> ## Documentation Index
> Fetch the complete documentation index at: https://docs.granvl.com/llms.txt
> Use this file to discover all available pages before exploring further.
# Strategy
> Products, ICPs, angles, and research: the evidence your agent argues from.
Strategy is the layer underneath every page granvl builds. It is a tree, and the shape is the point:
```
Product the thing you sell
├── ICP who you sell it to
│ └── Angle the argument for why that audience should care
├── Research the evidence an angle points back to
└── Copy lines what actually gets written, tagged with all three
```
You can skip it. Ask your agent for a landing page with an empty Strategy tab and you will get one. But the copy will be generic by construction, not by accident, because the rule your agent follows is:
A claim needs support that exists. If the proof is not in the Product, the ICP research, or the Angle, do not write the claim.
With an empty Strategy tab there is no proof anywhere for a claim to point at, so every headline retreats to something safe and unfalsifiable. Fill the tree in first and the same prompt produces something specific.
## How your agent uses it
Before generating or deploying copy, your agent pulls a **bounded brief** rather than reading your whole workspace:
* `get_strategy_brief` returns one Product + ICP + Angle with the evidence linked to them, and an explicit list of what is missing.
* `get_copy_brief` returns the same brief plus the copy lines that already exist and the exact gaps for the ad or page being written.
The "explicit gaps" part matters. A brief that says *this angle has no research behind it* is what makes your agent ask you for evidence instead of inventing some.
## The five tabs
Strategy lives at **Strategy** in the sidebar, scoped to the workspace's [active brand](/guides/settings#brands).
| Tab | What it holds | Created by |
| ----------------- | ------------------------------------------------------- | ------------------------------- |
| **Products** | What you sell, what it does today, what it does not do | You, or granvl's built-in agent |
| **ICPs** | Who buys, what stops them, the words they use | You, or granvl's built-in agent |
| **Offers/Angles** | One product plus one ICP plus one argument | You, or granvl's built-in agent |
| **Research** | Transcripts, competitor findings, audience quotes | You, or granvl's built-in agent |
| **Copywriting** | Every headline, subheadline, and ad line written so far | An agent, never a blank form |
You don't have to fill this in by hand. granvl's [built-in agent](/concepts/built-in-agent) can research your market and write these records as you talk to it, which is usually faster than starting at a blank form. Your own MCP-connected agent reads them through the brief tools but doesn't author them.
Field-by-field detail for each is in [Strategy fields](/guides/strategy-fields).
## Products separate fact from plan
A Product record deliberately splits **current capabilities** from **planned capabilities**, and keeps a **limitations** field alongside them. The two are never merged when a brief is built, so copy cannot promise a roadmap item as a shipped fact. The limitations field is the honest edge: it is what lets an angle disqualify the wrong buyer on purpose.
## Angles carry the whole offer
An angle is not just a headline idea. It holds the problem it leads with, the desired outcome, the **mechanism** (why the promise is believable), the **proof**, the objection it exists to defuse, who it qualifies, and who it disqualifies. Two angles on the same product and ICP that argue different mechanisms are a real experiment; two that reword the same claim are not.
## The copy library compounds
Copy lines are stored, not regenerated from scratch each session. Every line carries:
* **status**: `testing` (a live hypothesis), `winner` (a test settled on it), or `archived`
* **rationale**: one sentence on what the line is doing, which objection it answers, which reader it is for
* **links** back to the product, ICP, and angle it argues from
Your agent reads the library before writing anything, because adapting a winner beats inventing a fifth variation of it. Two rules hold this together:
Agents cannot promote their own work. Lines save as `testing` and tagged agent-written. Only warehouse-backed experiment results promote a line to `winner`. A self-declared winner would poison every future brief that trusts the label.
There is no **New** button on the Copywriting tab. Lines arrive from your agent, either directly or as a side effect of building a page. Clicking an existing line opens its detail panel, where you can edit the copy and read why it was written that way.
## The opening rule
One copywriting rule is enforced in code rather than advised. For ad body copy (`ad_primary`), **the first three lines have to earn the fourth**. Readers see a truncated preview before "See more", so on most impressions those three lines *are* the ad. `create_copy_lines` rejects an ad body whose opening is a single run-on line.
An opening fails when it greets the reader, names the brand before giving a reason to care, restates the offer the ad is about to make, or only makes sense after reading the rest.
## Research is the receipt
Research entries hold the actual quotes, transcript excerpts, or competitor findings, plus **takeaways** written to be read on their own. Link each one to the ICP, angle, or product it supports and a strategy decision has an actual receipt behind it. Entries can also sit unassigned while a campaign is still taking shape, then be attached once the records they inform exist.
The fastest first pass: one Product, one ICP, one Angle, and one Research entry with real customer language in it. That is enough for a brief to stop reporting gaps, and it takes about ten minutes, or one conversation with the built-in agent.