Manus SEO: the technical guide to being used and cited by an AI agent

How Manus researches a task#

Manus describes itself as “a virtual colleague with its own computer,” and that’s literal. Every task gets a fully isolated cloud virtual machine with networking, a file system and a browser. By December 2025 Manus had created more than 80 million of them. The output isn’t a chat reply you might click through from — it’s a report, slide deck, spreadsheet or website, and your page is either a numbered reference in it or absent.

That changes the question you’re optimizing for. With ChatGPT or Claude the question is whether a crawler can read your HTML. With Manus it’s whether an agent can finish its sub-task on your page: find the price, open the plan table, read the limits, and move on.

1. 1

   It turns the task into a plan

   Manus breaks a request into steps and keeps rewriting a to-do list as it works — “reciting its objectives into the end of the context,” per its engineering write-up (July 2025). A typical task takes around 50 tool calls.

   Your lever: Answer the sub-questions a plan will contain — pricing, limits, integrations, alternatives — each on a page of its own.

2. 2

   It searches with a tool, one attribute at a time

   Manus doesn’t document which search engine backs its search tool. Its system prompt, as extracted in March 2025, told it to prefer that tool over search-results pages and to “search multiple attributes of single entity separately.” Its task view still tallies them, in steps labeled like “Searched 6 queries.”

   Your lever: Make each attribute findable on its own: a pricing page, a limits page, an integrations page, each titled for the query that would find it.

3. 3

   It opens pages in a cloud Chromium browser

   The browser runs inside the task’s VM — E2B, whose microVMs Manus used in 2025, lists Chromium among its tools. It works from “data center IP addresses, not residential IPs,” identifies as Manus-User, and signs its requests.

   Your lever: Don’t challenge it. A CAPTCHA stops the task until the user takes over the browser by hand.

4. 4

   It reads what the browser extracts

   The 2025 prompt described page content extracted “in Markdown format” that “omits links and images,” alongside the interactive elements in the visible viewport. Search snippets “are not valid sources” — it must open the page. Once used, page content can be dropped from context “as long as the URL is preserved.”

   Your lever: Keep the fact in text, near the top, on a stable URL — where a screenshot, an extraction and a later revisit all find it.

5. 5

   It writes the deliverable and lists what it used

   The prompt told Manus to “provide a reference list with URLs at the end.” In the 2026 app, references show as small numbered pills; hovering one opens a card with the source’s title and hostname, and the list at the end reads [1] Title.

   Your lever: Give pages a <title> that says exactly what they are. It’s the most likely text for the reference line.

How Manus shows up at your server#

Manus reaches your site three ways, and only one of them is distinguishable from a person. Most tasks use the cloud browser. Wide Research runs many of those at once. Browser Operator is an extension that drives the user’s own Chrome or Edge.

Manus hasn’t published its user agent string; its help page for the bot, linked from Cloudflare’s directory, returned a 404 in September 2026. Known Agents records it as Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/151.0.0.0 Safari/537.36; Manus-User/1.0 — the Chrome version moves with each release, so match on the token.

Because any client can send that string, the identity that matters is the signature. Cloudflare Radar’s verified-bot directory (mirrored as open data) lists Manus Bot in the AI Assistant category, verified by Web Bot Auth — HTTP message signatures checked against a public key directory. Fingerprint (March 2026) likewise names Manus among operators “signing their requests today.” The directory is public:

# Manus's Web Bot Auth key directory (as served in September 2026)
curl -s https://api.manus.im/.well-known/http-message-signatures-directory
{"keys":[{"crv":"Ed25519","kty":"OKP","x":"KcGM6sqzNes25V0hn96KsrIwL6pbut1KQMGmyFwpboM"}]}
 
# The response is itself signed (tag="http-message-signatures-directory").
# Keys rotate: verify against the live directory, never a hard-coded key.

robots.txt still has a use: stating your policy. Just don’t expect it to enforce anything. Known Agents says a disallow “only communicates your preference,” and reports that 9% of top websites block Manus-User there (September 2026). If you want Manus in, say so and keep private paths out:

# Manus's cloud browser. Cloudflare lists it as not following
# robots.txt: this states your policy, it doesn't enforce it.
User-agent: Manus-User
Allow: /
Disallow: /account/
Disallow: /checkout/

Agent-readiness: pages an agent can use#

Manus hasn’t published guidance for site owners, so the clearest spec comes from Google. Its web.dev guide to agent-friendly sites (April 2026) says: “Prefer <button> and <a> tags over modified <div> and <span> elements,” link labels to inputs, keep layouts stable, and “avoid ‘ghost’ elements or transparent overlays.” Manus confirmed in 2025 that it used Browser Use, open-source software that extracts a page’s buttons and widgets for the model — the same accessibility-first reading.

Reachable without a challengeRequired

The cloud browser runs from data-center IPs. CAPTCHAs, interstitial checks, 403s and 429s stop the task until the user takes over. Allow the verified bot.

Facts in visible textHelps

Manus’s extraction “omits links and images.” Prices in a graphic, specs only in a PDF, or copy drawn on a canvas may never reach the report.

Real links, buttons and labelsHelps

Native <a href>, <button> and labeled form fields are what DOM-reading agents detect reliably. Manus’s 2025 prompt admits that “not all interactive elements may be identified.”

No overlays or interstitialsHelps

Cookie walls, newsletter modals and app-install banners cost the agent steps, and transparent overlays can hide the control it needs.

Stable layoutHelps

Google: “Agents that take screenshots will likely be confused if your website layout is constantly shifting.” Reserve space for late-loading elements.

Stable, specific URLsHelps

Manus keeps the URL when it drops a page from context, then revisits it. A plan, filter or tab that has its own URL can be reopened and cited; a state reached only by clicks can’t.

Server-side renderingHelps

Not required for Manus, whose browser runs JavaScript. It still makes the first extraction complete — and ChatGPT, Claude and Perplexity don’t render at all. See server-side rendering.

Structured dataUnconfirmed

Manus hasn’t said whether it reads JSON-LD. Extraction works on visible content, so keep every fact from your schema markup in the copy too.

Markdown for agentsUnconfirmed

Its browser is Chrome, which asks for HTML. Whether any other Manus fetch path sends Accept: text/markdown isn’t documented.

llms.txtUnconfirmed

Manus publishes an llms.txt for its own docs, but hasn’t said its agent reads other sites’ files.

robots.txt rulesNo effect

Cloudflare’s directory lists Manus-User as not following robots.txt. Use it to state policy; enforce policy at the edge.

The quickest check is whether your edge turns the user agent away. It can’t reproduce Manus’s IP reputation or its signature, so a pass here is necessary rather than sufficient:

# Does a UA or bot rule turn Manus away?
curl -s -o /dev/null -w "%{http_code}\n" \
  -A "Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/151.0.0.0 Safari/537.36; Manus-User/1.0" \
  https://yoursite.com/pricing
 
# 403, 429 or 503 = your edge blocks agents by user agent.
# 200 doesn't prove the real agent gets in: IP and signature
# checks only apply to real Manus traffic.

What an agent needs from a pricing page

• Every plan’s name, price, currency and billing period as text, in a real HTML <table>.
• What each plan includes and its limits — seats, usage caps, overage rates — on the same page, not behind tabs that change nothing in the URL.
• A visible ”last updated” date, because agents run the same comparison again next month.
• Links to docs, changelog and comparison pages as plain <a href> elements.
• No “contact sales” wall in front of numbers you are willing to publish. An agent can’t book the call.

What Manus uses and cites#

There’s no public study of which domains Manus cites as of September 2026, so treat citation statistics from ChatGPT or Perplexity as irrelevant here. What is known comes from Manus’s docs, from its system prompt as extracted in March 2025 — whose tools and architecture co-founder Yichao “Peak” Ji confirmed at the time — and from how its app renders references. Manus says it has “rebuilt our agent framework four times” (July 2025), so read the prompt-based signals as direction, not current rules.

• Pages it opened, not snippets

  Observed

  The prompt: “Snippets in search results are not valid sources; must access original pages via browser.” A page that blocks or stalls the browser can’t be the source, however well it ranks.

• Cross-checked facts

  Observed

  Manus was told to “access multiple URLs from search results for comprehensive information or cross-validation.” When your site, review sites and profiles disagree on a price, the agent has to pick — and may not pick you.

• Data providers before the open web

  Official

  Manus has built-in data sources for company profiles, financial data and SEC filings, and its prompt ranked “datasource API > web search > model’s internal knowledge.” Your profiles on those providers may be read before your site.

• Text the extraction can see

  Observed

  Extracted page text leaves out images and links, and only viewport elements are listed for interaction. Facts that live in visible copy near the top travel best.

• Pages that load for a data-center browser

  Official

  Manus documents that its cloud browser’s data-center IPs can trigger verification. Sites that challenge them lose the visit unless the user steps in.

• Sources inside the user's subscriptions

  Official

  With Browser Operator, Manus reads Crunchbase, PitchBook, SimilarWeb, the Financial Times, Semrush and Ahrefs through the user’s own logins. Your presence on those platforms is part of your footprint.

• One complete page per item

  Our read

  Wide Research gives each item — a product, a company, a location — its own agent. A dedicated page that covers the item completely is what that agent lands on and quotes.

• Titles that name the page

  Our read

  Each reference is listed as [n] plus a title, with your hostname on the hover card. A precise <title> is the likeliest text for that line.

Being read isn’t the same as being cited. Manus opens far more pages than it references, so the pages that earn the reference are the ones that supplied a fact the deliverable actually uses — a price, a limit, a date — stated plainly enough to copy into a table.

Wide Research, Browser Operator and scheduled tasks#

Wide Research: hundreds of agents, one item each

Wide Research launched on 31 July 2025, for Pro users first. It deploys “hundreds of independent agents that work in parallel,” each “a fully capable, general-purpose Manus instance” with its own context. Manus says it has been tested up to 250 items and takes “minutes for 50-100 items.” Its own examples are comparisons: 100 sneaker models, 250 AI researchers, Fortune 500 companies.

• Expect bursts. A comparison that includes you can send many agents to your site within minutes. Rate limits keyed on IP bursts can turn most of them away.
• Publish one page per entity. Each subagent researches one item; a complete page for each product, plan or location gives it one place to finish.
• Keep facts table-ready. Results are assembled into tables and decks. Consistent plan names, units and currencies across your pages survive that merge.

Browser Operator: the agent in your visitor's browser

Browser Operator is an extension for Chrome and Edge that runs Manus in a dedicated tab, grouped under the task’s name, in the user’s own browser. Manus says it has been available to all users since 22 November 2025, though its docs still describe a beta for paid plans. Because it uses the user’s sessions and local IP, “there are no unfamiliar login attempts. No CAPTCHA interruptions.”

For you, that traffic is your visitor. It carries their cookies, their IP and their browser’s user agent, so you can’t see it separately or block it without blocking them. It also reaches logged-in areas: if your customers ask Manus to pull data from your app or gated docs, it works there with their permissions.

Scheduled tasks: agents that come back

Manus can run tasks on a schedule — daily, weekly, monthly or custom. Its documented examples include “weekly checks for product updates, pricing changes, and new blog posts” and competitor price tracking. Your pricing page, changelog and blog index are the pages those tasks revisit; a visible date on each tells the agent what changed.

Measuring Manus traffic#

Your server logs are the reliable view. Manus-User hits show which pages tasks open, and weekly spikes on the same URLs suggest scheduled monitoring. Known Agents warns that “any bot can claim to be Manus-User,” so for decisions, rely on your CDN’s verified-bot flag — which checks the signature — rather than the string.

# Pages Manus's cloud browser opened, most-visited first
grep "Manus-User" access.log | awk '{print $7}' | sort | uniq -c | sort -rn | head -20
 
# Hits per day: the same pages every week suggests scheduled tasks
grep "Manus-User" access.log | awk '{print substr($4,2,11)}' | sort | uniq -c

Analytics sees less than you’d hope. As shipped in September 2026, Manus’s web app opens links in a task — in-line references, the reference list, plain links — with rel="noopener noreferrer", so the browser sends no referrer and those visits land in Direct. Websites Manus builds are published on manus.space subdomains, with sandbox previews on manus.computer; clicks from those carry their origin. Google’s AI Assistant channel names ChatGPT, Gemini, DeepSeek, Copilot and Grok — not Manus — so give it a custom channel:

# Session source — pages Manus built or hosts, linking to you
(^|\.)manus\.(space|computer|im)$
 
# Clicks from inside a Manus task carry no referrer: they land in Direct

The agent’s own visits can pollute analytics too. The cloud browser runs JavaScript, so your analytics tags may fire for Manus-User visits and record them as sessions. Compare log hits with GA4 sessions on your most-visited pages before trusting engagement numbers there.

Citations need a panel of your own. Run a fixed set of buyer tasks in Manus each month — the comparisons your customers would delegate — and record whether you appear in the reference list, and which of your pages it used. Tasks cost credits, so keep the set small and identical from month to month. See AI referral traffic for the wider picture.

Myths worth dropping#

The action checklist#

Everything above, in the order we’d do it. Tick items off as you go — your progress is saved in this browser.

Manus SEO: frequently asked questions#