The o3 Prompt Guide for GTM

This guide was written in June 2025, when o3 was OpenAI’s frontier reasoning model, and is republished here lightly edited for format. The specific models have moved on — see GPT-5: Reasoning Tokens Change the Math for the next chapter — but the prompt patterns and the reasoning-first framing held up.

Why o3 for GTM?

In my experience, o3 is the strongest model that exists as of today (June 19, 2025) in terms of pure reasoning. Other models will outperform it on speed, context window, and creativity.

Models like Claude shine in certain use cases like writing copy, but a large part of a GTM motion is done through reasoning first. Reasoning will take you much further than pattern-matching.

Three things make o3 specifically useful for GTM:

Tool use. Unlike o1, o3 can use other tools, including web search and image-scanning.
Complex automation. The automatable part of your GTM motion can be broken down into complex, multi-step paths that o3 excels at thinking through.
Price. API pricing has already dropped 80% as of June 10 — probably due to Gemini 2.5 Pro pricing pressure.

The prompt structure

Four rules cover most of it:

Define the role. Tell o3 what it is. “You are a seasoned researcher.” “You are an experienced business strategist.”
Provide context. Always supply as much data as you can to get the most relevant output.
Specify the format. Say exactly what shape you want the output in — bullet list, CSV, JSON, table.
Match the task. o3 is better for strategic, deep-thinking tasks than something quick and easy.

Know the limitations

Context window. o3 is limited by how much information it can process at once.
Chat deterioration. Using o3 in chat often deteriorates after multiple prompts and only goes back to normal when you create a new chat.
Tooling support. Still not heavily supported across most GTM tooling — n8n and Clay only allow o3-mini natively.
Speed. Slow compared to other models.
Hallucinations. o3 still hallucinates occasionally.

Where you can use it today: ChatGPT (o3 and o3-Pro), Operator, and Deep Research.

Use case 1: Operator for lead sourcing

The prompt structure that works has five parts:

Role: “You are a lead-sourcing analyst”
Goal: “Build a list of sales leader job posts from startups in the US”
Output: “Return a CSV with columns: Company, Website, Job Post link, Job Title, City, Post Date”
Quantify: “Stop after 60 listings”
Foolproof: “Before searching, list your assumptions. Once I confirm, execute and share the leads”

Best practices from running this:

Monitor timeouts. You can’t keep Operator running for more than a day. A task will time out after 8–16 hours and you will have to start over from scratch, so be attentive.
Stay for the setup. Keep it open for the first 3 minutes so you can log in to any platform you need.
Batch the work. Keep it to 30–50 rows of data at a time, then refine the inputs.
Set realistic targets. o3 is slower than 4o.
Go beyond lists. Ask questions about what your company can offer each lead, rank and score leads, or highlight identifiable triggers.

The honest tradeoff: a Pro subscription gives you effectively unlimited agent requests, but Operator is not as good as Manus, times out faster, and requires ChatGPT Pro.

Use case 2: Campaign ideation

Start with business context. Add all relevant information about your business and save it in ChatGPT memory — or keep it in a single prompt you paste when needed:

What your company does
How you are different from your competitors
Case studies on who you’ve helped and how
How your ICP is solving this problem today
A ranking of the pain points you solve

Then layer in triggers and insights:

Are there triggers that are immediately obvious for your target market?
What is something new you can teach a prospect that they don’t know yet?
What is the cost of inaction if they do nothing?
What is changing in their industry that makes this relevant now?

And the campaign parameters: the goal (book sales calls, nurture leads, educate the audience), the ICP (job titles, industry, location, company size, tech stack), and the constraints (what is a clear no-go for this campaign).

A simple version of the ask: “You are an outbound strategist. Generate 6 different ideas (not copy) for cold email campaigns aimed at {ICP}.”

Tips:

Front-load your constraints up top.
Be very specific about your desired output.
Iterate on the ideas you get.
Ask o3 to list and score ideas on novelty and messaging-fit.

Avoid:

Vague ICP descriptions.
Overloading email templates (kills the context window).
Asking for too many ideas at once (lowers quality).

The full one-shot prompt:

CONTEXT
- You are o3, my outbound campaign strategist. DO NOT write subject lines or email copy.

TASK
- Generate 8 distinct campaign ideas (concepts only) for {ICP}. Here are all the details about {ICP}:
- For each idea, return exactly four bullet points:
    - Title – ≤ 5 words
    - Core Angle / Value Proposition – 1 sentence
    - Suggested Proof Element – (stat, mini-case study, quote, etc.)
    - Why It Resonates – 1 sentence tied to ICP pain

INPUTS
- Goal: {Objective}
- Key Benefits: {Value props}
- Do-Not-Repeat Angles: {List}
- Email Library (reference only, not to be copied)

QUALITY
- No subject lines, no greetings, no body text.
- Novelty > 50% vs. Email Library.
- Each idea < 80 words total.
- If info is missing, please ask me for any further context I can share with you.

Use case 3: Sales call and demo prep

Required inputs:

Buyer context — industry, size, tech stack, anything you know that was not mentioned on the call
The transcript, from Fireflies, Fathom, or similar
Your objective for the meeting
Business context saved to ChatGPT memory

The one-shot prompt:

You are o3, my sales-call strategist.

TASK
Based on the inputs, create a prep pack for my upcoming 60-minute sales call with the VP of Sales from CompanyX.

OUTPUTS
Include sections:
1. TL;DR (≤ 100 words)
2. Meeting Goal
3. Attendee Snapshot (role, KPIs, likely priorities)
4. Key Pain Points (in their own words if possible)
5. Tailored Value Props (ranked, with proof source to reference)
6. Agenda Outline (timeboxed)
7. 5 High-Impact Questions to Ask
8. Top 3 Likely Objections & Concise Rebuttals
9. Demo Storyline
10. Clear CTA & Next-Step Criteria
11. Flag any missing info as a "?? Info Gap" item at end.

INPUTS
- Buyer context: Raised $70M last month. Recommended to us through ReferralX. VP of Sales joined 90 days ago
- Objective: Win approval on 6-month agreement.
- {{FULL TRANSCRIPT}}

Tips:

Handle long transcripts. If the transcript is exceptionally long, break it up and upload it in parts, or trim it down using 4o in a separate chat.
Clean transcription matters. Make sure your call tool cleanly exports Speaker 1 vs. Speaker 2. We use Fathom and it does this well for free.
Review the deck first. Upload your general sales deck as a PDF and ask for feedback before asking anything about this specific deal.

Is o3-Pro worth it?

Yes, if:

You are processing long documents frequently (RFPs, legal docs, long sales cycles)
You need strategic decisions that require deep reasoning
You want access to Operator on the Pro plan (nearly unlimited runs)
Even then, Pro is likely only worth it for 1–2 seats on the team at most

No, if:

You are using o3 primarily for high-volume tasks that require multiple prompts
Speed is important (o3-Pro frequently thinks for more than 10 minutes on a prompt)
o3 alone works great for most GTM use cases — especially if you are making LLM budget decisions for a large sales or growth org