← AIAP Tool GuidesAll programsHomeSearch
AIAP Tool Guides·Gemini·15:35

Gemini Gems: Saved Custom Assistants with Instructions and Knowledge Files, and Choosing Flash or Pro with Show Thinking

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
← PreviousGemini Overview: Multimodal Uploads, @-Tagging Google Apps, Saved Info Personalisation, an
Lesson 2 of 4 in Gemini
Next →Gemini Deep Research and Multimodal Analysis: Role-Task-Constraint Prompts, Editing the Re

The short version

  1. A Gem is a saved assistant: instructions plus up to 10 knowledge files, reused every time - for recurring work, not one-off questions (0:00-0:04).
  2. His interview-coach gem: a veteran marketing-careers coach persona, mid-level focus, mock interviews, a stop phrase, and his CV as knowledge (0:04-0:08).
  3. Built live: a "campaign manager" gem turned "new iPhone launch" into three tweets, a LinkedIn post and an Instagram image idea; then 2.5 Pro with show thinking versus 2.5 Flash (0:08-0:15).

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.

01Building a Gem: persona, task, stop phrase, knowledgeWrite the instructions once (role, audience, task, how to end), attach reference files, save, and chat with…›

Write the instructions once (role, audience, task, how to end), attach reference files, save, and chat with it whenever the task recurs.

Anatomy (0:04-0:07)

New gem (0:08-0:10)

↓ Full write-up of this concept

02Flash for everyday, Pro with show thinking for reasoningPick the model by the output you need;›

Pick the model by the output you need; for high-stakes answers open the reasoning trace.

Model selector (0:10-0:12)

Show thinking demo (0:12-0:14)

↓ Full write-up of this concept

The concepts in full

01

Building a Gem: persona, task, stop phrase, knowledge

how-to

A good instruction set was "a long 20 to 50 liner" that took 20 minutes - and saves time every day after.

Interview coach: "you're an award-winning veteran job interview coach with more than 20 years of experience specializing in marketing careers", mid-level managers, simulate interviews, summarise after the stop phrase "the end"; knowledge = his CV, so feedback cites his real projects. New gem typed live: "you are a campaign manager in marketing field in social media domain - daily I will give you a creative idea, you give me three tweets, one LinkedIn post and maybe one image for Instagram". Calling this "fine-tuning" is loose - it is prompt-level, not training.

Do it in this order
Why it matters

Reusable expertise without retyping context.

This idea elsewhereparallelBC9 Day 1: projects and reusable context in Claude/ChatGPT
02

Flash for everyday, Pro with show thinking for reasoning

"The onus is on you - what is the vision, what is the end objective, what is the output."

Why might vitamin D be low? 2.5 Pro's show thinking listed steps such as regional factors for tropical India and diet. "Vitamin D-rich foods" needs only 2.5 Flash. For anything like a medical case study, "I cannot straight away rely on AI data" - read the reasoning.

Why it matters

Matches cost and speed to the task.

Tools referenced

ToolCoverageMomentContext
Geminidemonstrated
Gemini Gemsdemonstrated
ChatGPTmentioned
Imagenmentioned

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 saysThe trainer actually means
    current / currents interview coachKaran's Interview Coach
    gym / gengem

    True on recording day — verify before relying

    ← PreviousGemini Overview: Multimodal Uploads, @-Tagging Google Apps, Saved Info Personalisation, anNext →Gemini Deep Research and Multimodal Analysis: Role-Task-Constraint Prompts, Editing the Re