AI vs Humans in 2025: Who Should Write When?

TL;DR: The Critical 10 is the ~10% of work that sets direction, carries risk or defines brand voice. Keep it human-led.  The other ~90% (research synthesis, outlines, some first drafts, variants and routine edits) can be AI-led so long as you have clear guardrails and keep human sign-off. Use the 60-second checklist at the end to decide who leads each piece.

The Wrong Debate: “AI vs Humans” Isn’t the Real Question

In 2025, the useful question is no longer AI vs humans generally, it’s which should lead each specific content job.

The answer depends on the what you’re writing and what’s at stake. That’s what the Critical 10 is about: it’s the 10% of writing and content that should always be human‑led.

The Critical 10 Framework: A Smarter Way to Think About Content

Let’s be brutally honest. Most of the in most organisations – as much as 90% of it – doesn’t matter that much.

Things like emails, internal reports and a lot of internal slide decks aren’t usually setting strategy, influencing reputation or defining identity. So they don’t need to be Pulitzer-winning quality. They just need to be clear, consistent and done on time.

For this type of writing work, AI is a game changer. It saves time, reduces effort and often does a much better job than humans, especially when the original author isn’t a strong writer.

But then there’s the other 10%*…

This is the content that sets direction, carries risk or defines brand voice. It’s your positioning and your story, the stuff that shapes how people see you and even if they see you at all.

Here, writing isn’t just functional, it’s strategic.

In this type of content, nuance, judgement and tone aren’t just nice-to-haves, they’re the whole point.

No matter how good your prompts are or how effectively you can use a Large Language Model (LLM) like ChatGPT, you shouldn’t be trusting it to lead this kind of content work. You (or a professional writer) will do it better, safer and with more authority.

Because this type of writing work isn’t about volume, it’s about shaping you others think and feel.

*Note: In some industries – think legal, health and financial services – more than 10% of writing work is likely to be critical. However, I’ve found 10% is an accurate guide for most sectors and organisations.

What AI can write: The 90%

So let’s look at the first category: the 90% of content you should be outsourcing to AI.

This is the writing work that’s repeatable, scalable and driven by pattern rather than persuasion. It often includes:

• • • Internal comms: summaries, email updates & status reports

  
  

  • • Basic web content: pages like how‑tos & FAQs

  
  

  • • Marketing support: some socials, minor product descriptions & basic customer emails

  
  

  • • Operational docs: onboarding guides, process docs & internal knowledge content

  
  

  • Sales material: first-draft proposal templates & slide copy (but not the persuasive stuff).

When scoped and prompted properly, AI can write these formats way faster than a human, often with better structure and tighter consistency. And it can go all day and night. (When I worked as an advertising copywriter, some agency owners thought we could too. But that’s another story…)

AI is excellent at this type of writing because it’s working within known patterns. It’s filling gaps and applying tone templates. And it can generate a draft (actually many drafts) in no time at all.

But here’s the thing: it only works when you use it right.

How to Write Your 90% With AI (Properly)

Just because AI can write the 90% doesn’t mean it’ll do it well.  At least, not by default.

You only have to read much of the content being generated in 2025 to know how true this statement is.

But most of the bad AI content I see isn’t the model’s fault. It’s the product of vague prompts, lazy workflows and people asking AI to think for them rather than with them.

The result is scaled mediocrity: samey prose, shaky facts and a voice that sounds like everyone else’s.

What AI Can’t Write (the Critical 10)

If a piece of content sets direction, carries risk or defines how your brand is perceived, you shouldn’t be handing it over entirely to a machine. That’s a Critical 10 piece.

Critical 10 pieces usually fall into one of five non-delegable categories.

1. Strategy & Narrative Writing

Examples: Category positioning, About pages, anything that involves POV (e.g. thought leadership).

Why: This is judgment work. It’s full of nuance, trade-offs and lived experience. AI doesn’t really know your market, your mind, your competitors or your “why”. Despite what it might tell you, it’s just completing patterns.

2. High-Risk or Regulated Writing

Examples:  A lot of financial, legal and health-adjacent writing still sits here (but not all of it). Crisis communications definitely fall here too. So shareholder comms and market announcements.

Why: This type of writing carries consequences. Get something wrong and AI won’t be sued. You will.

3. Brand-Defining Voice

Examples: Hero copy, branding, positioning, naming and taglines.

Why: This is your brand. You don’t outsource your identity to autocomplete.

4. Revenue-Critical Offers

Examples: Major proposals, pricing pages and high stakes tenders.

Why: This work has direct commercial impact. So nuance, intuition and empathy matter.

5. Final Sign-Offs

Examples: The last set of eyes on any public-facing work.

Why: Tone, accuracy, ethics, risk and accountability all need human judgment. It’s not AI’s skin in the game, it’s yours.

Work type	Lead	Why
Positioning, narrative, POV	Human	Strategy, judgement, brand risk
Shareholder letters, crisis/regulated comms	Human	Liability, reputation
Home hero, taglines, naming	Human	Identity; long‑tail consequences
Research synthesis & summaries	AI	Speed; human sets sources
Outlines, first drafts, variants	AI	Patterned; volume & iteration
Line edits, style consistency	AI	Consistency but human final skim
Thought leadership	Hybrid	Human thesis & ideas; AI for structure

How to Use AI When Writing Critical 10 Content

AI shouldn’t own the Critical 10, but it can still play a role.

Used properly, AI is a sharp assistant that can help you think faster. It can also draft alternatives, spot gaps or pressure-test your message. It just shouldn’t be making the call.

Here’s how to use it, even for your most important writing work.

1. Surface blind spots

AI’s great at spotting what you might be missing. Use it to:

• • • List possible counterarguments

  
  

  • • Identify logical gaps

  
  

  • Highlight unanswered questions

Prompt: “What would a smart critic say about this argument? Where are the weak points? What’s the loose thread a doubter will pull on?”

2. Draft variations for structure or emphasis

Got a POV but want it framed three ways for different audiences? Use AI to explore tone, structure and rhythm – but only after a human sets the core message.

Prompt: “Rewrite this for [audience]. Keep all original facts, just shift the emphasis.”

3. Stress-test for risk

Don’t rely on AI to fact-check. Do use it to spot where a reader might get confused, offended or misled.

Prompt: “List any potential risks – legal, reputational, ethical – in this draft. Return as {risk, severity, mitigation}.”

4. Apply style consistency (after the human writes)

Once the message is final, you can use AI to clean up your writing by calling out things like:

• • • Tone inconsistencies

  
  

  • • Passive voice

  
  

  • • Overused phrases

  
  

  • Sentence length variation

Think of this as “AI-as-copyeditor”. That means not using it as a writer but as a second set of eyes on your work.

5. Organise thinking before the draft

AI is brilliant at mapping arguments and helping you turn your scattered ideas into a coherent plan. Feed AI messy notes and it can quickly turn them into a structured outline.

Prompt: “Turn this into an outline with clear sections. Put the strongest point up front. Flag where data or examples are needed. Also flag any weak points which need to be omitted.”

So… What Now?

If you’re still debating “AI vs humans,” you’re behind.

Smart teams aren’t choosing sides, they’re choosing use cases. That means letting humans handle the work that really matters and handing the rest to the machines, but with serious guardrails in place.

Here’s how we can help you do the same.

• • • AI Writing Workshops. Learn the prompts, strategies and workflows that empower your team to write more effectively & efficiently.

  
  

  • • Done-in-a-Day. If your pitch is fuzzy or your brand voice is off, we’ll fix it. In 24 hours. Guaranteed.

  
  

  • Content Lab. Need the kind of strategic writing only a human should be trusted with? We give you the very best.

AI vs Humans: Who Should Write This? (60-Second Scorecard)

Run these five checks and match your score below. Tick yes/no: ☐ Stakes & risk: Could a wrong line cost money or trust? ☐ Brand-defining: Is this a first-impression or identity-defining asset? ☐ Strategy-setting: Does it shift positioning, narrative or POV? ☐ Legal/regulatory: Is it governed by regulations? Does it involve high risk or crisis protocol? ☐ Judgement call: Does it need lived context, trade-offs or data you can’t paste in? Score → Decision: 2–5 yes: Human-led. AI assists. You lead. 1 yes: Hybrid. You shape, AI drafts, you sign. 0 yes: AI-led (with guardrails). Human skims.

 

FAQ

What should AI write and what should stay human?

What belongs to AI vs humans again?

Use the 60-second checklist above. If it involves stakes/brand/strategy/regulated/final sign-off it remains human-led (Critical 10). Patterned, repeatable work is AI-led with guardrails.

Can AI write thought leadership?

Sometimes. Best to treat it as Hybrid: a human sets the thesis and argument. AI helps with structure, examples, stress-testing and alternatives. Humans always own the ideas and opinions.

What about regulated industries?

Keep claims and conclusions human-led with legal/compliance review. Use AI for synthesis, options and formatting—never as final authority.

Where does AI help most in sales?

Use it for structure and variants for outreach, follow-ups and deck formatting. Pricing pages, negotiation emails and final proposals belong to the Critical 10 and should be human-led.

How do we keep quality and voice?

How do we keep brand voice?

Generate a Voice Pack (3 great examples, 1 “do-not”, tone rules, taboo list). Use AI to check consistency post-draft. Humans should always own the final pass on Critical 10 content.

Will AI make us sound generic?

Only if you let it. Provide a POV, specifics and a taboo list. Insist on concrete examples. Human-edit the headlines.

What’s the quickest workable process?

Brief → Sources → Outline approval → Constrained draft → Claims + Risk list → Human sign-off. Most teams can run this process in as little as 15 minutes to protect quality and voice.

How do we maintain trust and accuracy?

Do we need to disclose when AI helps?

Disclose when AI’s role is substantial, especially in regulated contexts. A simple line works: “Drafted with AI assistance; facts and voice reviewed by [role].” We asked AI to help with these FAQs (but didn’t let it run wild).

How do we stop hallucinations?

Use approved sources only, give it an “ask-don’t-invent” rule and a claims list with citations. Always conduct a human fact/risk review before publishing. This is the guardrail set we teach in our Writing Workshops.

Can AI produce citations we can trust?

AI can format citations but it shouldn’t originate them. You supply sources; AI maps claims → citations; a human verifies every reference/URL.

Do AI detectors work?

They’re unreliable. Don’t rely on them for gatekeeping. Judge work by truth, usefulness, originality and brand voice.

How do we avoid plagiarism or mimicry?

Don’t prompt “write like X.” Use your Voice Pack, cite data properly, and run a light similarity check when stakes are high.

How do we run the 90/10 split at scale?

Which model/features should we use?

Choose a current model with browsing, advanced data analysis and document handling. We use ChatGPT-4o/-o3 and Claude, and can train you on major LLMs including Google Gemini, Copilot and Llama. If you’re unsure, our AI Writing Workshops cover setup and guardrails.

Should we store prompts?

Yes. Build a versioned Prompt Library by role/use-case so you can reuse and improve them. Pair each prompt with examples and expected outputs. Review quarterly.

How often should we retrain teams?

At least quarterly: debrief on new features, update prompts, add fresh examples and tune governance based on incidents and feedback.

How do we operationalise governance?

Document a sources policy, claims checklist, risk thresholds, approval roles and an audit trail. Make the final human sign-off explicit.

How does AI affect SEO and discoverability?

Will AI-written content hurt our SEO?

No. Bad content hurts SEO, not the tool. AI can be fine when you brief well, cite real sources, structure for audience/intent and keep a human sign-off (especially for Critical 10 work).

Does Google penalise AI content?

Google penalises unhelpful, unoriginal, low-value pages -not AI. Produce accurate, people-first content with unique value, then keep it fresh.

How do we get picked up by AI Overviews/SGE?

Answer intent fast (150–180-char summaries), cover key entities, add concise FAQs, cite credible sources and link from a strong pillar. Build authority and update regularly.

Are there legal or data risks with AI writing?

Is AI usage safe for confidential data?

Use enterprise plans that disable training on inputs, apply redaction/DLP and store sources internally. Stay in line wth Australian privacy expectations. Don’t paste secrets into consumer tools.

Who owns the copyright of AI-assisted work?

Check your model’s terms. Define ownership in contracts, keep an audit trail and seek legal advice for high-value IP.

Updated July 2025