Brand Voice Tools Compared: Jasper, Writer and Custom Instructions

For most small businesses, custom instructions are the best starting point: a shared ChatGPT or Claude project holding your voice guide costs nothing beyond the chat plan. Jasper (Pro at $69 a month per seat, $59 billed annually) earns its price for one marketer producing weekly campaigns, and Writer suits teams needing banned terms enforced.

The three options differ less in how well they imitate your voice than in how they hold it. Jasper and Writer store your voice as settings inside the tool and apply them to every draft. Custom instructions are text that each person or project carries, which the model follows most of the time and quietly ignores some of the time. Enforcement and ownership, not writing quality, decide which one fits.

Follow me on Instagram@sagnikteaches

How each option stores your voice

Before comparing prices, it helps to know what each tool actually keeps about you, because that is what you are paying for.

Connect on LinkedInSagnik Bhattacharya

Jasper: Brand Voice, Knowledge and a Style Guide on Business

Jasper describes Brand Voice as letting its AI analyse your content's tone of voice, after which you set tone and formatting rules so its writing sounds like you. Alongside voices sit Knowledge assets (facts about your company) and Audiences (who you are writing for). The Pro plan includes two Brand Voices, five Knowledge assets and three Audiences. The Business plan removes those caps and adds a Style Guide, which applies your rules on punctuation, capitalisation and terms when content is generated, flags deviations, and can match a rule only when capitalisation is exact, which matters for product names.

Subscribe on YouTube@codingliquids

Writer: voice profiles, termbanks and a connected style guide

Writer builds a voice profile from examples: at least 300 words, with 500 or more recommended, pasted into up to eight text boxes. Its own guidance is to use one content type per voice, so a voice built from client emails should contain only client emails. You can also write the voice description yourself instead. Admins create team voices that others use. Two changes this year matter for brand control: since March 2026 you can connect a termbank (your approved and banned terms) to a team voice so banned words are removed automatically, and since April 2026 you can connect your style guide to a voice so its suggestions apply when outputs are generated. The Starter plan allows one team voice and one organisation-wide voice; Enterprise has no limit on shared voices.

Custom instructions: text you control, in a chat tool you choose

"Custom instructions" covers three different places, and the difference matters for teams:

• Personal instructions. ChatGPT's custom instructions, in the personalisation settings, apply to all of one person's chats. Paid plans such as Plus and Business allow up to 5,000 characters, up from the old 1,500 limit. They belong to the individual, so five staff means five versions.
• Shared projects. A project holds instructions plus reference files, and every chat inside it uses them. OpenAI says ChatGPT project sharing works on every plan, including Free, and Business adds sharing with groups or the whole workspace. Claude projects can be shared on Team and Enterprise plans, with members set to "can view" (use it, not change it) or "can edit"; Claude's Free plan allows up to five projects, but custom project instructions need a paid plan.
• Organisation-wide skills, and Google's equivalent. On Claude Team and Enterprise, owners can upload a skill (a packaged set of instructions and files) for the whole organisation; it is switched on for everyone by default, and individuals can turn it off. On the Google side, Gemini Gems can be shared, and recipients can see a Gem's instructions and files. Google is replacing Gems with skills, though: Gem support ends for personal accounts in November 2026 and for Workspace business accounts in March 2027, with existing Gems recreated as skills automatically. If you are setting this up now, build it as a skill.

Prices and plan limits side by side

List prices in USD, checked in September 2026. The chat-assistant figures come from the plans you would buy anyway for general AI use, which is why the custom-instructions route often costs nothing extra.

Two details catch people out. Jasper Pro is a one-person plan: the second user sends you to a sales conversation about Business. And ChatGPT Business and Claude Team both need at least two seats, so a sole trader comparing Jasper Pro with "custom instructions" is really comparing it with a $20 individual plan. (Claude's Free plan can hold projects, but not the custom instructions that carry your voice.)

Requested, applied or blocked: how banned words are handled

Every brand has words it never wants to see. For a law firm that list usually includes "guarantee", because a promised outcome is a professional risk as well as a style problem. How each tool treats that list is the most practical difference between them.

Here is an illustrative test. The prompt, run inside a shared project whose instructions say "Never use the word guarantee":

Write a 90-word update for clients announcing our new
fixed-fee review of commercial leases. Use our house voice.

An illustrative output:

We have introduced a fixed-fee review for commercial leases.
For one agreed price, a lawyer will read your lease, mark the
clauses that carry cost or risk, and explain each one in plain
words. There are no hourly surprises: the fee we quote is the
fee you pay, guaranteed. Reviews are usually returned within
five working days. To book, reply to this email with a copy of
your draft lease and the date you need to sign.

The slip is easy to explain: "guaranteed" got through because the rule named only "guarantee", and the model treated the other form as a different word. The five-working-day turnaround also needs checking against what the firm can actually deliver. The remedy on the instructions route is to list every variant (guarantee, guaranteed, guarantees, guaranteeing) and add a last line to the instructions: "Before finishing, list any word from the banned list that appears in your draft." Then search the final text for the stem "guarant" before it goes out. On Writer, a connected termbank removes banned terms automatically, and on Jasper Business the Style Guide flags the breach at generation. On Jasper Pro and the instructions route, the rule is a request, so the find step stays in your process for good.

A full banned-phrase list, with the variants to include, is in the tutorial on AI words to avoid.

A one-afternoon trial: an eight-person law firm tests all three

An illustration of how to choose on evidence rather than demos. An eight-person law firm (five lawyers, a practice manager and two assistants) wants client updates, website pages and a monthly newsletter to sound like the firm rather than like a template. Three people write client-facing content. The firm already pays for Claude Team for all eight, so the custom-instructions route costs nothing extra.

The practice manager set up a fair test in about three hours:

1. Chose five samples the partners would sign today: two client updates, two website service pages and one newsletter.
2. Wrote a one-page voice guide with testable rules and a banned list of twelve words, with variants.
3. Loaded the same material into a Jasper Pro trial (7 days), a Writer Starter trial (14 days) and a shared Claude project.
4. Ran the same five briefs through each: a fee change notice, a new-service page, a newsletter lead item, a reply to a client asking about delays, and a short social post.
5. Printed all fifteen drafts without labels and had two partners rate them blind.

The filled-in scorecard (each criterion out of 5):

Then the yearly cost for the three writers:

• Shared Claude project: $0 extra, since the Team seats are already paid for.
• Jasper Pro: $708 a year at $59 a month, but it covers one writer. Three writers means a Business quote with a 12-month commitment.
• Writer Starter: three seats at the $29 to $39 range trackers list would be about $1,044 to $1,404 a year, to be confirmed at sign-up.

Writer scored best, by two points out of thirty. The firm decided those two points were worth a find-and-replace step, not $1,000 or more a year and a second tool to administer. It kept the shared Claude project, put the voice guide into an organisation skill as well so it applied in every chat, and added a rule that the practice manager is the only person with edit rights. The decision gets revisited if the content volume doubles or a banned word reaches a client. That last condition is the useful one: it turns "is Writer worth it?" into something the firm will notice.

When Jasper is worth $59 to $69 a month

Jasper is built for marketing output, and it pays off when one person produces a lot of it. A good fit looks like a financial planning firm with a single marketing manager who writes the website, a fortnightly email, social posts and seminar invitations. That person gains two things over a chat assistant: marketing-specific workspaces and agents, and two separate Brand Voices on Pro. The second voice matters more than it sounds. The firm's client letters should be measured and careful; its seminar invitations can be warmer. Two voices keep those apart instead of averaging them into one bland middle.

Watch-outs before you buy:

• The seat is one person. If a second colleague needs to draft, you are into a Business quote, and Business starts with a 12-month commitment.
• The Style Guide, the feature that enforces punctuation and terms rules, is Business-only. On Pro your rules are part of the voice, not checked separately.
• Some usage-based features on Business, such as agents, run on credits. Jasper says standard content creation in Chat and Canvas is not what drives credit use, but ask how your planned use would be billed.
• Use the 7-day trial on real briefs. A week is short, so have your samples, voice guide and five test briefs ready before you start it.

When Writer is the better fit

Writer's strength is control across a team. It suits two to five people producing content where terminology must be exact: an insurance broker whose product names, cover descriptions and excluded phrases have to be right every time, or a firm whose compliance adviser has given it a list of words that must never appear. The connected termbank does the policing, and the admins, not every writer, own the voice.

A realistic mistake shows the main limitation. A recruitment agency on Writer Starter built its one team voice from a mix of job adverts, candidate emails and LinkedIn posts, because it wanted one voice for everything. The results came out as a hybrid: job adverts that sounded like chatty emails, and candidate emails padded with the punchy lines of an advert. Writer's own guidance is one content type per voice. With a single team voice on Starter, the agency rebuilt it from client-facing emails only (the content where tone mattered most commercially) and handled adverts through a playbook with its own instructions.

Other points to weigh:

• Starter is capped at five users. Beyond that, Enterprise is a sales conversation.
• Seats come with fixed credit limits, so ask what a month of your normal volume uses before committing.
• Writer says it does not train or improve its models on customer data by default, which helps if your drafts contain client details.
• The 14-day trial is long enough to run the five-brief test twice, once before and once after connecting the termbank, so you can see what the termbank changes.

When custom instructions are all you need

For a business of one to fifteen people writing a few pieces a week, a shared project in the chat assistant the team already uses is usually enough. You keep one tool, one login and one bill, and the voice guide sits next to the other work people do with AI. The guide to setting up Claude projects as a shared team assistant and the one on using ChatGPT projects cover the mechanics for each tool.

The instructions route breaks in three predictable places:

1. Everyone writes their own. If the voice lives in personal custom instructions, five people produce five voices. Use a shared project or organisation skill, not personal settings, for anything customers see.
2. Rules are requests. The model follows them most of the time. Keep a final find step for banned words and a human read for anything with fees or promises.
3. Nobody owns it. Shared instructions with edit rights for everyone get changed, lengthened and contradicted. Give one person edit rights and everyone else view access.

One thing not to build: a custom GPT. OpenAI is retiring them, and existing GPTs stop running on 11 December 2026. Its replacement turns a GPT into a plugin, with the instructions becoming a skill, but for a brand voice a shared project does the job with less to maintain.

Building the instructions version so it holds

Most weak voice set-ups fail on the wording of the rules, not the tool. Vague rules give the model nothing to test itself against. Compare:

Here is a filled-in set of project instructions for an illustrative twelve-person recruitment agency, the kind of block that fits comfortably inside the 5,000-character limit:

WHO WE ARE
Twelve-person recruitment agency placing finance and office
staff with small and mid-sized employers. We fill permanent
and temporary roles. We do not do executive search.

VOICE
Direct and warm. We sound like a consultant who has read the
CV properly, not like a job board.
- Address candidates and clients by first name.
- Sentences mostly under 25 words. No exclamation marks.
- Name the specific role, salary band and location type
  (office, hybrid, remote) when we know them.

NEVER WRITE
rockstar, ninja, fast-paced environment, competitive salary
(give the band instead), guarantee / guaranteed / guarantees,
"exciting opportunity", "don't miss out".

FACTS (use only these; ask if something is missing)
- Standard permanent fee: [see attached fee sheet]
- Replacement period: [see attached terms]
- Office hours: [hours]

BEFORE YOU FINISH
List any NEVER WRITE term in your draft and remove it.
List any figure you used that is not in FACTS.

And an illustrative test. The prompt: "Write a short email to a client, [first name], introducing two shortlisted candidates for their accounts assistant role." The model's final lines read: "Both candidates are available to start immediately and are a great fit for your fast-paced team. Checked: no NEVER WRITE terms used. Figures not in FACTS: none." Two problems: "fast-paced" appeared despite the ban, because the banned entry was "fast-paced environment" and the model used a different noun, and the self-check reported nothing, which is why the self-check can't be the only check. "Available to start immediately" is also an invented fact unless both candidates said so. The fix is to ban the stem ("fast-paced" in any phrase) and to keep a human reading every client email before it is sent. Writing the guide itself is covered in how to build a brand voice guide that AI can follow, and the mechanical rules (numbers, dates, capitals) belong in a house style guide alongside it.

Where voice set-ups drift after the first month

Whichever tool you choose, the voice degrades in the same ways, usually within a quarter:

• Rule creep. Every awkward draft prompts a new rule, until forty rules contradict each other and the model satisfies none of them well. Cap the list: if you add a rule, remove or merge one.
• Stale samples. The voice was built from last year's service pages, so drafts keep naming a package you have renamed. Refresh samples when services or prices change.
• Personal settings fighting the shared ones. One consultant's personal instructions say "be upbeat and use emojis". Inside a shared project the two sets can mix. Ask staff to clear personal tone instructions for client work.
• Model updates. Vendors update the models underneath all three options without asking you. The same instructions can produce a noticeably different tone after an update.

The defence is cheap: keep the five test briefs from your trial and re-run them once a month. Compare the new drafts against the ones the partners approved. It takes about twenty minutes, and it catches drift before a client does. If you prompt ad hoc rather than from a shared project, the tutorial on writing prompts that sound like your business covers the prompt-level version of the same discipline.

Keeping your voice portable between tools

Tools change plans, prices and names often enough that you should assume you will switch at least once. Three habits make that painless:

1. Keep the master copy outside every tool. The voice guide, banned list and fact sheet live in your own document. Whatever sits in Jasper, Writer or a project is a copy of it. This matters more than it sounds: ChatGPT Business workspaces, for example, have no self-service data export.
2. Keep your sample set. The five approved pieces you used for the trial are also what you would feed Writer's voice builder, Jasper's Brand Voice or a project's files. Store them together, labelled by content type.
3. Keep your test briefs and their approved outputs. When you switch, run the same briefs in the new tool and compare. If the new drafts score lower, you know before customers see them.

With those three in place, moving from a shared project to Writer, or from Jasper back to a project, is an afternoon's work rather than a rebuild. The tool becomes replaceable, and the thing of lasting value, a written, tested description of how your business sounds, stays with you.

Choosing a brand voice tool: other questions

Do Jasper or Writer train their models on our content?

Both say no. Jasper states that data entered into it is not used to train the underlying models. Writer says it takes a zero data retention approach and does not train or improve its models on customer data by default. On the custom-instructions route, ChatGPT Business and Claude Team don't train on business content by default either; on individual plans, check the model-training switch in privacy settings.

Is a custom GPT still a sensible place to keep our brand voice?

No. OpenAI is retiring custom GPTs across ChatGPT plans, and they stop running on 11 December 2026. OpenAI's migration turns a GPT into a plugin whose instructions become a skill. For a shared brand voice, use a shared ChatGPT project, a Claude project, a Claude organisation skill or a Gemini Gem (becoming a skill from November 2026) instead, and keep the master copy in your own document.

How many writing samples does a brand voice need?

Writer asks for at least 300 words of examples and recommends 500 or more, split across up to eight text boxes, all from one content type. For a shared project in ChatGPT or Claude, three to five strong examples per content type work well. More samples help less than better ones: pick pieces a partner would sign today, not your oldest material.

Further reads

• How Agencies Keep Brand Voice Consistent in AI-Written Content — How agencies hold several clients' voices without mixing them up.
• How to Build Gemini Gems for Repeat Business Tasks — Google's route for Workspace firms: Gems, now becoming skills.
• Custom GPT vs Claude Project vs Gemini Gem: Which Should You Use? — The three chat-assistant containers compared in detail.
• Tone of Voice Examples: AI Copy Before and After Editing — Before-and-after edits that show what good voice rules produce.
• ChatGPT vs Claude vs Gemini vs Copilot for Small Business (2026) — Which chat assistant plan to pick before you add a voice tool.
• How Much Does Human Editing of AI Content Cost? — What a human line edit costs if the tool still misses your voice.
• How to Write Menu Descriptions With AI That Still Sound Like You — A five-stage method for AI-drafted menu descriptions that keep your kitchen's voice, with a voice card template, a drafting prompt and a before-and-after dish.
• Why Your AI Marketing Copy Sounds Generic (and How to Fix It) — Why AI defaults to the average of every ad it has read, the six fingerprints of generic copy, and the inputs and edits that make it sound like your business.
• Grammarly vs ChatGPT for Checking Business Writing — Grammarly or ChatGPT for checking business emails and documents: what each catches, where each goes wrong, and a worked choice for a small team.
• How to Build a Product Glossary AI Must Use in Every Draft — How to build a product and brand glossary AI uses in every draft: eight term types, a filled-in table, a compact AI block, loading, checking and upkeep.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Jasper pricing page, Jasper help pages (Style Guide) and trust pages; WRITER plans page and WRITER Knowledge Base (voice calibration, What's new at WRITER); Claude help pages (projects, organisation skills); OpenAI help pages and release notes (custom instructions, custom GPT retirement). Checked September 2026.