By the end of this lesson you will be able to
- Use AI for the parts of marketing where it genuinely helps
- Avoid the sameness trap that makes AI content invisible
- Build a reusable voice brief instead of re-describing your tone every time
Marketing got the biggest immediate productivity gain from AI and the biggest strategic problem. Producing content is now nearly free, which means producing content is worth nearly nothing. What is scarce is having something specific to say.
The sameness trap
Ask any model for "a LinkedIn post about the importance of customer service" and you get the same post your competitor got. Not similar — structurally identical, because you both asked an averaging machine for the average answer.
Produces beige
- "Write a blog post about X"
- No specific claim to make
- Tone described in adjectives
- No examples of your own work
- Published as generated
Produces something yours
- "Here is what I think about X, and why. Structure it."
- One specific, arguable claim
- Tone shown via two of your own posts
- Your real numbers, names, and stories
- Edited until it sounds like a person
Build a voice brief once
The highest-leverage thing a marketing team can do is write one voice brief and paste it into every prompt. Describing your tone from scratch each time produces inconsistency and wastes the part of the prompt that matters.
We are [company], [what we do] for [who].
Our voice:
- [adjective], [adjective], never [adjective to avoid]
- We use short sentences and concrete nouns
- We do not use: "leverage", "seamless", "game-changing",
"in today's fast-paced world", em-dash-heavy constructions
- We say "you", not "our clients"
- We admit what we do not do
Here are two pieces that sound exactly right:
[PASTE TWO OF YOUR BEST]
Match that voice and structure.Where it genuinely helps
| Task | Fit | Note |
|---|---|---|
| Repurposing one piece into five formats | Excellent | Source material is right there |
| Turning a customer call into a case study draft | Excellent | Their words, your structure |
| Twenty subject lines to react to | Excellent | Volume as raw material, not output |
| Editing your draft for length and clarity | Very good | Better editor than writer |
| Alt text and meta descriptions at scale | Very good | Tedious, mechanical, low risk |
| Structuring a piece you have already thought about | Very good | Outlines are its strength |
| Writing the original insight | Poor | It averages; insight is the opposite |
| Anything with a statistic in it | Risky | Verify or delete. It invents these |
Repurposing: the best return in the function
You wrote one good thing. Most teams then publish it once. Turning it into five formats used to cost half a day and now costs ten minutes, which changes the economics of ever writing anything long.
[VOICE BRIEF]
Here is a piece we published:
[PASTE THE ARTICLE]
Produce:
1. A LinkedIn post — the single most useful idea, 120 words,
no hook question, no "Here's the thing:"
2. Three short posts, each making one specific point
3. A 4-sentence email newsletter intro linking to it
4. Five subject lines, none of them clickbait
Rules:
- Every output must contain a specific detail from the piece,
not a summary of its theme
- Do not add statistics or claims that are not in the sourceCase studies from customer calls
The case study everyone means to write and nobody writes. Record a fifteen-minute customer conversation (with permission), paste the transcript, and you have a draft.
Ask about the before, not the after
"What was it like before?" produces the material that makes a case study persuasive. Most people only ask about results, which produces a testimonial rather than a story.
Keep their words
Instruct explicitly: quote the customer verbatim where possible rather than paraphrasing. The paraphrased version always sounds like marketing, because it is.
Ban invented numbers
"Use only figures the customer stated. If a metric would strengthen the piece but was not given, mark it [NEED FIGURE]." This is the single most important line in the prompt.
Send it to them before publishing
Not just for approval — they will correct details, and the corrections usually make the piece better.
A note on SEO content
Mass-producing keyword-targeted articles was briefly effective and is now actively risky — search engines have adjusted, and a site full of thin AI pages can damage the pages that were genuinely good.
- Write fewer pages, each answering one question properly and completely.
- Include something only you could write — your data, your customers, your mistakes.
- Use AI for structure, editing, and the mechanical parts. Not for the substance.
- If you would not send the page to a customer as an answer, do not publish it.
Will Google penalise AI-written content?
The stated position is that it rewards useful content regardless of how it was produced, and penalises thin content produced at scale to game rankings. In practice that means AI-assisted good content is fine and AI-generated filler is worse than nothing. The distinction is usefulness, not authorship.
Can it write in our founder's voice?
Reasonably well, given four or five samples of their actual writing — much better than any description of their style. It will not produce their genuinely original opinions, which is usually the part that made their writing worth reading.
How do we stop everything sounding the same?
Start with a claim rather than a topic. "Write about customer service" averages; "I think most customer service training is theatre, here is why" cannot be averaged, because nobody else supplied that premise.
Key takeaways
- Volume is free now, so volume is worthless. Having something specific to say is the scarce part.
- Write one voice brief, including a banned-words list, and paste it into every prompt.
- Repurposing one good piece into five formats is the best return in the function.
- Never let it supply a statistic. Mark gaps as [NEED FIGURE] instead.
- The test: could a competitor publish this with their logo on it?