← Blog

ChatGPT, Claude, or Gemini?

A PM's honest guide to which LLM for which task — and the ceiling they all share.

The takeaway

All three LLMs are useful for PM work. ChatGPT is your fast generalist, Claude is your deep thinker, Gemini is your Google power tool. But they all share the same ceiling: zero memory between sessions. The real question isn't which is smartest — it's when you're spending more time providing context than you're saving.

LLM comparison for PM work
ChatGPT
Best for
Brainstorming & ideation
First-draft PRDs
Meeting summaries
Custom GPTs for repeatable tasks
Watch out
Verbose output
Confidently wrong
Zero session memory
Claude
Best for
Long-document analysis (200K)
Nuanced spec review
Edge case identification
Strategic synthesis
Watch out
Overly cautious / hedges
No plugin ecosystem
Zero session memory
Gemini
Best for
Google Workspace integration
Web-grounded research
Multimodal analysis
Data analysis in Sheets
Watch out
Inconsistent structure
Weaker long-form writing
Google ecosystem lock-in

The honest truth: all three are useful. None of them are built for you.

Where general LLMs break down
Works well
One-off drafts & brainstorming
Quick analysis & email polish
Meeting summaries & ad hoc research
Breaks down
Ongoing product context
Cross-referencing past decisions
Constraint-aware recommendations

The tipping point isn't "is the AI smart enough?" It's "am I spending more time providing context than I'm saving?"

0
tokens remembered
between sessions

Every session starts from scratch. Your positioning, user segments, tech constraints, past decisions — all gone. ChatGPT's "memory" stores your job title. Not the trade-offs you made last quarter.

5 copy-paste prompt templates

While you're using general LLMs (and you should), these templates get better output. Notice what they all require: you manually providing context every time.

1 PRD Review
Review this PRD as a senior PM with 10 years of experience. Check for: logical inconsistencies, missing edge cases, assumptions stated as facts, vague success criteria, anything a skeptical eng lead would push back on. Be specific. Quote the exact section. Don't suggest rewrites. [Paste your PRD]
2 User Story Generator
Generate user stories for [feature]. Format: As a [user], I want [action], so that [benefit]. Context: Product: [1 sentence] | Users: [who] | Constraint: [the one limit] Include: 5-8 happy path + 3-5 edge cases with testable acceptance criteria. Do NOT include admin stories unless I specified admin as a user type.
3 Competitive Positioning
Position [my product] vs [competitor]. My product: [2-3 sentences]. Competitor: [2-3 sentences]. My real advantages: [...]. My real weaknesses: [...]. Generate: 1) positioning statement (under 20 words), 2) three differentiation angles ranked by defensibility, 3) the one comparison to avoid, 4) a "switch trigger" pain point. Be honest. Don't flatter my product.
4 Meeting Prep
Meeting in [time] about [topic]. Type: [stakeholder review / sprint planning / etc.]. Attendees: [roles]. My goal: [outcome]. Potential friction: [what's contentious]. Generate: 1) 3-point agenda with time allocations, 2) three goal-driving questions, 3) the one objection to prepare for + response, 4) proposed next step. Keep it tactical. No filler.
5 Feature Prioritization
Prioritize these features for next quarter. Context: [what you do, who for, stage] | Capacity: [size] | Top metric: [the one number] Features: 1. [A] 2. [B] 3. [C] ... For each: impact on metric (H/M/L + reasoning), effort (t-shirt + reasoning), strategic alignment (compounds or one-off?), opportunity cost. Recommend a sequence. Challenge pet projects.

These templates work. But notice: every one requires you to manually re-provide context. Your prompts get longer. You start keeping a "context doc" you paste into every session. You're building a manual memory system on top of tools designed to forget.

This is what we're building at DISKO.

An AI product partner that remembers your product, your decisions, and your context — so every conversation starts where the last one left off.

Join the beta →