) and ask for citations on every claim.
- Tell it to say so if it can't find reliable sources rather than filling gaps.
- Ask it to flag near-misses (related but not exact matches).
- Keep one research question per prompt; follow up for more depth.
- Use Research mode for multi-source reports; keep the question focused and specify angle and timeframe.
- Skip role-play personas and long style instructions; they add little to search quality.
The shape of a good prompt
[Specific question with key search terms] [Scope: timeframe, region, source types] [Output: format, item count, length] [Grounding: cite each claim; say if not found; note near-misses]
Avoid
- Single-word or vague queries ('heat pumps?')
- Uncapped 'list everything' requests
- Multiple unrelated questions in one prompt
- Long persona or creative-writing set-ups
- Pasting URLs instead of describing what to find
Example
Request are heat pumps worth it
Are air-source heat pumps more cost-effective than gas boilers for a 3-bedroom UK home in 2026? Compare upfront installation cost after current government grants, typical annual running cost, and payback period. Use UK sources from the last 12 months, preferring official or industry data. Present a comparison table followed by a 3-sentence verdict. Cite every figure, and say clearly if reliable recent data is missing.