Run Perplexity research in a fixed order. Scope the question so one canonical source can answer it, pick the mode for the job, and open the page behind every figure before it leaves the thread. Independent audits found that citations fail most often on facts spread in pieces across several pages.
When you need this
- Perplexity findings are reaching decks and decision documents with nobody opening the sources.
- Two people ask the same question and get answers with different figures.
- A buying or market recommendation rests on pages nobody on the team has heard of.
The playbook
Before research: scope the question
Does the question name the company, market, or period, and the figure you need?
In practiceAsk a specific question in a Project. Set preferred domains and standing instructions of up to 8,000 characters. Haus Research found citations strongest for facts published in one canonical place. Headcount passed in 82.0% of citation pairs. The current CEO passed in 44.3%, where sources did not publish the fact in full in one place.
Choose the mode for the job
Is this a quick fact, a report, or a decision that needs a second model's answer?
In practiceUse Search with Best for quick facts. Use Research for reports that need dozens of searches across hundreds of sources. Research picks its models itself. On Max and Enterprise Max, Model Council asks three models the same question and shows where they disagree.
During research: open the cited sources
Does the cited page say what the sentence says?
In practiceTrace every figure leaving the thread to its page. Haus Research found that 34.7% of 1,826 Sonar API citations led to pages that would not open or held none of the sentence's figures. A finance staffer at connectome.design also found figures for two brewers that did not match securities filings.
Weigh the source behind the label
Is the source a primary document, an independent report, or a page written to be cited?
In practiceTreat Government, Academic, and Trusted labels as a first signal. Perplexity says each label covers a whole site and does not replace reading the source. Trellner found that 59.8% of 7,534 Sonar citations for software recommendations pointed outside the 100,000 most-visited domains.
Before sharing: record what was checked
Can the reader see which claims were verified, and against what?
In practicePut the primary source beside every figure in the deliverable. Mark any figure that nobody has checked against its page.
Keep company data in its lane
Does the question carry client, personal, or regulated data?
In practiceKeep company files in Projects or admin-approved connectors and follow the company's data classes. Texas A&M approves Perplexity for public, university-internal, and FERPA data, requires approval before HIPAA data, and prohibits all other confidential data.
Questions to settle with each team
- Which of our recurring research questions concern facts published in one canonical place, and which need a primary source every time?
- Who signs off a figure that came from Perplexity before it goes to a client or the board?
- Which Projects should carry standing instructions and preferred domains for each team?
- When is a single answer not enough for a decision, so it should go to Model Council or a second person?
Red flags
- A figure in a deck whose only source is a Perplexity thread.
- Citations to pages behind a login or paywall that nobody on the team can open.
- Machine-generated 'best software' lists as the evidence for a buying recommendation.
- Deep Research reports forwarded without anyone opening their sources.
Products in this piece
Sources
- Official documentation
- 6
- Public source
- 4
- Perplexity Help Center: What is Research mode?
- Perplexity Help Center: What's New in Advanced Deep Research
- Perplexity Help Center: What is Model Council?
- Perplexity Help Center: Understanding source labels
- Perplexity Help Center: What are Projects?
- Perplexity Help Center: What advanced AI models are included in my subscription?
- Haus Research: A third of Perplexity's citations don't contain the number they're cited for (Sep 2, 2026)
- Trellner: Three sites made 215,128 "best software" pages for AI. Perplexity cites them (Sep 2, 2026)
- Suganthan Mohanadasan: How Perplexity Actually Picks Sources (Jul 27, 2026)
- Texas A&M Technology Services: Perplexity Enterprise Pro FAQ (Sep 28, 2026)