ChatGPT for marketing: a 2026 guide

You opened ChatGPT this morning and pasted a prompt that produced great copy last week. The output came back fast. You read it once and something feels off. The voice shifted. A word your brand never uses showed up in the third paragraph. The structure is right, the grammar is clean, the tone is not yours.

This is the story of ChatGPT for marketing in 2026. Roughly 90% of content marketers now use AI writing tools in their workflows. The output is fast and the productivity gains are real, but the consistency problem is real too. Most marketing teams have not solved it.

This piece covers what ChatGPT does well for marketing teams, the five places it breaks at scale, 15 prompts that work, and the layer underneath that makes those prompts hold.

What ChatGPT does well

ChatGPT is the most-used marketing tool of 2026 for a reason. The breadth and speed are real. Five use cases where it earns its place in the stack:

Campaign brainstorming at speed. Ten ideas for a launch theme in the time it takes to make coffee. The output is rarely your final answer, but it gets you past the blank page faster than any tool that came before.

First-draft generation. Blog post drafts, email bodies, social posts, internal memos. ChatGPT handles the structural lift. You edit on top instead of writing from scratch.

Competitor and customer feedback summarization. Paste a competitor landing page or a batch of customer support transcripts and get a structured read in seconds. The signal-to-noise ratio is good enough to act on.

Research synthesis across sources. Pull together findings from multiple articles, papers or interview notes. The synthesis is usually directionally right, useful for shaping a position.

Headline and subject-line iteration. Twenty variants in 30 seconds. You pick the two you like, refine and run a test. The volume of variants is what makes ChatGPT useful here, not the quality of any single output.

These five jobs are what put ChatGPT inside every marketing team. The breadth keeps it there. The quality is good enough to ship after light editing on most outputs. ChatGPT for content marketing, ChatGPT for digital marketing and ChatGPT for social campaigns all share the same five jobs underneath, which is why the same prompt patterns transfer across them.

Where it breaks

For teams that scale ChatGPT past one or two contributors, five failure modes show up almost without exception. They are not theoretical. They appear within the first quarter of serious use.

Voice drift

The same prompt produces a slightly different brand voice in March than in May. Model updates shift the defaults, your conversation history shifts the priors, and the output drifts. You cannot tell because each session feels normal in isolation. The drift only shows up when you read three months of output side by side.

Team scaling

Five marketers using ChatGPT for the same brand will produce content that sounds like five different companies. Custom Instructions help at the individual level. They do not transfer. Each teammate is writing their own version of the brand into their own private prompt stack.

Prompt decay

Every edit to a saved prompt erodes the original intent. A prompt that was carefully constructed in February ends up rewritten five times by May, with no one tracking which version was canonical. The decay is silent and it compounds.

The audit gap

A piece of copy goes live and lands wrong. You want to trace it back to the prompt that produced it. ChatGPT cannot tell you which prompt was used, which version or which teammate ran it. The chain of custody breaks at the moment of generation.

The handoff break

The teammate who built your prompt setup leaves. The institutional knowledge of which prompts work, why they are structured the way they are and what edge cases they handle leaves with them. The next teammate inherits a Notion page and no operating manual.

Why AI writing sounds off-brand walks through the same failure pattern from the brand voice angle, with more detail on the human side of the breakdown.

15 ChatGPT prompts for marketing

Below are 15 ChatGPT prompts for marketing that work across most use cases. Each one is structured to accept a brand voice block: a short paragraph of rules describing how your brand sounds, what words it uses and what it avoids. The placeholder text is [BRAND VOICE RULES]. Replace it with your own rules before running.

The prompts work better when the brand voice block is real. Generic instructions produce generic output. Specific instructions produce on-brand output.

1. Audience research

Use case: building a working profile of a target audience before campaign planning.

You are a senior marketing strategist. Build an audience research brief for the following segment: [SEGMENT]. Cover demographic profile, top three pain points, top three buying triggers, top three objections, and three publications or communities this audience reads. Apply these brand voice rules to all output: [BRAND VOICE RULES]

You are a senior marketing strategist. Build an audience research brief for the following segment: [SEGMENT]. Cover demographic profile, top three pain points, top three buying triggers, top three objections, and three publications or communities this audience reads. Apply these brand voice rules to all output: [BRAND VOICE RULES]

You are a senior marketing strategist. Build an audience research brief for the following segment: [SEGMENT]. Cover demographic profile, top three pain points, top three buying triggers, top three objections, and three publications or communities this audience reads. Apply these brand voice rules to all output: [BRAND VOICE RULES]

You are a senior marketing strategist. Build an audience research brief for the following segment: [SEGMENT]. Cover demographic profile, top three pain points, top three buying triggers, top three objections, and three publications or communities this audience reads. Apply these brand voice rules to all output: [BRAND VOICE RULES]

Good output is structured, specific and free of marketing jargon. Watch for generic phrasing like "savvy professionals" that signals the model fell back to its defaults.

2. Buyer persona

Use case: producing a one-page persona document.

Draft a buyer persona for [PERSONA NAME] at [COMPANY TYPE]. Include role, daily responsibilities, top three goals, top three blockers, preferred information sources, and one quote that captures their relationship to [PROBLEM YOU SOLVE]. Apply these brand voice rules: [BRAND VOICE RULES]

Draft a buyer persona for [PERSONA NAME] at [COMPANY TYPE]. Include role, daily responsibilities, top three goals, top three blockers, preferred information sources, and one quote that captures their relationship to [PROBLEM YOU SOLVE]. Apply these brand voice rules: [BRAND VOICE RULES]

Draft a buyer persona for [PERSONA NAME] at [COMPANY TYPE]. Include role, daily responsibilities, top three goals, top three blockers, preferred information sources, and one quote that captures their relationship to [PROBLEM YOU SOLVE]. Apply these brand voice rules: [BRAND VOICE RULES]

Draft a buyer persona for [PERSONA NAME] at [COMPANY TYPE]. Include role, daily responsibilities, top three goals, top three blockers, preferred information sources, and one quote that captures their relationship to [PROBLEM YOU SOLVE]. Apply these brand voice rules: [BRAND VOICE RULES]

The quote is the test. If it sounds like a real person, the persona is usable. If it sounds like a slogan, run the prompt again with more constraint.

3. Headline generation

Use case: producing headline variants for a blog post or landing page.

Write 20 headline variants for an article on [TOPIC]. The audience is [AUDIENCE]. The angle is [ANGLE]. Apply these brand voice rules: [BRAND VOICE RULES]. Number each variant. Do not use exclamation marks or em dashes

Write 20 headline variants for an article on [TOPIC]. The audience is [AUDIENCE]. The angle is [ANGLE]. Apply these brand voice rules: [BRAND VOICE RULES]. Number each variant. Do not use exclamation marks or em dashes

Write 20 headline variants for an article on [TOPIC]. The audience is [AUDIENCE]. The angle is [ANGLE]. Apply these brand voice rules: [BRAND VOICE RULES]. Number each variant. Do not use exclamation marks or em dashes

Write 20 headline variants for an article on [TOPIC]. The audience is [AUDIENCE]. The angle is [ANGLE]. Apply these brand voice rules: [BRAND VOICE RULES]. Number each variant. Do not use exclamation marks or em dashes

The "do not use" instructions matter. Without them, ChatGPT defaults to its own punctuation habits, which are not yours.

4. Blog post outline

Use case: building an H1 through H3 outline before writing.

Build a detailed outline for a blog post titled [WORKING TITLE]. The target keyword is [KEYWORD]. The audience is [AUDIENCE]. Include H1, six to eight H2 sections with H3 subsections where useful, a short note on the angle of each section, and a target word count per section. Apply these brand voice rules: [BRAND VOICE RULES]

Build a detailed outline for a blog post titled [WORKING TITLE]. The target keyword is [KEYWORD]. The audience is [AUDIENCE]. Include H1, six to eight H2 sections with H3 subsections where useful, a short note on the angle of each section, and a target word count per section. Apply these brand voice rules: [BRAND VOICE RULES]

Build a detailed outline for a blog post titled [WORKING TITLE]. The target keyword is [KEYWORD]. The audience is [AUDIENCE]. Include H1, six to eight H2 sections with H3 subsections where useful, a short note on the angle of each section, and a target word count per section. Apply these brand voice rules: [BRAND VOICE RULES]

Build a detailed outline for a blog post titled [WORKING TITLE]. The target keyword is [KEYWORD]. The audience is [AUDIENCE]. Include H1, six to eight H2 sections with H3 subsections where useful, a short note on the angle of each section, and a target word count per section. Apply these brand voice rules: [BRAND VOICE RULES]

The angle notes per section are the part that saves time. They turn the outline into a writing brief.

5. Landing page hero

Use case: drafting the headline and subhead for a landing page.

Write three headline-subhead pairs for a landing page targeting [AUDIENCE] who want to [OUTCOME]. Each pair: one headline under nine words, one subhead under 20 words. Apply these brand voice rules: [BRAND VOICE RULES]

Write three headline-subhead pairs for a landing page targeting [AUDIENCE] who want to [OUTCOME]. Each pair: one headline under nine words, one subhead under 20 words. Apply these brand voice rules: [BRAND VOICE RULES]

Write three headline-subhead pairs for a landing page targeting [AUDIENCE] who want to [OUTCOME]. Each pair: one headline under nine words, one subhead under 20 words. Apply these brand voice rules: [BRAND VOICE RULES]

Write three headline-subhead pairs for a landing page targeting [AUDIENCE] who want to [OUTCOME]. Each pair: one headline under nine words, one subhead under 20 words. Apply these brand voice rules: [BRAND VOICE RULES]

Three pairs is the right number. Two feels thin. Five is too many to choose from in one session.

6. Email subject line

Use case: producing subject line variants for a campaign.

Write 15 subject lines for an email about [TOPIC] to [AUDIENCE]. The desired action is [ACTION]. Apply these brand voice rules: [BRAND VOICE RULES]. Number each. Each subject line must be under 60 characters

Write 15 subject lines for an email about [TOPIC] to [AUDIENCE]. The desired action is [ACTION]. Apply these brand voice rules: [BRAND VOICE RULES]. Number each. Each subject line must be under 60 characters

Write 15 subject lines for an email about [TOPIC] to [AUDIENCE]. The desired action is [ACTION]. Apply these brand voice rules: [BRAND VOICE RULES]. Number each. Each subject line must be under 60 characters

Write 15 subject lines for an email about [TOPIC] to [AUDIENCE]. The desired action is [ACTION]. Apply these brand voice rules: [BRAND VOICE RULES]. Number each. Each subject line must be under 60 characters

The character limit is non-negotiable. Without it, ChatGPT happily writes 80-character subject lines that get truncated in every major email client.

7. Email body

Use case: drafting the body of a marketing email.

Draft a [LENGTH] email for [AUDIENCE]. The topic is [TOPIC]. The desired action is [ACTION]. Lead with a one-sentence hook tied to a real reader frustration. Apply these brand voice rules: [BRAND VOICE RULES]. End with a single call to action. No PS line unless the brand voice rules allow it

Draft a [LENGTH] email for [AUDIENCE]. The topic is [TOPIC]. The desired action is [ACTION]. Lead with a one-sentence hook tied to a real reader frustration. Apply these brand voice rules: [BRAND VOICE RULES]. End with a single call to action. No PS line unless the brand voice rules allow it

Draft a [LENGTH] email for [AUDIENCE]. The topic is [TOPIC]. The desired action is [ACTION]. Lead with a one-sentence hook tied to a real reader frustration. Apply these brand voice rules: [BRAND VOICE RULES]. End with a single call to action. No PS line unless the brand voice rules allow it

Draft a [LENGTH] email for [AUDIENCE]. The topic is [TOPIC]. The desired action is [ACTION]. Lead with a one-sentence hook tied to a real reader frustration. Apply these brand voice rules: [BRAND VOICE RULES]. End with a single call to action. No PS line unless the brand voice rules allow it

The PS instruction matters. ChatGPT adds PS lines by default. Most brand voices do not use them.

8. LinkedIn post

Use case: drafting a LinkedIn post tied to a recent piece of content.

Write a LinkedIn post promoting [CONTENT]. Open with a one-sentence hook that is a counter-intuitive claim or a real observation, not a question. Use short paragraphs of one to two sentences. Apply these brand voice rules: [BRAND VOICE RULES]. End with one question for the audience. No hashtags unless the brand voice rules allow them

Write a LinkedIn post promoting [CONTENT]. Open with a one-sentence hook that is a counter-intuitive claim or a real observation, not a question. Use short paragraphs of one to two sentences. Apply these brand voice rules: [BRAND VOICE RULES]. End with one question for the audience. No hashtags unless the brand voice rules allow them

Write a LinkedIn post promoting [CONTENT]. Open with a one-sentence hook that is a counter-intuitive claim or a real observation, not a question. Use short paragraphs of one to two sentences. Apply these brand voice rules: [BRAND VOICE RULES]. End with one question for the audience. No hashtags unless the brand voice rules allow them

Write a LinkedIn post promoting [CONTENT]. Open with a one-sentence hook that is a counter-intuitive claim or a real observation, not a question. Use short paragraphs of one to two sentences. Apply these brand voice rules: [BRAND VOICE RULES]. End with one question for the audience. No hashtags unless the brand voice rules allow them

LinkedIn rewards short paragraphs and specific hooks. The instruction against questions as openers stops the model from defaulting to "Have you ever wondered..." which signals AI immediately.

9. X post

Use case: drafting a post for X about a content piece or product update. Prompts eight and nine together cover ChatGPT prompts for social media marketing across the two most-used channels.

Write five X posts about [TOPIC]. Each post is one to three sentences. No hashtags. No emojis. Apply these brand voice rules: [BRAND VOICE RULES]. The posts should work as standalone or as the first post in a thread

Write five X posts about [TOPIC]. Each post is one to three sentences. No hashtags. No emojis. Apply these brand voice rules: [BRAND VOICE RULES]. The posts should work as standalone or as the first post in a thread

Write five X posts about [TOPIC]. Each post is one to three sentences. No hashtags. No emojis. Apply these brand voice rules: [BRAND VOICE RULES]. The posts should work as standalone or as the first post in a thread

Write five X posts about [TOPIC]. Each post is one to three sentences. No hashtags. No emojis. Apply these brand voice rules: [BRAND VOICE RULES]. The posts should work as standalone or as the first post in a thread

X has its own voice. The brand voice rules still apply, but the length constraints force the model to compress harder than other channels.

10. SEO meta description

Use case: writing meta descriptions for a batch of pages.

Write a meta description for a page titled [TITLE] targeting the keyword [KEYWORD]. The audience is [AUDIENCE]. Include the keyword once, naturally. Maximum 155 characters. Apply these brand voice rules: [BRAND VOICE RULES]

Write a meta description for a page titled [TITLE] targeting the keyword [KEYWORD]. The audience is [AUDIENCE]. Include the keyword once, naturally. Maximum 155 characters. Apply these brand voice rules: [BRAND VOICE RULES]

Write a meta description for a page titled [TITLE] targeting the keyword [KEYWORD]. The audience is [AUDIENCE]. Include the keyword once, naturally. Maximum 155 characters. Apply these brand voice rules: [BRAND VOICE RULES]

Write a meta description for a page titled [TITLE] targeting the keyword [KEYWORD]. The audience is [AUDIENCE]. Include the keyword once, naturally. Maximum 155 characters. Apply these brand voice rules: [BRAND VOICE RULES]

The character cap is what makes this prompt useful. Without it, the model writes 200-character descriptions that Google truncates.

11. Competitor positioning

Use case: comparing your positioning against a named competitor.

Analyze the positioning of [COMPETITOR] based on their homepage at [URL]. Identify their main value proposition, target customer, and three claims they make most prominently. Compare to my brand positioning: [YOUR POSITIONING]. Return three positioning angles I could own that they cannot. Apply these brand voice rules: [BRAND VOICE RULES]

Analyze the positioning of [COMPETITOR] based on their homepage at [URL]. Identify their main value proposition, target customer, and three claims they make most prominently. Compare to my brand positioning: [YOUR POSITIONING]. Return three positioning angles I could own that they cannot. Apply these brand voice rules: [BRAND VOICE RULES]

Analyze the positioning of [COMPETITOR] based on their homepage at [URL]. Identify their main value proposition, target customer, and three claims they make most prominently. Compare to my brand positioning: [YOUR POSITIONING]. Return three positioning angles I could own that they cannot. Apply these brand voice rules: [BRAND VOICE RULES]

Analyze the positioning of [COMPETITOR] based on their homepage at [URL]. Identify their main value proposition, target customer, and three claims they make most prominently. Compare to my brand positioning: [YOUR POSITIONING]. Return three positioning angles I could own that they cannot. Apply these brand voice rules: [BRAND VOICE RULES]

The "angles I could own that they cannot" framing is what makes the output strategically useful. Without it, the output reads like a competitive analysis from a freshman marketing class.

12. Customer interview guide

Use case: building a discovery interview guide.

Build a 12-question customer interview guide for a conversation with [CUSTOMER SEGMENT] about their use of [PRODUCT CATEGORY]. Cover: current workflow, biggest frustration, tools they tried and dropped, what triggered the search for a new tool, evaluation criteria, decision process. Apply these brand voice rules: [BRAND VOICE RULES]. Phrase questions to elicit stories, not yes-or-no answers

Build a 12-question customer interview guide for a conversation with [CUSTOMER SEGMENT] about their use of [PRODUCT CATEGORY]. Cover: current workflow, biggest frustration, tools they tried and dropped, what triggered the search for a new tool, evaluation criteria, decision process. Apply these brand voice rules: [BRAND VOICE RULES]. Phrase questions to elicit stories, not yes-or-no answers

Build a 12-question customer interview guide for a conversation with [CUSTOMER SEGMENT] about their use of [PRODUCT CATEGORY]. Cover: current workflow, biggest frustration, tools they tried and dropped, what triggered the search for a new tool, evaluation criteria, decision process. Apply these brand voice rules: [BRAND VOICE RULES]. Phrase questions to elicit stories, not yes-or-no answers

Build a 12-question customer interview guide for a conversation with [CUSTOMER SEGMENT] about their use of [PRODUCT CATEGORY]. Cover: current workflow, biggest frustration, tools they tried and dropped, what triggered the search for a new tool, evaluation criteria, decision process. Apply these brand voice rules: [BRAND VOICE RULES]. Phrase questions to elicit stories, not yes-or-no answers

The "stories, not yes-or-no" instruction is the difference between a useful interview and a boring one.

13. Press release draft

Use case: drafting a press release for a launch or company news.

Draft a press release announcing [NEWS]. Open with a headline under 12 words and a two-line subhead. First paragraph covers the who, what, when, where, why. Include two quotes: one from [SPOKESPERSON] and one from [CUSTOMER OR PARTNER]. Apply these brand voice rules to body copy: [BRAND VOICE RULES]. Quotes should sound like people, not press releases

Draft a press release announcing [NEWS]. Open with a headline under 12 words and a two-line subhead. First paragraph covers the who, what, when, where, why. Include two quotes: one from [SPOKESPERSON] and one from [CUSTOMER OR PARTNER]. Apply these brand voice rules to body copy: [BRAND VOICE RULES]. Quotes should sound like people, not press releases

Draft a press release announcing [NEWS]. Open with a headline under 12 words and a two-line subhead. First paragraph covers the who, what, when, where, why. Include two quotes: one from [SPOKESPERSON] and one from [CUSTOMER OR PARTNER]. Apply these brand voice rules to body copy: [BRAND VOICE RULES]. Quotes should sound like people, not press releases

Draft a press release announcing [NEWS]. Open with a headline under 12 words and a two-line subhead. First paragraph covers the who, what, when, where, why. Include two quotes: one from [SPOKESPERSON] and one from [CUSTOMER OR PARTNER]. Apply these brand voice rules to body copy: [BRAND VOICE RULES]. Quotes should sound like people, not press releases

The "quotes should sound like people" instruction is critical. Without it, every quote in the press release sounds the same.

14. Sales objection response

Use case: building a response library for common sales objections.

A prospect raises the objection: [OBJECTION]. Write three response approaches. Each approach is two to three sentences. The first response leans on a customer story, the second on a product capability, the third on a reframe of the objection itself. Apply these brand voice rules: [BRAND VOICE RULES]

A prospect raises the objection: [OBJECTION]. Write three response approaches. Each approach is two to three sentences. The first response leans on a customer story, the second on a product capability, the third on a reframe of the objection itself. Apply these brand voice rules: [BRAND VOICE RULES]

A prospect raises the objection: [OBJECTION]. Write three response approaches. Each approach is two to three sentences. The first response leans on a customer story, the second on a product capability, the third on a reframe of the objection itself. Apply these brand voice rules: [BRAND VOICE RULES]

A prospect raises the objection: [OBJECTION]. Write three response approaches. Each approach is two to three sentences. The first response leans on a customer story, the second on a product capability, the third on a reframe of the objection itself. Apply these brand voice rules: [BRAND VOICE RULES]

Three approaches gives the salesperson flexibility. The reframe option is the one most teams forget to build, and it is often the strongest move.

15. Long-form repurposing

Use case: turning a blog post into social posts and email content.

Read the article at [URL]. Produce the following derivatives: one LinkedIn post, three X posts, one email teaser of 80 words or fewer, one Instagram caption. Each derivative pulls a different angle from the article. Apply these brand voice rules to all outputs: [BRAND VOICE RULES]

Read the article at [URL]. Produce the following derivatives: one LinkedIn post, three X posts, one email teaser of 80 words or fewer, one Instagram caption. Each derivative pulls a different angle from the article. Apply these brand voice rules to all outputs: [BRAND VOICE RULES]

Read the article at [URL]. Produce the following derivatives: one LinkedIn post, three X posts, one email teaser of 80 words or fewer, one Instagram caption. Each derivative pulls a different angle from the article. Apply these brand voice rules to all outputs: [BRAND VOICE RULES]

Read the article at [URL]. Produce the following derivatives: one LinkedIn post, three X posts, one email teaser of 80 words or fewer, one Instagram caption. Each derivative pulls a different angle from the article. Apply these brand voice rules to all outputs: [BRAND VOICE RULES]

Pulling different angles is what makes the derivatives feel like content, not summaries of the article.

The prompt library trap

The natural next move after a list like the one above is to save the prompts into a shared doc. A team prompt library. Most marketing teams build one within a month of going serious on ChatGPT.

The library helps for the first quarter and then it breaks. Prompts get copied and edited until no one knows which version is canonical. Two teammates fork the same prompt for two different uses and both forks survive. The brand voice rules embedded in the prompts go stale because they live in 20 different prompts and no one updates all 20 when the rules change. The library becomes a graveyard of half-current prompts that nobody fully trusts.

Prompt libraries versus voice infrastructure covers the failure pattern in full, but the short version is this: a prompt library is a storage system, not a governance system. Storing the prompt does not store the voice.

The layer underneath

The fix is to move brand voice out of the prompt and into the layer underneath the prompt. That layer is called voice infrastructure, the system of documents, tools and processes that codify how a brand sounds and make it reusable across teams and AI models.

Voice infrastructure operates on three layers. Foundation covers the brand voice principles like "confident, not arrogant." Codification covers the voice chart, brand lexicon and style guide that translate those principles into specific word choices. Activation covers the templates, prompts and AI voice models that apply the rules at the point of writing.

The 15 prompts above sit inside the Activation layer. They produce on-brand output only when the Foundation and Codification layers exist and are connected. Without those upper layers, the prompts are an activation system trying to run on rules that live in someone's head.

Brivvy is one example of voice infrastructure built for marketing teams. It takes the brand voice rules from the Foundation and Codification layers and makes them available to ChatGPT, Claude and other AI tools through the Model Context Protocol (MCP). The prompts above use a [BRAND VOICE RULES] placeholder. Voice infrastructure is what makes that placeholder actually hold across every prompt, every teammate and every session.

For a broader look at brand voice as a category, see what is brand voice.

Claude versus ChatGPT for marketing

The natural follow-up question for any marketing team running ChatGPT is whether Claude is a better fit. The short answer is that both work, and the gap that matters is not between the two models. Claude for marketing has become its own category as Anthropic's market share has grown, but the same brand voice problem shows up on both tools.

ChatGPT is faster, broader and more capable on quick-turn tasks. Claude is stronger on long-form output and complex reasoning, with a 200,000 token context window that handles entire campaign briefs or competitor analyses in a single pass. Most marketing teams use both. ChatGPT for the high-volume work, Claude for the long-form and strategic work.

Both have the same fundamental gap for marketing use. Neither enforces brand voice at generation. The 15 prompts above work with Claude as well as they work with ChatGPT, and both improve when the brand voice rules live in voice infrastructure rather than inside each prompt. The AI stack for startups in 2026 covers the comparison across eight tools, including both.

Frequently asked questions

What now

Run three checks on your current ChatGPT for marketing setup. First, audit the last 20 pieces of AI-generated content for voice consistency. Second, count the number of saved prompts across your team and ask whether anyone knows which version is canonical. Third, try one of the 15 prompts above with your real brand voice rules in the [BRAND VOICE RULES] placeholder, and notice the difference in output. The drift you find in the first two checks is the layer beneath. The improvement you see in the third is what fixing that layer looks like.