Manus is an autonomous general AI agent: you give it a goal in plain language and it plans the steps, does the work across many steps in the cloud, and hands back a finished deliverable — a report, a spreadsheet, a slide deck, or a website. So the way you prompt it is different from a chatbot. You are not sending a message; you are writing a delegation brief for a worker that will run unsupervised for many minutes.
The whole formula fits in one line: Goal + Constraints + Exact deliverable + Audit rules. Nail those four and Manus produces verifiable, on-target output. Skip them and it wanders. Below is each part with copy-paste example prompts, then six complete briefs. For a broad set of ready-made prompts, see the best Manus prompts roundup.
The Manus prompt formula
A strong Manus prompt has four parts: a clear goal/role, hard constraints, an exact deliverable, and audit rules. Write all four every time and Manus's autonomy works for you instead of against you.
- Goal / role — one clear objective: "Research and compare…", "Build a spreadsheet of…", "Create a 12-slide deck that…".
- Hard constraints — scope, timeframe, budget, geography, number of items, sources to use or avoid, and any decision weights or thresholds ("weight cost 40%, support 30%, ease 30%").
- Exact deliverable — the format you want back, with named columns, sections, or slide count: "a spreadsheet with columns X, Y, Z", "a 10-slide deck", "a one-page report".
- Audit rules — cite every claim with a source URL, leave a field blank rather than guess, tag uncertain claims High/Med/Low, and add a sources appendix.
Here is the formula itself, ready to keep next to your keyboard:
Goal + Constraints (scope, timeframe, budget, sources) + Deliverable (exact format) + Audit rules (cite sources, don't guess, confidence tags)Why it works: Every part removes a decision Manus would otherwise make for you across dozens of autonomous steps, so the finished result matches what you actually needed.
Pin down constraints so autonomy doesn't wander
Give Manus a scope, a timeframe, a budget, explicit sources, decision weights, and a stop condition — because ambiguity early compounds over many steps. A single vague word ("recent", "affordable", "top") becomes a different guess in every sub-agent, and you only find out at the end of a long run.
Weights and thresholds are the highest-leverage constraint. If a choice depends on trade-offs, tell Manus how to score them so its ranking is yours, not its own.
Compare project-management tools for a [15-person design agency]. Scope: only tools with a free or sub-$12/user/month plan, native time tracking, and a public API. Timeframe: pricing and features as of [current month/year] only. Score each on a 100-point scale using these weights: price 40%, ease of use 30%, integrations 20%, support 10%, and rank them by total. Stop when you have evaluated exactly [8] tools. Deliverable: a ranked comparison table. Cite the pricing page URL for every price; if a plan detail isn't published, leave it blank and tag it Low confidence.Why it works: Explicit weights turn a subjective "best" into a repeatable score, and the stop condition caps the run so it doesn't sprawl into 30 tools.
Research the [electric cargo bike] market for a buyer in [Germany]. Constraints: bikes available new in the EU, priced €2,000–€5,000, with a payload over 100kg. Ignore listings older than [12 months] and ignore marketplace resellers. Use only manufacturer sites and [two named review outlets]; do not use forum posts. Deliverable: a shortlist of [6] models with a one-paragraph rationale each. Ask me clarifying questions first if any constraint is ambiguous.Best for: Any research task where "good" is subjective — the geography, price band, source whitelist, and freshness window stop Manus from drifting.
Specify the exact deliverable
Name the format and its structure: the exact columns for a spreadsheet, the slide count and section order for a deck, or the sections for a report. Manus can produce spreadsheets, slides, reports, dashboards, and websites — so telling it which one, and its precise shape, is what determines what lands in your workspace.
Build a spreadsheet of [SaaS competitors] in [the email-marketing space]. One row per company, with these exact columns: Company, Website, HQ country, Founded year, Pricing model, Entry price/month, Free plan (Y/N), Standout feature, G2 rating, Source URL. Include [20] companies. Sort by Entry price/month ascending. Leave any cell blank if the data isn't publicly stated rather than estimating, and put the source URL you used in the final column for every row.Why it works: Naming every column gives Manus a schema to fill, so you get a clean, sortable table instead of a prose write-up you have to reformat.
Create a [10]-slide deck in Manus Slides pitching [our new analytics product] to [Series A investors]. Use this exact section order: 1) Title, 2) Problem, 3) Solution, 4) Product demo, 5) Market size, 6) Business model, 7) Traction, 8) Competition, 9) Team, 10) The ask. One core idea per slide, minimal text, and a speaker note under each. Base the market-size and competition slides on cited research and add the source URLs in the speaker notes.Best for: Decks and reports — fixing the slide count and section order keeps the narrative tight instead of ballooning to 30 slides.
Make output auditable
Add four audit rules to every research 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. Autonomous output is only useful if you can spot-check it, and these rules make that possible in minutes.
For every factual claim in your report, add an inline citation with the exact source URL. Where a number is estimated or inferred rather than directly stated by a source, tag it [Low confidence] and explain the inference in one line. Never fill a gap with a plausible guess — leave it blank and note "not found". End with a Sources appendix listing every URL you used, grouped by section.Why it works: This block is a reusable footer — paste it onto any research brief and the finished deliverable arrives pre-annotated for review.
Add a "Confidence" and "Source" column to the spreadsheet. Confidence is High only when the value comes straight from an official source (company site, filing, or primary data), Medium for reputable secondary sources, and Low for anything inferred. Put the direct URL in Source for every populated cell. If two sources disagree, use the primary one and note the conflict in a Notes column.Best for: Lead lists and data tables — a Confidence and Source column per row lets you trust the High rows and manually verify only the Low ones. Reuse it with the Manus prompt cheat sheet.
Use the right Manus feature
Match the feature to the job. Wide Research launches many sub-agents in parallel for big lists (companies, products, papers — 50 to 200+ items). Scheduled Tasks run daily, weekly, or monthly for monitoring and lead gen. The Browser Agent opens a real headless browser to navigate live sites, fill forms, and extract data. Manus Slides builds decks. 1.6 Lite is fast and cheap; 1.6 Max reasons the deepest for complex planning. Knowledge stores reusable context like brand voice or your ICP. And "ask clarifying questions first" is your safety valve when scope is fuzzy.
Use Wide Research to profile these [120] companies in parallel: [paste list or attach file]. For each, use the Browser Agent to open the company's live site and pull: employee count, HQ city, primary product, pricing tier, and the name and email format of the head of [marketing]. Return one spreadsheet row per company with a Source URL column. Leave any field blank you can't verify on the site; do not guess emails — infer the format only from a stated example and tag it Low confidence.Best for: Large lists that need live-site data — Wide Research parallelizes the run and the Browser Agent gets facts from the source instead of stale search snippets.
6 full example prompts
Six complete delegation briefs, one per common Manus job. Each is copy-paste, uses [bracketed placeholders], and carries all four parts of the formula. Combine them with the Manus prompt templates to build your own.
1. Deep research brief
Goal: Research the [state of AI coding assistants] for a [head of engineering] deciding what to roll out to a [40-developer team].
Constraints: Cover only tools generally available as of [current month/year]. Focus on [enterprise plans]. Sources: official product docs, pricing pages, and [two named analyst outlets] only — no vendor blog hype. Compare exactly [7] tools.
Deliverable: A one-page report with these sections — Executive summary, Comparison table (Tool, Model, Price/seat, IDE support, Data-privacy stance, Standout strength, Weakness), Recommendation with rationale, and Risks.
Audit rules: Cite every claim inline with a source URL. Leave any unpublished detail blank and mark it "not found". Tag any inferred number [Low confidence]. End with a Sources appendix.
Use Manus 1.6 Max. Ask me clarifying questions first if scope is unclear.Why it works: It fixes the audience, the tool count, the source whitelist, and the exact report sections, so the autonomous run has nothing left to invent.
2. Lead list at scale
Goal: Build a B2B lead list of [Shopify agencies] in [the UK] for our outreach.
Constraints: Companies with [5–50] employees, active in the last [12 months], with a live website. Exclude freelancers and marketplaces. Target [80] companies. Use Wide Research to run these in parallel and the Browser Agent to confirm each site is live.
Deliverable: A spreadsheet with columns — Company, Website, City, Employee band, Services, Contact name, Role, Email, Email source, LinkedIn URL, Confidence, Source URL.
Audit rules: Only include an email if it's published on the site or inferred from a stated pattern (tag inferred ones Low confidence). Leave blank rather than guess. One source URL per row.
Stop when you have [80] verified rows.Why it works: Wide Research plus a hard stop condition and a per-row Confidence column turns a huge scrape into a list you can trust and act on.
3. Data analysis
Goal: Analyze the attached [12 months of sales data] ([CSV]) and tell me what changed and why.
Constraints: Treat [currency] as [GBP]. Exclude [refunded orders] and any row with a null date. Compare [H1 vs H2] and [this year vs last year] where data allows. Do not pull in outside data.
Deliverable: A dashboard plus a short written summary with these sections — Headline metrics (revenue, orders, AOV, MoM growth), Top and bottom [5] products by revenue, Trend chart of monthly revenue, [3] insights, and [3] recommended actions.
Audit rules: State every calculation's basis (which columns, which filter). Flag any metric affected by missing data as [Low confidence]. Do not infer causes you can't support from the data — label those as hypotheses.
Use Manus 1.6 Max.Why it works: Naming the filters, the comparisons, and the exact dashboard sections keeps the analysis reproducible and stops Manus from inventing causes.
4. Slide deck
Goal: Create a [12]-slide deck in Manus Slides for a [quarterly business review] with [the leadership team].
Constraints: Audience already knows the product, so skip basics. Cover [Q2]. Tone: direct, data-first. Pull metrics only from the attached [KPI sheet].
Deliverable: 12 slides in this order — Title, Quarter at a glance, Revenue vs target, Pipeline, Wins, Misses and why, Customer health, Product progress, Team, Risks, Next-quarter priorities, Asks. One idea per slide, a chart where relevant, and a speaker note under each.
Audit rules: Every number traces to a cell in the KPI sheet — cite the metric name in the speaker note. Don't fabricate figures; if a KPI is missing, add a slide note flagging it.
Ask me to confirm the [Q2] date range before building.Why it works: A fixed slide count and section order plus "one idea per slide" produces a tight executive deck instead of a text-heavy sprawl.
5. Website build
Goal: Build a responsive one-page marketing website for [my new SaaS product], ready to preview and export.
Constraints: Brand [name] is [X]; primary color [#HEX]; tone [confident, plain-spoken]. Mobile-first. No stock-photo clichés. Use only the copy points I give below: [paste 5–8 bullet points].
Deliverable: A single responsive page with these sections in order — sticky nav, hero (headline + subhead + CTA), 3 feature cards, how-it-works (3 steps), social-proof strip, pricing (3 tiers), FAQ (5 Q&A), and footer with links. Include working anchor links and one primary CTA button repeated in nav, hero, and footer.
Audit rules: Use only the copy I provided — do not invent claims, testimonials, or pricing. Leave a clearly labeled placeholder where I haven't given content.
Show me the live preview when done.Why it works: Listing every section and forbidding invented copy gives you a real, on-brand page you can ship rather than a hallucinated one you have to gut.
6. Scheduled monitor
Goal: Set up a Scheduled Task that tracks [my 5 competitors] and appends changes to a running spreadsheet.
Schedule: Run every [Monday at 9am].
Constraints: Monitor these sources per competitor — pricing page, blog, and LinkedIn: [list URLs]. Only report changes since the last run; ignore cosmetic edits. Use the Browser Agent to load each live page.
Deliverable: Append rows to the existing sheet with columns — Date, Competitor, Change type (Pricing/Product/Content/Hiring), What changed, Source URL. If nothing changed for a competitor, add one row noting "no change".
Audit rules: One source URL per row. Never report a change you can't point to on the live page. Tag anything ambiguous [Low confidence].
Email me a one-paragraph summary after each run.Why it works: Scheduled Tasks plus the Browser Agent turn competitive tracking into a hands-off weekly digest, and the "changes since last run" rule keeps the sheet signal, not noise.
Frequently Asked Questions
What makes a good Manus prompt?
A good Manus prompt is a delegation brief, not a one-liner. It has four parts: a clear goal or role, hard constraints (scope, timeframe, budget, sources, and any decision weights), the exact deliverable format (spreadsheet columns, slide count, report sections), and audit rules (cite every claim with a source URL, leave blank rather than guess, tag uncertain claims with a confidence level, add a sources appendix). Manus works autonomously across many steps, so the more you pin down up front, the less it wanders.
Why do I need to add constraints if Manus is autonomous?
Because autonomy compounds ambiguity. Manus plans and executes dozens of steps on its own, so any vague instruction early on multiplies into wasted work later. Giving it a scope, a timeframe, a budget, decision weights or thresholds, and a stop condition (for example, stop when you have 50 verified rows) keeps every sub-agent pointed at the same target.
How do I make Manus output verifiable?
Add audit rules to the prompt. Tell Manus to cite every claim with a source URL, to leave a field blank rather than guess, to tag uncertain claims with a confidence level of High, Medium, or Low, and to include a sources appendix listing every URL it used. This turns an autonomous run into an output you can spot-check instead of trust blindly.
When should I use Wide Research?
Use Wide Research whenever your task is a big list — companies, products, papers, keywords, or contacts — usually 50 or more items. Instead of working the list one item at a time, Manus launches many sub-agents in parallel, so 100 companies get researched roughly simultaneously. Name it explicitly in the prompt and pair it with an exact spreadsheet schema.
Should I tell Manus to ask clarifying questions first?
Yes, when scope is fuzzy or the task is expensive. Adding "ask me clarifying questions before you start" makes Manus surface assumptions up front rather than committing a long autonomous run to the wrong interpretation. For tightly specified briefs where you have already given scope, format, and audit rules, you can skip it and let it run.
What is the difference between Manus 1.6 Lite, 1.6, and 1.6 Max?
They are effort tiers. 1.6 Lite is fast and cheap for simple, well-scoped jobs; 1.6 is the balanced default; and 1.6 Max applies the deepest reasoning for complex research, analysis, and multi-step planning. Use Lite for a quick lead pull or formatting task, and Max when the quality of the plan itself matters.
How do I get Manus to build a spreadsheet, deck, or website instead of a report?
Name the deliverable tool and its exact shape in the prompt. Say "build a spreadsheet with columns A, B, C," "create a 12-slide deck in Manus Slides with this section order," or "build a responsive one-page website with these sections." Manus routes the task to the right builder, so specifying the format and its structure is what determines what you get back.