AIAP Tool Guides·Perplexity·15:59

Perplexity Deep Research, Citations and the Model Selector: Three Search Tiers, Back-Checking Every Line, and Choosing a Model

Karan Rana Brand/marketing manager, YouTuber, podcaster and course creator ("a non-tech guy"); presents the AI for Marketing, NotebookLM and Manus tool guides

▶LMS lesson

The short version

Three tiers: plain answer, web search for live data, and Deep Research for reports with more sources and reasoning (0:00-0:05).
Ten lines for a PhD literature review on nuclear development and geopolitics: 18 sources and three ways to back-check each line (0:05-0:09).
Deep Research on the same topic took minutes and explored tangents; the model selector lists GPT-4.1, Gemini 2.5 Pro, Grok 3 Beta, Claude Sonnet 4, o3 and "Best" (0:09-0:15).

At a glance, three clicks deep

Skim here first: the closed row is the glance, open is the study card with the key points and timestamps, and the ↓ link drops to that concept's full write-up below.

01Know when to use Deep Research - and when not toDeep Research costs more compute and minutes;›

Deep Research costs more compute and minutes; use it where accuracy and sourcing matter, not for captions.

Tiers (0:00-0:05)

Citations demo (0:05-0:09)

Deep Research (0:09-0:13)

↓ Full write-up of this concept

02The model selector and "Best"Choose the underlying model per query, or let "Best" pick;›

Choose the underlying model per query, or let "Best" pick; experiment to learn their strengths.

Selector (0:13-0:15)

↓ Full write-up of this concept

The concepts in full

Know when to use Deep Research - and when not to

how-to

"The wisdom to know when not to use Deep Research."

Same PhD prompt in normal search gave ten cited lines; in Deep Research it planned a search, listed sources, explored international cooperation and nuclear economics, and produced a long report with frameworks (realism, liberalism, dependency theory) - still with per-line citations. Yes for "most iconic Indian ads and their psychological underpinnings"; no for an Instagram caption or a LinkedIn-to-thread repurpose.

Do it in this order

01Ask in normal search first for quick facts.
02Switch on Deep Research only for multi-source, high-stakes questions.
03Read the plan and report; check citations line by line.

Why it matters

Spends time and credits where they pay off.

The model selector and "Best"

o3 is "known for better reasoning - not a creative writer".

Options at recording: GPT-4.1 (fine for Instagram captions), Gemini 2.5 Pro, Grok 3 Beta, Claude Sonnet 4, o3 and a Perplexity reasoning option. He advises fiddling with models to learn, rather than always using auto-select.

Why it matters

Model fit matters as much as prompt wording.

This idea elsewhereparallelAI Studio: comparing models side by side

Tools referenced

Tool	Coverage	Moment	Context
Perplexity	demonstrated
Perplexity Deep Research	demonstrated
OpenAI o3	mentioned
Grok	mentioned
Claude Sonnet 4	mentioned
Gemini	mentioned

Action items

Resources mentioned

Resources

docLesson page on the platform

Extraction notes

This page was built from an auto-generated transcript, which garbles product and people's names. Those were corrected silently in everything above and logged here for transparency. The warnings flag claims that were true on the recording day but change fast.

Transcript corrections applied

The transcript says	The trainer actually means
cloud 4.0 sonnet	Claude Sonnet 4
first aid sources	first eight sources (likely)

True on recording day — verify before relying

Vendor UI walkthrough: menu names, layout, plan limits and pricing change monthly - check the live product before following click paths.
Model list and the "Best" option ("the latest feature") are mid-2025; no plan limits given (0:13-0:15).