AI Tool Selection Guide — NotebookLM, Claude, ChatGPT, Perplexity

Overview

Different AI tools have distinct strengths. Choosing the right tool for the task — and knowing how to prompt effectively — is the difference between output that reads as AI-generated and output that's genuinely useful. This guide summarizes the core use cases, strengths, and techniques for the four tools in active use at Asymmetric.

Context: This guidance emerged from a walkthrough during the ^[1], with a team AI professional development session planned for October. See also ^[2].

The Core Principle: Feed It Context

The most common failure mode with AI tools is under-prompting. Generic prompts produce generic output — which is why clients sometimes notice it reads as AI.

Two strategies for grounding AI output:

1. Expansive prompts — Provide the company URL, describe the product/service, and verify the AI understands before asking it to write. Coax it step by step until you're confident it knows what you're talking about.
2. Document-grounded projects — Upload all relevant source material (emails, briefs, decks, docs) into a project or notebook so the AI reasons from your actual context rather than the open internet.

"The secret to AI is that you've got to give it an expansive prompt… coax it a little bit until you're pretty sure that it knows what you're talking about. Then you let it write." — Mark Hope

Tool-by-Tool Guide

NotebookLM (Google)

Best for: Working exclusively within a defined set of documents; preventing hallucination; summarizing large document sets.

How it works:
- Create a new notebook per project (e.g., one per client)
- Upload all source material: PDFs, TXT, Markdown, Google Docs, website links, pasted email text
- PowerPoint/PPTX files should be converted to PDF first
- Supports up to ~200 source documents
- Can connect directly to Google Drive

Key behavior: NotebookLM will only reason from the documents you provide. It will not go out to the internet. This makes it ideal when you want to stay on-brand and on-brief without drift.

Outputs available: Chat Q&A, in-depth reports, mind maps.

When to use it:
- Synthesizing a large client brief or email history
- Generating content that must stay within provided source material
- Onboarding to a new account quickly

Claude AI

Best for: Creative writing, strategy, editing, and code review. The go-to tool for most content work (~90% of daily use).

How it works:
- Use the Projects feature to upload source documents (same concept as NotebookLM)
- Within a project, Claude reasons from your uploaded files rather than the open web
- Accepts a wide range of file types

Key strengths:
- Strong at nuanced writing and tone
- Excellent at editing and critiquing output from other tools
- Reliable for marketing strategy documents

Cross-tool technique: Run output through multiple tools to improve quality and reduce detectable AI patterns. For example: generate a draft in ChatGPT → paste into Claude and ask "what do you think of this?" → Claude will edit and improve it. Each pass makes the output less uniform and harder to identify as AI-generated.

When to use it:
- Writing and editing content (blog posts, strategy docs, ad copy)
- Marketing strategy development
- Code review and improvement
- As a second opinion on ChatGPT output

ChatGPT

Best for: General reasoning, brainstorming, and tasks where Claude isn't available. Also useful as a second opinion on Claude output.

Limitations to know:
- Cannot create Google Sheets or Google Docs natively
- More prone to hallucination than Perplexity for factual claims
- Statistics and data points should always be verified

Hallucination mitigation: Add this to your prompt — "Don't give me any facts, statistics, or details unless you can support them with a citation." This changes how it writes and reduces fabricated numbers.

Pro tip: After getting output, ask: "Are all of the statistics you gave me true, accurate, and can you support them with citations?" It will often revise and correct itself.

When to use it:
- Brainstorming and ideation
- Tasks where you want a second opinion on Claude's output
- General writing when Claude isn't the right fit

Perplexity

Best for: Internet research with citations; competitor discovery; finding what others are saying about a topic.

How it works:
- Functions like a high-speed, AI-powered browser
- Every claim is backed by a cited source with a clickable link
- Surfaces sources you might not find through a standard search (Reddit threads, niche industry sites, regulatory filings, etc.)

Key strengths:
- Citation-backed output eliminates unsourced hallucination
- Excellent for competitive landscape research
- Can reveal whether a client's SEO is working (if their site appears as a source, it's indexing well)
- Spaces feature allows document-grounded projects, similar to NotebookLM/Claude Projects

Limitations:
- Not strong at creative or strategic thinking
- Not ideal for long-form content generation

When to use it:
- Researching a new industry or client vertical
- Finding competitors and third-party commentary
- Verifying claims or finding supporting data with sources
- Checking a client's SEO visibility

Gemini (Google)

Best for: Tasks that require direct Google Workspace integration.

When to use it:
- Creating or editing Google Docs or Google Sheets directly
- Accessing Google Drive content
- Generating structured Google Workspace outputs

Note: ChatGPT and Claude cannot create Google Sheets natively. Use Gemini for this.

Quick Reference

Advanced Techniques

The Multi-Tool Pass

Generate output in one tool, then pass it to another for critique and editing. Each tool's edits make the output less uniform and less detectable as AI-generated. Works especially well for code and long-form content.

Verify AI Statistics

Never publish statistics from ChatGPT or Claude without verification. Either:
- Use Perplexity to find cited sources for the same claims, or
- Add to your prompt: "Only include statistics you can support with a citation."

Stay Able to Speak to the Output

AI-assisted content is only useful if you can present and defend it. Read everything before it goes to a client. Skimming and approving without understanding creates risk in client conversations.

Upcoming: AI Professional Development Session

A team training session is planned for October 2025 (after the client health check cycle). Mark Hope will present and walk through live examples. Required attendees include account managers, Raphael, and Gavin.

→ See ^[3] for context and action items.
→ Owner: ^[4] (scheduling)