Building a Brand Voice Context, Step by Step

TL;DR: A brand voice context is one saved entry in Prompt Architects' Context Library: a name, a Content field holding your voice rules and real samples, and an optional shortcut ID, capped at 5,000 characters. Build one from five real samples, save it, then attach it to any prompt so ChatGPT, Claude, and Gemini output matches your brand by default.

What is a Context, and why start with brand voice?

A Context in Prompt Architects is a small saved object: a name, a block of text in a field called Content, and a couple of optional settings. You write it once, save it, and attach it to future prompts instead of retyping the same background every time. It is not a preset or a template category. Under the hood it is just a name and a content string, so what you put in that content is entirely up to you: brand voice, a product spec, a client's constraints, your own working style.

Brand voice is the strongest first context to build for one practical reason: it is the one piece of background that never changes per task. Your audience shifts by campaign, your constraints shift by client, but the way your brand actually writes, the words it favors, the words it bans, the rhythm of its sentences, stays the same across a LinkedIn post, a support reply, and a product description. That stability is exactly what makes it worth building once and reusing everywhere, rather than re-describing it inside every new chat.

If you want the deeper argument for why this matters and a broader framework for what a voice document should contain, see our brand voice guide. This post is the narrower, mechanical half: the actual steps to build a brand voice context inside the product, save it correctly, and get it applying to real prompts.

Pull five real writing samples instead of describing your voice from memory

Skip the adjectives. "Friendly but professional" describes nothing a model can act on, and it is usually not even true, it is what you wish your voice sounded like rather than what it actually reads like on the page. The reliable starting point is samples, not description.

Pull five pieces of real, already-published or already-approved copy: an email, a landing page section, a social post, a support reply, and one more from whatever channel you write in most. Choose pieces that felt right when you sent them, not pieces that needed three rounds of edits to get there. Three samples is the minimum for a model to find a pattern instead of mirroring one example too literally; five gives it enough range across formats without bloating your character budget once you turn them into rules next.

A quick filter for whether a sample belongs in the set:

How do you turn five samples into rules a model can follow?

Paste the five samples into a chat and ask the model to find the pattern: average sentence length, whether it uses contractions, how it opens and closes, what kind of claims it makes, what it never says. A model is better at pulling out a pattern from examples than a person is at describing their own writing from the inside, mostly because a writer's habits are invisible to the writer.

Turn what comes back into instructions, not impressions. "Confident" is an impression. "Makes direct claims, never opens with a question, avoids 'we believe' or 'it seems'" is an instruction a model can actually follow. Do the same for a short vocabulary list: five or so words or phrases your brand actually uses, and five or so it never does. The never-use list does more work than any other single line in the context, because it is the fastest way to strip out the generic AI-sounding words that creep back in no matter how good your examples are.

Here is a template you can paste straight into the Content field and fill in. Keep the finished version under roughly 900 words so it clears the 5,000-character cap with room to spare:

BRAND VOICE CONTEXT — [Brand or product name]

Voice rules (write each as an instruction, not an adjective):
1. [e.g., Uses "you," never "clients" or "users."]
2. [e.g., Sentences average under 18 words. No triple-clause sentences.]
3. [e.g., Makes direct claims. No "we believe," "we think," or hedging.]
4. [e.g., Contractions on. No exclamation points in body copy.]

Always use: [3-5 words or phrases your brand actually says]
Never use: [3-5 banned words, e.g. seamless, leverage, revolutionary]

Sample 1 ([format, e.g. product email]):
[Paste 50-150 words of real, approved copy]

Sample 2 ([format, e.g. landing page section]):
[Paste 50-150 words of real, approved copy]

Sample 3 ([format, e.g. social post]):
[Paste 50-150 words of real, approved copy]

Apply to: [what this voice profile covers — marketing copy, support replies, internal docs]

Where do you actually create the context inside Prompt Architects?

Open the Context Library from the sidebar (it lives at /context in the dashboard) and select Create Context. The form asks for three things: a Name (something plain like Brand Voice or Company Voice Guide), the Content field where the template above goes, and an optional Variable ID. Save, and it appears immediately in your library, ready to attach to a prompt.

This is a Pro-and-above feature. It is not part of the free plan at all, so if the option is missing or shows an upgrade prompt instead of the create form, that is expected on a free account rather than a bug.

How much room do you have, and what happens if you go over?

The Content field enforces a flat 5,000-character limit, and that number is the same regardless of plan: Pro, Advanced, and Team all get the same cap for a single context's content. The form counts characters as you type and blocks saving past the limit, with a plain message telling you the content is over the cap if you paste in more than it holds.

5,000 characters is roughly 800 to 900 words, which comfortably fits four to six voice rules, a short vocabulary list, and two or three samples once you trim each to its most representative 100 to 150 words. If your first draft runs long, cut samples before you cut rules, a fourth rule earns its place more reliably than a third full-length example does.

Separately from the content length, Prompt Architects also caps how many contexts you can save in total, and how many you can attach to a single prompt at once. Both limits are set per plan rather than published as a fixed number anywhere customer-facing, so treat any number you see quoted for this elsewhere as unverified and check your own Context Library page instead, it shows your saved count against your actual limit.

Give it a shortcut ID so you can drop it into any prompt

Below the Content field is an optional Variable ID box. Leave it blank and you will always attach the context through the picker in the prompt enhancer. Fill it in, and you get a second, faster route: typing the bracketed ID directly inside any prompt substitutes the full context content at that spot, no picker required.

The product will suggest an ID from your context's name the moment you type one, and you can edit it. It has to start with a letter or digit, run 2 to 32 characters of letters, digits, or underscores, and it always gets a c_ prefix once saved, so "Brand Voice" becomes something like c_brand_voice. Typing [c_brand_voice] anywhere in a prompt, in the enhancer or pasted into the extension sidebar, pulls in the full saved content at that point. This is the fastest way to fold your voice profile into a prompt you are otherwise writing from scratch, rather than stopping to open a selector.

Should teammates be able to use it too?

If your plan includes team seats, the context form has a second toggle: keep it private, or share it with your team. A shared context stays editable only by whoever created it, teammates can select it and, if you gave it a shortcut ID, reference that ID in their own prompts, but they cannot change the underlying rules or samples. That asymmetry is deliberate: one person owns the canonical voice, everyone else consumes it.

This is the difference between a brand voice context and a shared document sitting in a folder somewhere. A shared context updates for every teammate the moment the owner edits it, with no separate copy to keep in sync and no message reminding four people to grab the latest version.

How do you apply a saved context to a real prompt?

Two ways, and they are not exclusive. In the prompt enhancer, turn on Personal Context, then pick your brand voice context (and any others you want) from the list that opens. It gets folded into that request, labeled by name, alongside whatever you typed. Or, if you gave it a shortcut ID, skip the toggle entirely and type the bracketed ID straight into your prompt text; it resolves to the full content before the request goes out.

Both routes work the same whether you are enhancing a general text prompt, an image prompt, or a video prompt, since the context data travels with the request regardless of prompt type. And because a context lives on your account rather than in one browser tab, the same picker shows up wherever you have access to the enhancer, the web dashboard and the Chrome extension sidebar alike. If most of your prompting happens through a client like Claude Desktop or Cursor instead, our guide to setting up the MCP server in Claude Desktop covers connecting that surface to the same account.

Test it, then refine it, this is the part everyone skips

Building the context is the easy half. Run the exact same prompt twice, once with your brand voice context attached and once without, and read the two outputs side by side. You are checking for two things: did the banned words from your never-use list actually disappear, and does a colleague who knows your brand recognize the output as yours without being told which version they are looking at.

Open the context from your library, edit the Content field, and save whenever the test comes back weak. There is no version history on a context itself, so if you want to keep the previous draft around, copy it somewhere before you overwrite it. Most brand voice contexts need one or two rounds of this before the never-use list actually catches what slips through, and then they settle for months at a time.

What if Contexts isn't available on your plan, or you've hit your limit?

If you are on the free plan, Contexts is not available at all, not a smaller version of it, the feature itself is gated to Pro and above. If you are already on a paid plan and see a limit-reached message when you try to save, your options are the same as with any capped feature: delete an older context you are not using, or upgrade. Nothing you already saved stops working while you decide; the limit only blocks new saves past your current count.

Product managers who want the same mechanism for a different kind of standing context, project scope, audience, and constraints rather than voice, can follow the same six steps with different content; see keeping product context in every AI chat for that angle. And if brand voice is only one of several standing facts about how you work, building a professional context profile covers the broader version, role, audience, and tone together, using the same underlying feature.