What guidelines do
The guidelines library lives at the workspace level: one set per brand, applied across every board that opts in. They carry more weight than a reference document. On a board with guidelines enabled, your agent is required to read them before writing anything and to treat them as binding rules.
The biggest objection to AI-written marketing is generic, off-brand output. That comes from a missing brief rather than from the model. Guidelines are the standing brief.
The five kinds of guideline blocks
- Text
- Tone of voice, do/don't rules, product facts, audience notes. The workhorse.
- File
- Your logo, brand fonts, an example email your agent should imitate. Stored privately in the workspace.
- Link
- A public source of truth, such as your brand page or a published style guide.
- Color palette
- Named brand colors with exact hex values, so "our pink" is never approximately pink.
- Voice
- The "Prevent AI language" preset: switchable rules that stop copy from sounding AI-written. More below.
The fast path: let your agent research your brand
If your brand lives on a website, there's no need to type your identity into forms. Have your agent go look at the site and write the guidelines itself. The prompt below is filled in for our demo brand; swap in your own site and workspace name.
Analyze https://bookstore.plainpaper.io. I want to understand: 1. What this brand does and who it sells to 2. Its tone of voice: how it talks, with real phrases as evidence 3. Its visual branding: colors (exact values where you can get them), typography, and how imagery is used Read the homepage and a few deeper pages. If you can browse with Chrome, take screenshots of the homepage and one product page and work from what you actually see. Then set up the guidelines for my workspace "P&P Bookstore" in Plainpaper: - a text block "Tone of voice" with 5 do rules and 5 don't rules, each backed by an example phrase from the site - a color palette block with the exact hex values you found - a text block "What we sell & who we sell to" in 5 sentences Keep every block short and write every rule as binding. Show me what you created when you're done.
Browsing depends on your agent's setup. Claude can fetch and read public pages on most plans; screenshots need a browser tool such as Claude in Chrome. If your agent can't browse, use the next prompt instead; it builds guidelines from what you tell it.
No public website, or a brand that lives mostly in your head? Brief the agent directly:
I'll describe my brand; turn it into workspace guidelines in Plainpaper for the workspace "P&P Bookstore". We're an independent online bookstore. We sound like a well-read friend: warm, curious, never academic. We recommend, we don't push. Short sentences. We say "readers", never "customers". Our colors are cream, deep navy, and a warm red. We never discount below 20% and we never use urgency tricks like countdown timers. Create: a tone-of-voice text block with do/don't rules, a color palette block (pick tasteful exact hex values for cream, deep navy, warm red), and a text block with the two hard business rules. Ask me up to 5 questions first if anything important is missing.
Setting them by hand
Everything the agent can do here, you can also do yourself. Open Guidelines in the app sidebar: you'll find a suggestion gallery of eight starter blocks (Prevent AI language, brand logo, tone of voice, example HTML file, color palette, brand fonts, do/don't rules, and a public source link). Click one, fill it in, done. Most teams end up with a mix: agent-drafted blocks, edited by hand where it got the nuance wrong.
Turn on "Prevent AI language"
The voice block is a preset with switches, and it's the single highest-impact block in the library. Each toggle is a concrete writing rule your agent follows verbatim:
- No AI punctuation: kills the em-dash-everywhere habit.
- Cut fluff: no "in today's fast-paced world" filler.
- No AI hype: bans "game-changing", "unleash", "elevate".
- No formulaic structure: stops the "It's not X, it's Y" tic and rule-of-three padding.
- No emoji: keeps emoji out of your copy entirely.
"Won't it sound like ChatGPT?" is the question every stakeholder asks. This block is the switch that answers it.
Which boards use the guidelines
Guidelines apply per board, and the default is on: when you create a board there's a checkbox to opt out for boards where brand rules would get in the way (a competitor-research board, say). Your agent can see whether a board has guidelines enabled, and on enabled boards it must fetch and apply them before authoring anything.
Verify it worked
Don't take the setup on faith. Test it the way you'd test a new freelancer:
Read the guidelines for my Plainpaper workspace back to me as a short list. Then write one example Instagram caption for a new arrivals post, strictly in our voice, so I can check whether the guidelines actually land.
If the caption sounds off, fix the rule, not the caption. Tell the agent what was wrong and have it update the block:
That caption was too salesy: we never open with the discount. Update the tone-of-voice guideline block with a rule that covers this, then rewrite the caption under the new rule.
Keep blocks small and single-topic. Five short blocks beat one long manifesto: your agent applies "never discount below 20%" more reliably when it isn't buried in paragraph twelve. Prune anything you wouldn't enforce.
Related
- Brand guidelines: the full reference for the library.
- Draft email campaigns: see the guidelines shape real output.
- Prompt library: more guideline prompts, including seasonal-refresh and multi-brand setups.