Perplexity Search, Pro Search, Research and Model Council

#01What does each mode actually do?

Perplexity says Research is also rolling out as a Computer skill. You reach it from the + menu or by typing "/" in the input. Older help articles and screenshots still say "Deep Research".

#02How do you choose quickly?

Ask yourself three questions:

1. Is the answer one fact? Use Search.
2. Does it need comparing or combining several sources? Use Pro Search.
3. Will someone make a decision or publish based on it? Use Research, and use Model Council if the claim is contentious.

Our view: most teams overuse Research for things Pro Search handles in 30 seconds, then underuse it for the decisions that matter. Use Research for anything that ends up in a deck.

#03How do you write a good Research prompt?

Research works best with a brief, not a question. Use this template:

Goal: [decision this research supports]
Scope: [market, region, segment, time window]
Questions to answer:
1. ...
2. ...
3. ...
Sources: prefer [vendor docs, analyst reports, filings, reviews]; avoid [listicles, affiliate sites].
Output: [table + 300-word summary + list of open questions]
Date rules: only use information published or updated after [date]; flag anything older.

Example for a positioning check:

Goal: decide whether to lead our Q1 campaign with "AP automation for multi-entity finance teams".
Scope: mid-market (200–2,000 employees), US and UK, last 12 months.
Questions: Which AP vendors already claim multi-entity support? How do they describe it? What do reviewers complain about?
Sources: prefer vendor product docs, G2/Capterra review text, finance community threads; avoid affiliate "best of" lists.
Output: comparison table (vendor, claim, evidence URL, date), summary, gaps.

#04How should you check a Research report?

Research reports read well, which is exactly why you need to check them. Do these steps every time:

1. Open the sources list first. Count the source types. If more than half are vendor blogs or affiliate lists, rerun with source rules.
2. Spot-check three claims. Pick the most surprising number, the most important claim and one random claim. Open the cited page and find the sentence.
3. Check dates. Pricing and feature claims older than six months need verifying.
4. Look for false precision. Market sizes without a named source, and percentages without a sample size, should be removed.
5. Note what's missing. Ask: "Which relevant sources did you not find or could not access?"

#05When is Model Council worth it?

Model Council sends one question to several models at once and gives you a combined answer that marks where they agree, where they disagree, and what only one model found. Good marketing uses include:

• Positioning claims. "Is it accurate to say we are the only AP tool with X?" Disagreement is a warning sign.
• Regulatory or compliance copy. E-invoicing mandates, data residency and accessibility claims.
• Pricing comparisons. Models often disagree about competitor prices that have changed recently.
• Strategy questions. "What are the strongest counterarguments to this campaign thesis?"

Don't use it for drafting copy or simple lookups, because the cost and time aren't worth it.

#06What about choosing a model?

Paid plans let you choose a model per query in Pro Search. The model list changes often, so we don't name a "best" one here. A practical rule:

• Keep the default for most searches. Perplexity tunes it for retrieval.
• Try a reasoning model for multi-step logic (for example, building a pricing comparison from several pages).
• Use Model Council instead of switching models by hand when you want a cross-check.

Record which model and mode produced any number you publish, so you can reproduce it.

#07How do you export and reuse outputs?

• Export. Research reports can be exported (for example to PDF or a document) from the answer menu.
• Share. Share a thread link with teammates, or keep work in a shared Project (see lesson 8).
• Pages. Perplexity has temporarily retired "Create page". Perplexity says "Convert to Page" is returning. Don't build a process around Pages until it's back.
• Follow-ups. Ask follow-ups in the same thread. Context carries over, so you don't need to re-explain your scope.

#08Worked example: Ledgerline

Ledgerline is a fictional accounts-payable software company. Details are illustrative.

Ledgerline needed three outputs in one week: a definition for a glossary page, a pricing comparison against Competitors A-D, and a market scan of e-invoicing mandates in Europe for a webinar.

Walkthrough

1. Search: copied the definition sources into the glossary brief and wrote original copy.
2. Pro Search: "Compare published pricing tiers for Competitors A, B, C and D as of this month. Table with tier name, price, billing basis, source URL, date checked."
3. Model Council: pasted the table and asked "Which rows are disputed or likely out of date?" Competitor B's row came back disputed. The team checked the live page by hand and found a new tier.
4. Research: used the template above with "Only use government or official sources for mandate dates." Exported the report and spot-checked five dates.
5. Logged mode, date and source URLs in the research log.

#09Common mistakes

• Using Research for definitions. It wastes time and limits.
• Trusting the summary over the sources. Always open the sources.
• Publishing a number without the source. Every number needs a URL and a date.
• Ignoring disagreement. When Model Council shows a split, check by hand.

#10Try it yourself

Take one question from your list of ten. Run it in Search, Pro Search and Research. For each one, record the time taken, the number of sources, the share of sources you'd trust, and one wrong or outdated claim if you find one. Decide which mode you'll use by default for this type of question.

#11Quick quiz

1. Which mode would you use for a competitor pricing table that will go into a sales deck?

Answer: Pro Search to build it, and Model Council or a manual check to verify it before use.

2. What is Deep Research called now?

Answer: Research. It is also available as a skill inside Computer.

3. What should you open before reading any report?

Answer: The sources list.

#12What should you check before the next lesson?

• You have a default mode for each of your common question types.
• You have saved the Research prompt template.
• You have added a "Claims to verify" section to your template.

Previous: Perplexity plans and pricing. Next: Prompting and source control. Or return to the Perplexity for Marketers track. Related: Competitor research with Perplexity, ChatGPT deep research for competitive intel.

#13Common questions

What happened to Deep Research? Perplexity now calls it Research, and it is also rolling out as a skill inside Computer alongside Model Council. The idea is the same, with many searches and a written report.

Is Model Council worth it for marketers? It is useful for high-stakes questions such as positioning or pricing claims, where you want to see where different models disagree. For routine lookups it is overkill.

How long does a Research report take? Usually a few minutes. Perplexity runs dozens of searches and reads many sources before writing the report.