> For the complete documentation index, see [llms.txt](https://docs.scribeware.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.scribeware.com/into-the-weeds/ai-assistant-mcp/teaching-it-your-style.md).

# Teaching It Your Style

Standing Instructions, Extended Guidance and per-entry AI notes — how you get the AI writing in your voice instead of generic inspector-speak.

Out of the box, an AI writes like an AI. It'll say "outlet" when you say "receptacle", hedge where you'd be blunt, and quietly invent a house style that isn't yours. ScribeWare gives you three places to fix that, and they're all worth filling in.

Two of them live in **Settings → AI**, under the enable switch.

## Standing Instructions

The short box. Whatever you put here is sent to the AI with **every single request** — it can't be missed, ignored or forgotten halfway through a long conversation.

Because it goes out every time, keep it tight. The character counter under the box turns amber past 350 characters and red past 400, which is a nudge rather than a hard limit. Use it for rules about *behaviour*, not for a dictionary:

```
Before writing or editing observations, read scribeware://company/guidance
for terminology and style. When corrected on any result, call submit_mcp_feedback.
If the correction reflects a standing preference, also update the company guidance.
```

That example is the default suggestion, and it's a good starting point: it tells the AI where the real detail lives, and what to do when you correct it.

## Extended Guidance

The long box, directly underneath. This one is loaded **on demand** — the AI fetches it when it's about to do something that needs it, rather than carrying it around all day. So it can be as long as it's useful.

This is where the substance goes. A structure that works well:

```
Terminology:
  - Use "receptacle" not "outlet"
  - Use "service panel" not "breaker box"
  - Use "service disconnect" for the main shutoff

Writing style:
  - Lead each observation with location, then condition, then recommendation
  - Severity levels: Safety Hazard > Major Defect > Maintenance Item

Things to avoid:
  - Do not estimate repair costs or timelines
  - Do not use "up to code" — use "consistent with installation
    standards at time of build"

Workflows:
  - When asked to review a report before delivery: flag observations with
    spelling or grammar errors; flag defect-level observations with no photo;
    flag observations that lack a clear recommendation.
```

That last section is the one people underestimate. **Workflows are saved prompts.** Write down what "review this report" should mean once, and from then on *"review this report"* is a two-word instruction that does the full checklist.

{% hint style="info" %}
Both boxes are **company-wide**. Everyone on your team who connects an AI gets the same instructions, which is exactly what you want — it's how you stop three inspectors producing three different-sounding reports.
{% endhint %}

## Just Telling It Works Too

You don't have to go into Settings every time. Say it in the chat:

> From now on, always use "receptacle" instead of "outlet".

The AI can write that into your guidance itself. It'll usually tell you it's done so. Worth knowing what's happening under the hood: a standing preference gets saved to your company guidance permanently, so it applies tomorrow, and to your colleagues, not just to the rest of today's conversation.

If it *doesn't* save something you meant to stick, ask it directly — *"save that to the company guidance"* — or open **Settings → AI** and type it in yourself.

## AI Guidance on a Library Entry

The third place is per-observation, and it's the sharpest instrument of the three.

Open any library entry for editing and you'll find an **AI guidance** box at the bottom. It's notes about *that one observation* — when to reach for it, and how to fill it in:

> When using this observation, set the modifier to 'major concern' if a safety issue is mentioned.

This text is AI-only. It never appears in a report, clients never see it, and it doesn't show up in library searches. Use it for the judgement calls that are obvious to you and invisible to everything else — which of two near-identical entries is the right one, what modifier normally goes with it, what has to be filled in before it's usable.

## Correcting It

When the AI gets something wrong, tell it so plainly. Two different things can happen, and they're worth telling apart:

* **It broke** — the wrong section got edited, a tool threw a confusing error. The AI files that as feedback to ScribeWare, and we use it to fix the integration.
* **You have a preference** — "don't write it that way". That's a settings change, and it belongs in your guidance.

You don't need to manage this distinction yourself; the AI is instructed on how to sort them. But if you notice the same correction not sticking a third time, say *"add this to the company guidance"* and it will.


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter:

```
GET https://docs.scribeware.com/into-the-weeds/ai-assistant-mcp/teaching-it-your-style.md?ask=<question>&goal=<endgoal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.
