Perplexity Prompts Cheatsheet - Research & Retrieval-Augmented Q&A Techniques
Perplexity tends to get used as a smarter chat window, which is a problem because the citations it volunteers make people stop verifying. The useful framing is that you are outsourcing retrieval: the model finds and ranks, while you constrain what it searches, to what date, and in what shape it must answer. This table groups material by question phrasing, focus modes, date and source constraints, output-format constraints, chained follow-ups, and ready templates, with the emphasis on rewriting vague questions into checkable retrieval requests. Two rules for judging a result: every factual sentence should carry a clickable source, and when you ask what remains uncertain, the model should admit the boundary instead of improvising. Reach for it when a conclusion must be current, when you are writing a sourced research brief, when comparing several options within hours, or when you need a strict comparison table. The dividing line from NotebookLM is source scope: this one searches the open web and must stay traceable, while that one can only answer from materials you supplied.
Questioning Style 5
Write a full question, not keyword soupState goal and constraints: audience/depth/useScope it: region/industry/timeOne question at a time; split complex asksAsk for an actionable conclusionFocus Modes 5
Web / AcademicPro Search (deep research)Research / Reddit / YouTubeWriting / Math / FinancePick the mode by task to raise relevanceDate & Source Constraints 5
...from 2025 or later onlyUse only reputable sourcesCite every factual claim (default; emphasize if needed)Prefer official / primary sourcesExclude marketing-pitch sourcesFormat Constraints 5
Present A/B differences as a comparison tableOutput a step-by-step action guideCited bullet summaryMark "consensus / dispute / unknown"State confidence or limitationsChained Follow-ups 5
Pro Search chains for deeper digsFollow up to clarify ambiguityAsk it to list what is still uncertainSwitch Focus mode to cross-verifyAfter free quota, fall back to base modeTemplates 5
After 2025, what reliable progress on {topic}? Reputable sources only, cite.Compare A vs B for {scenario} in a table, flag disputesLiterature review of {question}: consensus/dispute/unknownUse Academic mode for top-cited {field} papers in 3 yearsFrom the above, give 3 actionable recommendations with evidenceCommand Examples
Recent developments: date floor, source bar, citations
After 2025, what reliable advances are there in {topic}? Use only trustworthy sources and attach citations.Three constraints in one sentence: the date floor prunes stale conclusions, the source bar excludes marketing pages, and citations make every advance clickable and checkable.
Decision comparison: table with disputed points called out
Compare A and B for {useCase} in a table, and mark the points that are disputed.The disputed-point clause is what makes this useful: comparison tables default to presenting two columns as equally settled, and this forces disagreement into the open.
Literature review: split into consensus, disputes, and unknowns
Produce a literature review on {researchQuestion}, separated into consensus, disputed, and unknown.The three-way split beats a conventional review because the unknown column is literally your next search queue, instead of you inferring the gaps from conclusions.
Academic lookup: mode and time window pinned
Using Academic mode, find highly cited {field} papers from the last three years.Pinning mode, field, and window at once keeps recall from drifting into generic web literature; take the list and cross-check it once in Web mode.
Common Pitfalls
- Writing the question as a stack of nouns: its strength is answering questions, not keyword matching, and recall quality drops accordingly.
- Asking for recent developments with no date floor, so training-era versions of the facts sit next to current ones with no version note.
- Asking once in a single Focus mode and treating the result as settled: that mode source bias lands in the answer unflagged.
- Omitting a format constraint and receiving an unverifiable wall of prose that can only be copied wholesale or discarded.
- Resubmitting one broad question after the Pro Search free runs run out, instead of splitting it into rounds of one question each at identical cost but far more checkable intermediate steps.
Tips
- Perplexity's paradigm is "Q&A as search": full questions beat keyword soup, and every claim is cited by default — great for traceable research.
- Recency is its strength — "from 2025 or later only" plus "Use only reputable sources" cuts noise sharply.
- Wrong Focus mode drops quality: Academic for papers, Pro Search for depth (5 free/day), Reddit for community sentiment.
- For hard topics chain with Pro Search and ask it to list "what is still uncertain" — exposing knowledge boundaries beats false omniscience.
FAQ
Why write a full question instead of a pile of keywords?
The paradigm is question-and-answer where the search happens: the model retrieves and ranks, you decide what it looks for. A full sentence tells it which facet to search, and adding the entity full name plus your audience, depth, and intended use beats keyword stuffing. Narrowing by region, industry, or time shrinks the retrieval surface further, and one question at a time beats one giant vague ask.
How do I pick a Focus mode?
Pick by the kind of source you need, because a mismatch degrades quality. Web for the open internet, Academic for literature, Pro Search for multi-step deep research at five free runs per day; then Research for survey work, Reddit for community sentiment, YouTube for video, and Writing / Math / Finance for their respective tasks. Asking the same question in two modes and cross-checking beats trusting one, since a single-mode answer tends to inherit that mode source bias.
How do I keep the facts both current and credible?
Constrain on two axes. Time: a limit such as from 2025 or later only prunes stale material. Sources: Use only reputable sources raises the bar, then go further by asking it to prefer official and first-party sources and exclude marketing-softcontent pages. Citation on every factual sentence is the default behavior and worth restating explicitly. Close the ask by demanding actionable conclusions rather than a list of information.
How do I turn a wall of prose into something I can paste into a doc?
Specify the shape up front with a format constraint: a comparison table for A-versus-B decisions, a step-by-step guide for how-to questions, a cited bullet summary for literature reviews, and explicit consensus / disputed / unknown labels for contested topics, plus a confidence or limitations note. Marking uncertainty is not weakness; it surfaces the overconfident parts so you know what still needs checking.
How do I work a complicated research question?
Drive it through Pro Search as a chain: refine on the previous turn rather than restating the whole thing. Ask it to list what remains uncertain, and follow up to narrow the ambiguous parts. When the free runs run out, drop back to the basic mode instead of resubmitting the same broad question; iterating one question at a time buys far more than a single catch-all ask.
Official References
Each command links to its official documentation below, so you can verify the latest usage and read deeper.
Maintained by LaoHand
Publicly updated on Aug 21, 2026, continuously proofread against official docs.
Contact Us
Wrong command or description? Send us corrections, business inquiries or product feedback by email.
Contact Us