Grok 4 is the smartest AI, but is it a good marketer? (Full Test)

We put Grok 4 through its marketing paces with three use cases designed to showcase the AI model's strengths.

I put Grok 4 through its marketing paces with two use cases designed to showcase the AI model's strengths for marketing.

Grok 4 was released last week, and X was alive with hype - "it's nuts", "insane", "game changer", and all the other lingo that works great on YouTube.

It certainly performs incredibly well against all benchmarks.

But for all of us who want to use AI to market and grow our businesses, is it truly "insane"?

The following use cases test Grok 4 against its most powerful attributes: reasoning and access to X data.

You can apply all of these in various use cases to your marketing via other LLMs, but for this post, we will exclusively run them on Grok4.

I’ll give the summary below on my assessment of Grok 4 for marketing.

Note: If you’re not on the paid subscriber list, next weeks paid tutorial is how to create a virtual, world-class team, to help and critique your work. It’s a great tutorial.

Also if you’re not signed up for free, upcoming posts include

• Turn ChatGPT in a super power with context + memory

• Veo 3 Tutorial

• The AI Marketing Strategy

• And much more

Let’s get back to this post.

1. Mimic a Creator's Style

Here's a great prompt that leverages an AI use case I love, along with some of Grok 4's core strengths. It creates a writing style based on X/twitter posts of a creator you select.

It will research a select group of X posts within a specific engagement range and, based on that, create a writing style ruleset. It then uses that ruleset to craft a piece of content on a topic and format your pick.

Let's get to the prompt

---

PROMPT

// =========================================================

// SINGLE-SHOT STAGED PROMPT — Creator-Style Content Engine

// =========================================================

// CRITICAL: Global writing rule keeps output tight & human-friendly.

<GLOBAL_WRITING_RULE>

Write in plain, conversational English—short, clear sentences with no jargon, clichés, hype, or fluff—and state requests directly and honestly.

</GLOBAL_WRITING_RULE>

// ---------- USER VARIABLES (edit these only) --------------

<INPUTS>

Add the user to input the following variables. Only move onto next step after the user has added all variables.

CREATOR_HANDLE = "{CREATOR_HANDLE}" // X/Twitter handle

TIME_RANGE_HOURS = {TIME_RANGE_HOURS} // how far back to pull tweets

MAX_TWEETS_TO_ANALYSE = {MAX_TWEETS_TO_ANALYSE} // cap for token safety

MIN_RETWEETS = {MIN_RETWEETS} // engagement threshold (adjust)

TOPIC = {TOPIC}

FORMAT = {FORMAT} // newsletter | twitter_thread | linkedin_post

</INPUTS>

// ----------------------------------------------------------

// CRITICAL: Three fenced stages isolate logic yet share context.

<STAGE 1 — RESEARCH (silent)>

1. Retrieve all tweets by {CREATOR_HANDLE} posted within the past {TIME_RANGE_HOURS} h (max {MAX_TWEETS_TO_ANALYSE}).

2. Keep only tweets with ≥ {MIN_RETWEETS} retweets.

3. For each kept tweet capture:

• tweet_id • text • retweets • replies • char_count

4. Sort primarily by descending char_count, secondarily by retweets.

5. OUTPUT a compact CSV code-block labeled DATA_START … DATA_END for internal use only.

</STAGE 1>

// CRITICAL: Store rules in working memory for later recall.

<STAGE 2 — STYLE-RULE EXTRACTION (silent)>

1. From the top 10 tweets in the sorted list, derive a **4-bullet Style Rule Set** with these headings:

• Tone

• Syntax quirks

• Rhetorical devices

• Common hooks

2. **Self-check:** Are the four rules distinct and non-overlapping?

• If **Yes**, continue.

• If **No**, refine the rule set until Yes. (Loop max 3 times.)

3. **SAVE** the final rule set to working memory under the label **STYLE_RULES**.

</STAGE 2>

// CRITICAL: Explicit recall of STYLE_RULES ensures transfer.

<STAGE 3 — CONTENT GENERATION>

Goal   : Write a {FORMAT} on the topic "{TOPIC}" while mirroring STYLE_RULES.

Guidelines:

1. Start with a compelling title/headline hook.

2. Use STYLE_RULES consistently for voice and structure.

3. Add smooth transitions, emotional triggers, and actionable insights.

4. **Avoid hashtags, emojis, and exclamation marks** unless STYLE_RULES explicitly requires them. // CRITICAL: Negative constraint

5. End with a strong CTA relevant to marketers.

<OUTPUT_SCHEMA>

{

"style_rules": ["…4 bullets copied verbatim…"],

"draft": {

"title": "...",

"body_md": "...", // markdown-formatted main content

"cta": "..."

}

}

</OUTPUT_SCHEMA>

</STAGE 3>

// CRITICAL: Model must output **only** the JSON object above—no tweets, no intermediary notes.

---

It’s a very fun prompt. Here’s the output for the creator Shaan Puri:

The style rules are very valuable. One of the fun things to do is to combine creators to create a mix of people's styles.

As always, never cut and paste someone else's style, but they can help you refine your own.

This prompt is a great test for Grok 4, specifically in reasoning, multi-step analysis, and analyzing X's data.

The content it created is fine. I believe the value is in how it can extract a creators writing style allowing you to play around with different styles to create one of your own.

2. Create Marketing Campaign

The next test for Grok is its ability to craft the nugget of a world-class marketing campaign.

Why do I say nugget?

AI isn't going to create an idea you'll ship, or at least one that you'll ship and rise above the noise. It's anchored to its training set so struggles with new/innovative ideas. But, it can give you the jumping off point to a great idea, a great piece of content, or a great campaign.

This prompt is designed to be your marketing campaign assistant, providing quick-fire ideas that you can turn into fully fledged marketing campaigns.

Note: I’m using fake data as my company and persona

---

PROMPT

// ====================================================

// GROK 4 – ONE-SHOT CAMPAIGN IDEATOR & PLANNER

// (edit <CONTEXT>; leave everything else)

// ====================================================

// ---------- 1) USER-SUPPLIED CONTEXT ----------

<CONTEXT>

OFFER = "ClipSpark instantly turns long webinars into 30-second, captions-on social videos, letting lean SaaS marketing teams post daily without editors or contractors."

PERSONA_SNAPSHOT = "A 32-year-old solo demand-gen marketer at a Series-A SaaS; drowning in webinar recordings; craves a constant LinkedIn presence but lacks video-editing skills or time."

ACTION_AND_KPI = "Generate 250 self-serve trial activations in 10 days; aim for 25 % upgrade-to-paid within 30 days."

BUDGET_TIMEBOX = "€8 000 total media spend from 1 Aug 2025 to 10 Aug 2025 (inclusive)."

BRAND_VOICE_SAMPLE = "Bold, cheeky, clarity-obsessed. Example: “Your webinars shouldn’t rot in a content graveyard—zap them into binge-worthy clips before tomorrow’s coffee gets cold and watch your pipeline perk up all day.”"

</CONTEXT>

// ----------------------------------------------

// ========== 2) MODEL ROLE & GLOBAL TONE ==========

<ROLE>

You are a veteran product strategist, conversion-focused copywriter, and growth analyst.

</ROLE>

<GLOBAL_WRITING_RULE>

Write in plain, conversational English—short, clear sentences; zero jargon or fluff.

</GLOBAL_WRITING_RULE>

// =================================================

// ========== 2.5) BREAKOUT CAMPAIGN RULE-SET ==========

// ── Each concept MUST include concrete answers for every token below ──

<RULE_SET>

1. CULTURAL TENSION → [TENSION]

// Live debate, fear, or aspiration in the ICP’s world (e.g. “Meetings are killing creativity”).

2. HERO STATEMENT → [HERO_PHRASE]

// Two- to four-word rally-cry that frames the tension (e.g. “Dream Crazy”, “Own Your Data”).

3. SIGNATURE ARTIFACT → [ARTIFACT]

// Tangible/digital shareable object (badge, bottle, filter, leaderboard card).

4. PERSONALISATION LOOP → [PERSONAL_FIELD]

// How the artifact becomes uniquely “theirs” (name, personal stats, custom colours).

5. PUBLIC SCOREBOARD → [CHALLENGE_MECH]

// Visible ranking or recognition (hashtag tally, daily top-5 feed, interactive map).

6. EMOTIONAL STORY → [STORY_BEAT]

// Narrative linking product benefit to identity, justice, or pride.

7. DISTINCTIVE VISUAL CODE → [VIS_CODE]

// Instantly recognisable colour, motif, or layout (Spotify green waveforms, Barbie pink).

8. TIME-BOUND URGENCY → [DEADLINE]

// Compressed timeframe (48-h drop, one-week challenge, annual reveal).

9. COMMUNITY CATALYSTS → [CATALYST_LIST]

// Influencers or power users seeded first to amplify buzz.

10. FRICTIONLESS CTA → [CTA_LINK]

// One-click path from curiosity to trial/purchase—no extra hoops.

</RULE_SET>

// ===============================================================

// ========== 3) TASK FLOW & OUTPUT SPEC ==========

<OUTPUT_REQUEST>

// ---------- Stage A – Campaign Ideation (internal) ----------

• Brainstorm 3 bold concepts tailored to OFFER & PERSONA_SNAPSHOT.

• **Do NOT use the product name or feature words (clip, trim, edit, video) in titles or descriptions.** // <- NEGATIVE CONSTRAINT

• Explicitly weave ALL 10 Rule-Set elements into every concept.

• Draw on at least one creative seed: guerrilla stunt, irony, nostalgia, or odd partnership. // <- INSPIRATION SEED

• Score each on Relevance, Viral Potential, Feasibility, Thumb-stop Power (1-5).

• If the top idea’s HERO_PHRASE or ARTIFACT echoes the product’s core feature, discard it and pick the next. // <- ORIGINALITY GATE

• Select the highest-scoring compliant idea.

// ---------- Stage B – Final Output (visible) ----------

Return exactly seven sections with Markdown headings.

Keep any prose block ≤ 120 words.

1. **Campaign Big Idea**

• Title (≤ 10 words), 50-word description, three-word hook trio.

• **Campaign DNA** table mapping the 10 Rule-Set tokens to their concrete execution.

2. **Problem & Persona Lens**

• Two-sentence pain summary.

• 30-word day-in-the-life vignette *as the user*.

3. **Key Differentiators**

• Bullet list (≤ 5) linking OFFER to the Big Idea.

4. **Risks & Mitigations**

• Bullet list (≤ 5) of red flags + one-line fixes.

5. **One-Week Validation Plan**

| Channel | Traffic goal | CTR % | Target actions | Cost (€) |

End with ROI formula: `(actions × LTV – cost) ÷ cost`.

6. **Messaging Kit**

• Three 120-character ad headlines—each for a different nuance in PERSONA_SNAPSHOT.

• Maintain BRAND_VOICE_SAMPLE tone.

7. **Channel-Mix Budget**

| Channel | % of budget | Rationale (≤ 20 words) |

Table must sum to 100 %.

</OUTPUT_REQUEST>

// ===============================================================

// ========== 4) OPERATIONAL GUARDS ==========

<INTERNAL_CHECKS>

• Do **not** reveal Stage A brainstorming, scoring, or chain-of-thought.

• Reference every CONTEXT field explicitly.

• Validate maths before output.

• Output only sections 1-7—nothing else.

</INTERNAL_CHECKS>

// ===============================================================

---

This prompt is good. It uses a ruleset defined from some of the best marketing campaigns run over the past decade and then attempts to weave them into the campaign it creates, e.g., here is what it came up with for me (I used fake data for a fake company)

Here is an example of the output (just showing the campaign idea and ruleset)

In terms of the output from this prompt, everything other than the campaign idea is valuable. The campaign idea is weak. I’ve been iterating on the prompt for a number of hours to try force it to get more creative e.g.:

// ---------- Stage A – Campaign Ideation (internal) ---------- • Brainstorm 3 bold concepts tailored to OFFER & PERSONA_SNAPSHOT.

• **Do NOT use the product name or feature words (clip, trim, edit, video) in titles or descriptions.** // <- NEGATIVE CONSTRAINT

• Explicitly weave ALL 10 Rule-Set elements into every concept.

• Draw on at least one creative seed: guerrilla stunt, irony, nostalgia, or odd partnership. // <- INSPIRATION SEED

• Score each on Relevance, Viral Potential, Feasibility, Thumb-stop Power (1-5).

• If the top idea’s HERO_PHRASE or ARTIFACT echoes the product’s core feature, discard it and pick the next. // <- ORIGINALITY GATE • Select the highest-scoring compliant idea.

I believe I can get it to be much more creative. I suspect one of the issues is the lack of context e.g. I used fake context, it’s much better if you upload your real company data.

But, sometimes when you’re in the AI lab, you iterate for a long-time before you get it right. This is what real prompting looks like vs. what yo see on X where it seems like every single prompt you run = an insane result.

The campaign prompt is a strong foundation to build on. It’s motivated me to do an entire post on how I can have AI help create tasteful marketing.

The Summary:

I planned to get a third use case done to demonstrate Grok's capabilities by creating a micro-influencer campaign for your product, e.g., identifying micro-influencers for your product and creating a campaign they can promote. I'll do this in a future post. The campaign prompt above took me much longer than I had planned for.

The TL;DR for Grok 4, based on all the use cases I've run, is that it's fine. Nothing more than fine. It speaks to something important. In many cases, an AI model's performance against industry benchmarks doesn't necessarily reflect its effectiveness for Marketing, GTM, and knowledge worker use cases e.g. what the average person wants to do with the models.

Until Next Time,

Happy AI’fying

Kieran