How to turn scattered public data into a structured buyer model that makes your outreach feel like a conversation, not a pitch.
Most outbound teams using AI start at the wrong step. They ask AI to write the message before they've used AI to understand the buyer. The result is polished copy with zero depth - it reads well but converts poorly because it's built on assumptions, not evidence.
The standard AI outbound workflow: take a prospect's name + title + company, throw it into a prompt, and ask for a "personalized" cold DM. That's personalization theater. The model is guessing. And the prospect can feel it.
The approach in this guide is different. You spend 80% of your time on research and 20% on message generation. Opus 4.6 is strong enough to do pattern recognition across multiple data sources simultaneously — something that would take a human researcher hours per account.
<aside>
The shift: stop using AI to write at people. Start using it to think about people.
</aside>
• Holds and cross-references large context windows (200K tokens) - you can feed it an entire company's digital footprint in one session • Identifies language patterns and emotional undertones, not just keywords • Spots contradictions between what companies say publicly and what their users complain about • Builds structured mental models from unstructured data without losing nuance
The quality of your buyer model is directly proportional to the quality and variety of inputs you give the model. Feed it surface-level data, get surface-level insights. Below is the exact input stack I use, ranked by signal density.
<aside>
PRO TIP: Don't summarize the inputs yourself before feeding them to Opus. Give it the raw text. Opus is better at extracting signal from noise than you are at pre-filtering — that's the whole point.
</aside>
Knowing what to leave out is just as important as knowing what to put in. Feeding Opus bad inputs doesn't just waste tokens — it actively degrades output quality by diluting the signal.
<aside>
■ The single biggest mistake: feeding Opus a prospect's "About" page and asking it to personalize a DM. That's reading someone's resume and thinking you know them.
</aside>
Below are the exact prompts I use in sequence. Each builds on the previous output. Run them in order within a single Opus conversation so the model retains context across steps.
PROMPT 1 — Paste after uploading G2 reviews + Reddit threads