Prompting Perplexity Labs for Deliverables

TL;DR: Perplexity Labs promised reports, spreadsheets, dashboards and web apps from a single prompt when it launched in May 2025. It has zero mentions across Perplexity's 207 current help-centre articles, and the /labs URL now redirects to Computer. This guide covers what Computer, Deep Research and Model Council document today for deliverable-shaped output, with 28 copy-paste research briefs organized by what you're actually trying to produce.

If you searched for Perplexity Labs, you were probably looking for a way to turn a question into something more than an answer: a report, a deck, a spreadsheet, a chart you didn't have to build by hand. That capability is real. The name you searched for is not what Perplexity calls it anymore, and in one important sense it isn't documented at all.

Does Perplexity Labs still exist?

Not under that name. Perplexity's English help centre publishes 207 articles, enumerated in its own sitemap, and as of September 3, 2026 not one of them contains the word Labs. That is a fresh check, run the same way a prior post on this site checked Focus modes: pull the sitemap, search every URL, find nothing.

The stronger signal is the redirect. perplexity.ai/labs returns a live page today, but it forwards to perplexity.ai/products/computer, whose title tag reads "Computer | Perplexity AI" and whose meta description calls it "an agentic AI assistant that uses your computer to complete tasks for you, from booking travel to filling out forms." The word Labs does not appear anywhere on the page it lands you on.

Labs was real, and its own announcement is still live at perplexity.ai/hub/blog/introducing-perplexity-labs, dated May 29, 2025. Its pitch, in Perplexity's words: "Labs can craft everything from reports and spreadsheets to dashboards and simple web apps — all backed by extensive research and analysis." Pro subscribers could reach it, per that post, "from the mode selector in the input bar on Web, iOS, Android" — Mac and Windows access was listed as "coming soon". The same post used the moment to rename Deep Research to Research, a rename Perplexity has kept. Fifteen months later, the rename habit caught up with Labs itself.

I'm not going to describe a Labs mode selector, its steps, or its current limits, because no current Perplexity documentation describes them either. What I can do is show you where the capability that made Labs launch-worthy actually lives now, sourced from articles Perplexity is actively updating.

What replaced Perplexity Labs?

Computer. Perplexity's help centre defines it plainly: "Computer is an independent digital worker that completes tasks and workflows for you." Among the capabilities it lists is "Document creation for reports and presentations" — Labs' pitch, in a currently maintained article last updated July 16, 2026.

The mechanism is named now, where Labs only ever said it ran on "a suite of tools". Computer runs on Skills: "Skills are more than reusable instruction sets — they deploy agents." A skill hands Computer a methodology for one type of task, and Computer dispatches dedicated sub-agents, often several in parallel, to carry it out and report back. The built-in set includes a Slides skill, a Research skill, a Chart skill, and a Research Report skill that formats findings once the research is done, and Perplexity's own examples show how they activate: ask to "create a presentation about Q1 results" and the Slides skill activates; request "a detailed analysis of renewable energy trends" and the Research skill kicks in; say "show me a chart of monthly revenue" and the Chart skill handles visualization.

Access and cost are documented in the same article. Computer needs an active subscription, and it is not on the Free plan: the model-availability article lists "Computer models are available on the Pro, Max, Enterprise Pro, and Enterprise Max plans, provided you have credits available — none of them are Max-only." Credits work like this: Max subscribers get 10,000 a month (plus a stated limited-time bonus of 35,000), Pro subscribers get 4,000, and ordinary Search stays unlimited and outside the credit system entirely. Run out mid-task and the task pauses rather than cancels, resuming once your balance refills.

Why does a deliverable-shaped prompt need to name the format first?

Because the format is what selects the skill. Computer does not infer that you want a slide deck from a topic alone. It reads the noun in your sentence. Ask for "a detailed analysis" and the Research skill runs. Ask for "a presentation" and the Slides skill takes over instead, often after the same research. Say nothing about shape and you get Computer's default: a written answer, not a file.

This is the one place where the old prompting habit of leading with a persona is actively counterproductive, more so than in plain Search — our guide to search-shaped prompting covers why a persona preamble is dead weight even there. Perplexity's own worked example shows the whole shape in one sentence: "Research our top 5 competitors, compare their pricing, create a slide deck with the findings, and email it to the team." That sentence names the topic, the comparison axis, the deliverable, and the delivery channel, in that order, and Perplexity's article confirms what happens next: "Computer will run parallel web searches, build a comparison table, generate a branded PPTX, and send it via your connected Gmail — all in one session." Four skills, one prompt, because the prompt named four things.

The general shape, adapted from that example: [research the topic] → [state the comparison or synthesis] → [name the deliverable] → [name the destination, if there is one]. Drop any of the last two and Computer has to guess, and a guess is where you get a chat answer instead of a file.

Deep Research or Computer: which do you brief for a report?

Both produce a report. They are not the same tool, and briefing the wrong one wastes the wait.

If the job ends at "give me a report I can read," Deep Research is the cheaper brief: no credits, no orchestrator model to pick, and it already streams into a file you can edit. If the job needs a second step, a deck built from that research, a sheet someone else can filter, an email that goes out on its own, Computer is the one documented to chain those steps in a single session. Model Council sits alongside both of these for a narrower job: as a standalone mode it is Max and Enterprise Max only, but Perplexity's own article states plainly that "In Computer, Model Council runs as a skill and is available in all paid plans" — so a Pro subscriber who wants three models cross-checking a decision can get it by asking inside Computer, without upgrading for the standalone mode. For the fuller mode-by-mode breakdown, including Search, Learn Mode and Comet, our mode-selection guide covers the rest of the selector. This post assumes you already know which mode you're in and focuses on getting a specific artefact out of it.

Copy-paste research briefs, by deliverable

Every brief below follows the shape from the section above: topic, comparison or synthesis, deliverable, destination. Fill the brackets and paste the whole thing; Computer or Deep Research reads the deliverable noun and picks its own skill. If what you actually need is a deeper library of research-only prompts, without a file at the end, our Perplexity research prompt templates has 24 more built around literature scans, due diligence and fact-checks.

Prompts for a report you can hand to someone

1. Research [topic] for a decision about [decision]. Produce a report with an executive summary, sections covering [dimension 1], [dimension 2] and [dimension 3], and a closing section listing what you could not verify. Export it to PDF.

2. Build a due-diligence report on [company] covering ownership, funding, leadership and any litigation. Cite a primary source for every claim, and flag anything sourced from only one outlet.

3. Write a market-landscape report on [category] as of [month, year]. For each named competitor, cover what they sell, who they sell to, and their published pricing. Mark any field you could not source as "not published."

4. Summarize the current state of evidence on [research question] as a report with a findings section and a limitations section. Keep peer-reviewed sources separate from press coverage and preprints.

5. Cross-reference [uploaded document] against current public sources on [topic]. Report which of its claims are still accurate, which are outdated, and which you cannot check, in three separate lists.

Prompts for a slide deck

6. Research [topic] and build a 10-slide deck: a title slide, a one-slide summary, and eight content slides covering [dimension 1] through [dimension 4]. Cite every figure on the slide it appears on.

7. Build a competitor comparison deck for [entity] versus [entity]. One slide per competitor on [criterion 1] and [criterion 2], closing with a recommendation slide naming the two data points behind it.

8. Turn [report or notes] into a client-ready deck: an executive summary slide, three findings slides, and one next-steps slide. Match the tone to a [industry] client and keep jargon out of the summary slide.

9. Build a board-update deck covering [metric 1], [metric 2] and [metric 3] for [time period]. One chart per metric. Note in the speaker notes any figure you had to estimate.

10. Build a 6-slide onboarding deck explaining [product or process] to someone joining the team this week. End with a one-page glossary slide defining every term you used.

Prompts for a spreadsheet or comparison table

11. Build a spreadsheet comparing [entity A], [entity B] and [entity C] across [column 1], [column 2] and [column 3]. Leave a cell blank rather than estimate, and add a Source URL column.

12. Compile pricing for [number] [category] vendors into a spreadsheet: plan name, monthly price, annual price, and the page you found it on. Flag any price you could not confirm within the last 90 days.

13. Turn [list of items] into a scored spreadsheet against [criterion 1], [criterion 2] and [criterion 3], with a total column and the scoring rule written in a header note.

14. Build a table tracking [metric] for [entity] over the last [number] [quarters or years]. One row per period, a source column, and a final row showing the trend direction.

15. Compare [option A] and [option B] on [criterion 1], [criterion 2] and [criterion 3] in a table. Mark any cell you could not source as "not published," not a guess.

Prompts for a dashboard, chart or small web app

16. Build a dashboard tracking [metric 1] and [metric 2] for [entity], with one line chart and one bar chart. Add a filter for [dimension] if the underlying data supports it.

17. Create a simple web app that lets me [action — e.g. "compare two vendors side by side"] using the data you just researched on [topic].

18. Chart [metric] for [entity] over [time period], and add a second chart showing [related metric] for the same period. Label the source under each chart.

19. Build an interactive comparison tool for [number] [category] options, with a control for [criterion] and a results table that updates when it changes.

20. Turn [dataset or research] into a one-page dashboard with three charts and a two-sentence takeaway written under each one.

Prompts for a briefing note or a spec

21. Write a one-page briefing note on [topic] for someone who has 90 seconds: what happened, why it matters, what to watch next, and one source per claim.

22. Draft a spec for [feature or process]: problem statement, requirements, an explicit out-of-scope list, and open questions. Research how [comparable products] solved the same problem and cite what you found.

23. Write a decision memo comparing [option A] and [option B] for [decision]. State a recommendation, then list the two facts that would change it.

24. Prepare a briefing document on [company or person] covering background, recent activity and public statements, with every claim traceable to a dated source.

If these deliverables are going to a client rather than your own team, our agency deliverables prompt set covers the surrounding work these briefs don't: scoping the engagement, writing the cover note, and reporting back once the file is sent.

Prompts for a recurring or scheduled deliverable

25. Every [weekday] at [time], check [sources] for anything new about [topic]. If nothing material changed, send nothing. Otherwise send a briefing note with three bullets and the links.

26. Once a week, rebuild the [topic] comparison spreadsheet from current data and email it to [recipient]. Note at the top anything that changed since last week's version.

27. Monitor [entity]'s public announcements and rebuild the competitor deck whenever something material happens. Tell me what changed before you tell me what stayed the same.

28. On the first of each month, produce a one-page dashboard of [metric] and file it in [connected app]. Skip the email if nothing moved more than [threshold].

Can you save a deliverable format as a reusable skill?

Yes, and this is the part of Computer that has no equivalent in what Labs ever documented. Beyond the built-in Slides, Research and Chart skills, Perplexity lets you upload your own: "Upload a .zip file with a SKILL.md file at the root level, or upload a .md file directly". That file "must include a YAML frontmatter block with a name and description". Perplexity's help centre gives a worked example, its own wording, its own instructions — its page renders the whole thing flattened onto one unbroken line, so the line breaks below are restored to make it valid YAML and Markdown, not something Perplexity's article itself shows:

---
name: weekly-summary
description: Summarize the week's key events into a concise executive briefing. Use when asked for a weekly summary, week-in-review, or status update.
---

# Weekly Summary

## Instructions
1. Gather the most important events and updates from the past week
2. Organize by theme or priority
3. Write a concise summary with key takeaways
4. Include action items if applicable

The description field does the routing: Perplexity states the description field "is critical — Perplexity uses it to decide when to activate the skill, so include keywords and trigger phrases". If you run brief 27 above every week by hand, this is the fix. Adapt the same shape:

---
name: competitor-snapshot
description: Research a named competitor's recent public activity and rebuild a one-page comparison. Use when asked for a competitor snapshot, competitive check, or "what changed with [competitor]".
---

# Competitor Snapshot

## Instructions
1. Search for the named competitor's public announcements in the last 7 days
2. Compare pricing, feature list, and any stated positioning against our own
3. Flag anything that changed since the last snapshot, at the top
4. Build a one-page table: what changed, what stayed the same, source for each row

Once uploaded, mentioning "competitor snapshot" is enough to trigger it. That's the deliverable-shaped habit this whole post is arguing for, written once and reusable instead of retyped.

Does a citation mean the deliverable was checked?

No, and this matters more here than in a plain search answer, because a Computer deliverable can leave your hands before you've read it: brief 6 above ends with "email it to the team," not "show me a draft."

A citation records that Perplexity's retriever found and used a page. It says nothing about whether the sentence sitting above the citation matches what that page actually says. The saving grace, same as in a plain search, is that the links are right there and checking them costs two clicks, not a research protocol.

The scoping matters for anyone tempted to cite a single number as "how accurate Perplexity is." The Tow Center's Columbia Journalism Review study, published March 6, 2025, ran 1,600 queries across eight AI search tools on one task: given a verbatim news excerpt, identify its headline, publisher, date and URL. It found "Perplexity answering 37 percent of the queries incorrectly," and, separately, that ChatGPT "incorrectly identified 134 articles" out of its 200 test queries — a 67 percent error rate, worse than Perplexity's. Overall, the eight tools "provided incorrect answers to more than 60 percent of queries."

Before Computer sends a deck or a spreadsheet anywhere on your behalf, open the sources behind the two or three numbers you'd be embarrassed to have wrong. That habit is cheaper than the mistake.

What will date this post first?

In roughly this order:

1. The Skills catalog. Slides, Research, Chart and Research Report are what's documented today. Perplexity is actively adding skills; expect more by the time you read this.
2. The credit numbers. Perplexity labels the 35,000-credit Max bonus as limited-time, stacked on top of a base 10,000. Base amounts and the bonus can both move.
3. The /labs redirect itself. A URL that forwards to a product page today can be repointed or retired outright tomorrow; it is not a stable citation the way a help-centre article is.
4. Model names inside Computer and Deep Research, which already disagree with the plan model-availability table elsewhere in Perplexity's own docs. The selector in your account, not any article, is the tiebreaker.

Where does a prompt tool fit into a Computer-shaped workflow?

Nowhere in the execution, honestly. Prompt Architects doesn't research the web, deploy sub-agents, or build the deck. Perplexity's Computer does that, end to end, once you've named the deliverable.

What's worth saving is the brief itself. A competitor-snapshot prompt, a due-diligence report prompt, a weekly-digest prompt: each one is worth writing precisely once, with the entity, date window and output format held as variables, rather than retyped with small differences each time you need it. That's a narrower job than a Computer skill, and a faster one to set up when you just need the wording right, not a deployed sub-agent.

One disclosure worth making plainly: our extension is listed on our integrations page as available and working on perplexity.ai. Checking our own FAQ while writing this, two of its answers list Perplexity as a currently supported model, and a third calls it an "upcoming" integration, in the same document. That's a real contradiction on our side, not a Perplexity one, and the integrations page is the one I'd trust. We should fix the FAQ.

The free plan includes 5 prompt enhancements per day, forever, per that same FAQ page, which is enough to test whether a saved research brief beats retyping one before you commit to anything paid.