AI Instructions
SwiftlyCRM's AI writes proposals, drafts contract clauses, analyses meetings and summarises pipeline health. Out of the box it writes in a sensible general style - which is fine, until it calls your product the wrong thing, writes in the wrong English, or promises something your legal team would never sign off.
AI Instructions let an admin define the rules the AI must follow. Every AI feature reads them before it writes anything.
Start With Your Company Profile
Before the AI can write for you, it needs to know who you are. Settings > Company Profile holds what your business does, your products and services, what sets you apart and who you serve. Every AI feature that writes on your behalf reads it - the About Us in proposals, contracts, project briefs and milestones, meeting suggestions and research.
Short on time? Enter your website and click Fill from website. The AI reads your home, about and services pages and drafts the profile for you to check and save (it uses 1 AI credit). Organisation admins can change the profile; everyone in the workspace can see it.
The AI never invents facts about you
If a proposal needs a detail the profile doesn't give - years in business, certifications, past clients - the AI leaves a [placeholder] for you to fill in rather than making one up. Without a profile at all, it describes your company only from the deal and its line items.
The profile says what your company is; the instructions below say how the AI should write.
Where to Set Them
Go to Settings > AI Instructions. You need admin access.
Write your rules in plain language - the same way you'd brief a new hire on their first day:
Always use British English spelling. Refer to our product as "the Platform", never "the software". Never promise delivery timelines in proposals - say "timelines confirmed at kickoff". Our tone is direct and practical, not salesy.
Scoping
Each set of instructions has a scope, so you can be general or surgical:
| Scope | Applies to |
|---|---|
| Global | Every AI feature |
| Proposals | Proposal generation only |
| Contracts | Contract drafting only |
| Meetings | Meeting analysis and summaries |
| Projects | Project briefs and milestones |
| Pipeline | Pipeline and deal-health insights |
| Research | Market research and problem definitions |
| Transcription | Voice note polishing |
You can also scope by branch, so a branch serving a different market can have its own terminology while the rest of the organisation keeps the house style.
How the rules combine
The most general rules apply first, then the more specific ones layer on top. An organisation-wide instruction plus a proposals-specific one means proposals follow both.
What Makes a Good Instruction
Be specific and testable. The AI follows concrete rules well and vague aspirations poorly.
| Works well | Works poorly |
|---|---|
| "Use British English spelling" | "Write professionally" |
| "Never quote a price without 'excluding VAT'" | "Be careful about pricing" |
| "Call the customer 'the Client' in contracts" | "Use good terminology" |
| "Keep proposal summaries under 150 words" | "Be concise" |
Things worth defining early: spelling and language variant, what you call your product and your customers, claims you must never make, required disclaimers, and your tone.
Turning Rules On and Off
Each instruction can be enabled or disabled without deleting it. Useful when you want to test whether a rule is actually helping, or pause a seasonal one.
Important
AI Instructions shape how the AI writes - they don't override the facts in your records. If a proposal shows the wrong figure, fix the deal or catalogue item; an instruction won't correct underlying data.
A Practical Starting Point
Rather than writing an essay on day one, add rules as you notice problems. The most common progression is:
- Set language and spelling
- Fix product and customer terminology
- Add the claims your team must never make
- Tune tone once the basics are right
Every AI output afterwards follows them, so you stop making the same edit over and over.