Search Handles Answers; Computer Executes Connected Work
Perplexity Academy draws a practical line between its Search and Computer tools: Search is for questions where a cited answer is the finished product, while Computer is for work that must move across systems, produce an artifact, or run repeatedly. The company says Search can handle complex analysis—from document summaries to valuations—so long as the result is bounded and verifiable; Computer is intended to connect data and tools, build deliverables such as dashboards or reports, and send scheduled updates.
The dividing line is whether an answer is the work
Search is for understanding; Computer is for execution. The useful distinction is not task complexity. Search can handle a sophisticated, multi-source question, but it is designed for situations in which the answer itself is the deliverable. Computer is for work that crosses steps, systems, or tools—or must be assembled, delivered, monitored, or repeated.
| Mode | Best when | Typical output |
|---|---|---|
| Search | The context fits in a query or attached file, and the answer is the deliverable | A cited summary, comparison, or analysis |
| Computer | The work crosses steps, systems, or tools, or must recur | A report, spreadsheet, dashboard, deck, website, or delivered update |
Search is positioned for summarizing a document, comparing a short list of options, or exploring a topic before deciding what to do next. A customer-call transcript can be attached and summarized; Kelvara, Northlane, and Orbition can be compared on security, cost, and implementation effort. The displayed comparison extends across primary strength, automation, analytics, collaboration, integrations, and best fit.
Nor is Search limited to simple lookup. One example asks for a three- to five-year discounted-cash-flow valuation of NVIDIA, compared with AMD and Broadcom using live market data. The displayed output includes a table of valuation assumptions and inputs: current price, market capitalization, annualized free cash flow, five-year growth, WACC, terminal growth, and net debt or cash.
The constraint is therefore the endpoint, not the amount of reasoning involved. A valuation, summary, or vendor comparison remains a Search task if a bounded answer resolves the request. The rule is: start in Search to clarify the thinking; move to Computer when the work needs to be completed, repeated, packaged, or delivered.
Search makes the evidence behind an answer visible
Search returns cited results from the web, premium sources included in a subscription, and connected knowledge. Its citations are part of the working interface rather than a trailing bibliography: users can hover over a citation attached to a claim, inspect the source preview, and open the underlying page.
In the displayed NVIDIA valuation example, a citation preview links to a page titled “NVIDIA Corp NVDA DCF Valuation - Growth Exit 5Y.” Facts and figures in answers are presented as backed by clickable citations, allowing a user to inspect where a particular claim came from.
That verification layer fits the Search role. It supports summarizing, comparing, and investigating while leaving the user with an answer that can be checked claim by claim.
Computer turns research across systems into deliverable work
Computer is described as a multi-model orchestrator for work spanning files, tools, systems, and the open web. It can research across those inputs and build reports, spreadsheets, decks, dashboards, websites, and other deliverables.
The additional capability is access and execution across connected services. The listed connectors include Google Drive, Slack, Salesforce, GitHub, and Snowflake, alongside more than 400 other tools.
One example begins with 30 Salesforce accounts. Computer is asked to research them, enrich them with product-usage and billing data, identify concrete expansion signals, and create a dashboard ranking the accounts by expansion potential. The displayed dashboard includes ranked accounts, a top-10 expansion-score view, and a chart comparing score with ARR, with bubble size representing opportunity.
This is the contrast with the NVIDIA example. Both requests require research and analysis. But the account request also requires data from named systems, enrichment, transformation into a dashboard, and a ranking artifact that can be used elsewhere.
Another example is explicitly recurring: monitor competitor pricing every week and summarize material changes in a go-to-market Slack channel. The shown Slack update reports five material changes across eight tracked competitors for the week. Computer can run on a schedule, monitor changes, and send proactive alerts; the interface also illustrates a daily update on U.S. state privacy-law developments and enforcement actions.
The same pattern applies to internal product feedback. Computer can pull material from Intercom support tickets, customer-success notes in Google Docs, and user surveys; turn it into a prioritized roadmap grouped by theme; and deliver both a PDF report and spreadsheet. The shown spreadsheet organizes themes alongside mentions, enterprise priority, revenue at risk, expansion, effort, and target quarter. Outputs can be downloaded or exported, including to Google Drive.
The handoff begins with a question and ends with an operating workflow
The proposed sequence starts with a bounded question in Search, then moves to Computer when the result needs to become an operating workflow.
The displayed example makes the split concrete. Search is asked to summarize key portfolio-company risks from an uploaded Q3 board deck. The summary is the endpoint: a contained analysis of a supplied document. Computer is assigned a different job—monitoring all 24 portfolio companies every Monday for material changes. That task has a schedule, portfolio-wide scope, and ongoing output.
The same division applies elsewhere. A cited vendor comparison can stop in Search; a ranking dashboard built from Salesforce, product usage, and billing data belongs in Computer. A pricing question can yield a one-time answer; weekly tracking and a Slack update require a recurring workflow.
Both modes run on desktop, mobile, and in voice mode. The intended progression is from asking a question, to getting a supported answer, to completing and delivering work where a team needs it.