This is a scannable reference for prompting Manus, the autonomous general AI agent that plans and does multi-step work in the cloud and hands back a finished deliverable. Keep it open in a tab: grab the formula, pick a mode, borrow the constraint and audit lines, and paste your brief. For the full picks, see the best Manus prompts roundup.
The prompt formula
A strong Manus prompt is a delegation brief, not a one-line message. You are handing off a job that runs autonomously across many steps, so pin down the ambiguity up front — vagueness early compounds into wasted work. Every good brief has four parts: Goal + Constraints + Deliverable + Audit rules.
GOAL: [one clear objective — "Research and compare…", "Build a spreadsheet of…", "Create a 12-slide deck that…"]
CONSTRAINTS: scope [what's in/out] · timeframe [last 12 months] · budget [under $X] · geography [region] · quantity [N items] · sources [use … / avoid …] · weights [cost 40%, support 30%, ease 30%]
DELIVERABLE: [exact format] — [a spreadsheet with columns X, Y, Z / a 10-slide deck / a one-page report with sections A, B, C]
AUDIT RULES: cite every claim with a source URL · leave a field blank rather than guess · tag uncertain claims High/Med/Low · add a sources appendix · ask clarifying questions first if scope is unclearFill each line, delete what doesn't apply, and paste the whole block. Add levers where they fit: tell it to use Wide Research for big lists, name the deliverable tool (spreadsheet / Manus Slides / website), and set a stop condition. Ready-made structures live in the Manus prompt templates, and the agentic tasks guide walks through building one step by step.
Modes & model tiers
Pick the mode for the shape of the work and the tier for the depth of reasoning. Most delegated tasks run in Agent/Task mode on standard 1.6; reach for the others when the job calls for it.
| Mode / Tier | What it does | Best for |
|---|---|---|
| Chat Mode | Fast, conversational back-and-forth without launching a full autonomous run. | Quick questions, refining a brief, deciding scope before you delegate. |
| Agent / Task (autonomous) | Plans and executes a goal across many steps in the cloud, then returns a finished deliverable. | The default for real work — research, lead lists, analysis, decks, sites. |
| Wide Research | Launches many sub-agents in parallel instead of working a list one item at a time. | Large lists (50–200+): companies, products, papers, keywords, contacts. |
| Scheduled Tasks | Reruns the same brief on a daily / weekly / monthly cadence. | Monitoring, recurring lead gen, price and competitor tracking. |
| Playbooks | Reusable saved task recipes you can rerun with new inputs. | Repeated workflows you run often with the same structure. |
| 1.6 Lite | Fast, cheap, lighter reasoning. | Quick lookups, simple formatting, shallow tasks. |
| 1.6 | Balanced default reasoning and cost. | Most research, lead generation, and deck or spreadsheet builds. |
| 1.6 Max | Deepest reasoning, slower and more expensive. | Complex analysis, multi-step strategy, high-stakes decisions. |
Constraint vocabulary
These are the levers that keep an autonomous run on target. Drop the ones that apply into the CONSTRAINTS line of the formula — the phrasings below are copy-ready.
| Constraint | Example phrasing |
|---|---|
| Scope | "In scope: [X, Y]. Out of scope: [Z]. Ignore [category]." |
| Timeframe | "Only sources and data from the last [12] months (since [date])." |
| Budget | "Only options priced under [$X per month] / within a [$X] total budget." |
| Sources (use / avoid) | "Prioritize [official docs, SEC filings, company sites]. Avoid [forums, undated blog posts, AI-generated summaries]." |
| Geography | "Limit to [companies / results] headquartered in [region / country]." |
| Quantity / N | "Return exactly [50] rows / [10] options / [12] slides. Do not exceed." |
| Weights & thresholds | "Score each on cost (40%), support (30%), ease of use (30%); only include those scoring above [7/10]." |
| Stop condition | "Stop when you have [50] verified rows or after checking [200] candidates, whichever comes first." |
Deliverable formats
Name the format you want back and specify its shape — columns, slide count, or sections. Manus produces these natively, so ask for the finished artifact, not a description of it.
| Deliverable | How to ask for it | What to specify |
|---|---|---|
| Report | "Deliver a [one-page / detailed] research report with a sources appendix." | Sections (summary, findings, recommendation), length, whether to include charts. |
| Spreadsheet | "Build a spreadsheet with one row per [item]." | Exact column names, sort order, one row per entity, a notes/source column. |
| Slides (Manus Slides) | "Create a [12]-slide deck in Manus Slides." | Slide count, the arc (problem → solution → proof → ask), audience, tone. |
| Website / App | "Build a responsive [landing page / web app] and give me a live link." | Pages/sections, key copy, colour or brand notes, any form or data behaviour. |
| Dashboard | "Build a dashboard visualizing [dataset / metrics]." | Which metrics, chart types, filters, and the time range to cover. |
| Video | "Produce a short [explainer / promo] video." | Length, aspect ratio, script or key points, and the intended audience. |
Audit & quality modifiers
These lines make an autonomous output verifiable — the difference between a result you can trust and one you have to redo. Paste the ones you need into the AUDIT RULES line.
| Modifier | Paste this line |
|---|---|
| Cite every claim | "Cite every factual claim with a direct source URL in a Source column." |
| Blank, don't guess | "If you cannot verify a field, leave it blank — never guess or fabricate." |
| Confidence tags | "Tag each uncertain claim with a confidence level: High, Med, or Low." |
| Sources appendix | "Add a sources appendix listing every URL you used, grouped by section." |
| Assumptions log | "Keep an assumptions log noting any judgement call you made and why." |
| Clarify first | "Before starting, ask me any clarifying questions if the scope is unclear." |
Copy-paste example prompts
Five ready briefs that follow the formula. Fill the [brackets], delete any constraint that doesn't apply, and paste into the Manus prompt box.
1. Research brief (deep report)
GOAL: Research and compare the top vendors in [market/category] and recommend the best fit for [my company type].
CONSTRAINTS: Only vendors serving [region]; data from the last 12 months; prioritize official docs, pricing pages, and SEC/press filings; avoid undated blogs. Score on cost (40%), support (30%), ease of use (30%).
DELIVERABLE: A one-page report with sections — Summary, Comparison table, Recommendation — plus a sources appendix. Run on 1.6 Max.
AUDIT: Cite every claim with a source URL. Leave a field blank rather than guess. Tag uncertain claims High/Med/Low. Ask clarifying questions first if scope is unclear.Why it works: tight scope, weighted decision criteria, a named deliverable, and audit rules turn a broad "research this" into a verifiable recommendation.
2. Lead list at scale (Wide Research)
GOAL: Build a B2B lead list of [target companies] matching my ICP: [industry], [size], [region].
CONSTRAINTS: Use Wide Research to process the list in parallel. Return exactly 50 verified companies. Prioritize company sites and LinkedIn; avoid data-broker aggregators.
DELIVERABLE: A spreadsheet, one row per company, columns — Company, Website, HQ, Employees, Relevant contact (name + title), Email/LinkedIn, Fit notes, Source URL.
AUDIT: Cite each contact with a source URL. Leave email blank if unverified — do not guess. Stop when you have 50 verified rows.Best for: large prospecting lists — Wide Research runs the companies in parallel while the blank-don't-guess rule keeps the sheet clean.
3. Data analysis + dashboard
GOAL: Analyze the attached [sales/CSV] data and surface the drivers of [the trend I care about].
CONSTRAINTS: Cover [date range]. Segment by [region] and [product line]. Flag any data-quality issues before analyzing.
DELIVERABLE: (1) A short findings report with the top 5 insights ranked by impact, and (2) a dashboard with charts for [revenue over time], [segment breakdown], and [top/bottom performers].
AUDIT: Keep an assumptions log for any cleaning or filtering decision. Note sample sizes. Tag any low-confidence finding as Low.Why it works: it asks Manus to clean and question the data first, then delivers both a narrative and a dashboard, with an assumptions log for traceability.
4. Slide deck (Manus Slides)
GOAL: Create a 12-slide pitch deck for [product] aimed at [audience, e.g. seed investors].
CONSTRAINTS: Follow the arc — problem, solution, market size, product, traction, business model, competition, team, ask. Keep copy tight; use my Knowledge for brand voice. Cite any market figure.
DELIVERABLE: A 12-slide deck in Manus Slides, one idea per slide, speaker notes under each.
AUDIT: Cite every market-size or growth stat with a source URL on the slide's notes. Leave a stat out rather than invent one.Best for: fast first-draft decks — naming the slide count, the arc, and Manus Slides gives it a clear structure to fill.
5. Scheduled monitor (recurring)
GOAL: Every weekday at 9:00am, check [these 5 sources/newsletters/competitor pages] for anything new about [topic].
CONSTRAINTS: Only items published since the last run. Use the Browser Agent to open each site. Skip duplicates already in the sheet.
DELIVERABLE: Append new items to my tracking spreadsheet, columns — Date, Source, Headline, Summary, Link, Why it matters. Set this up as a Scheduled Task (daily, weekdays).
AUDIT: Include the source URL for every row. Leave the summary blank rather than paraphrase inaccurately.Why it works: a Scheduled Task plus the Browser Agent turns a one-off search into hands-off daily monitoring that appends straight to your sheet.
Frequently Asked Questions
What makes a good Manus prompt different from a chatbot message?
Manus runs a goal autonomously across many steps in the cloud, so a good prompt is a delegation brief, not a message. It states one goal, pins down hard constraints (scope, timeframe, budget, sources), names the exact deliverable format, and adds audit rules. Vague prompts waste autonomous work; specific briefs come back closer to usable.
When should I tell Manus to use Wide Research?
Use Wide Research whenever your task is a large list — companies, products, papers, keywords, contacts — roughly 50 items or more. Instead of working the list one item at a time, Manus launches many sub-agents in parallel, so 100 rows finish far faster. Add the line "Use Wide Research" and name the list.
Which Manus tier should I pick: 1.6 Lite, 1.6, or 1.6 Max?
Pick 1.6 Lite for fast, cheap, shallow tasks like quick lookups or simple formatting. Use standard 1.6 for most research, lead lists, and deck building. Reserve 1.6 Max for the deepest reasoning: complex analysis, multi-step strategy, and tasks where a wrong assumption is expensive.
How do I stop Manus from making up facts?
Add audit rules to the prompt: cite every claim with a source URL, leave a field blank rather than guess, tag uncertain claims with a confidence level (High/Med/Low), and add a sources appendix plus an assumptions log. These make the autonomous output verifiable so you can trust or check each row.
Can Manus fill forms and pull data from real websites?
Yes. The orchestrator delegates to a Browser Agent that opens a headless Chromium browser in the cloud, navigates real sites, fills forms, clicks, screenshots, and extracts data. Name the sites and the fields you want, and give it a stop condition so it knows when the list is complete.
How do I set up recurring monitoring in Manus?
Use Scheduled Tasks. Write the task as a normal brief, then set a daily, weekly, or monthly cadence — for example "every day at 9am, check these 5 sources and append new items to the sheet." Manus reruns the same brief on schedule, which is ideal for lead gen, price tracking, and competitive monitoring.