We’ve reached the stage of the AI revolution where asking whether you should “use AI” feels a bit like asking whether you should “use the internet”.
Use it for what?
There’s an enormous difference between asking an AI tool to tidy up the notes from a meeting and allowing it to decide which employee gets promoted. There’s a difference between using it to generate 30 headline variations and trusting it when it confidently informs you that a court case, academic paper or historical event exists when it absolutely doesn’t. There’s quite a significant difference between asking it to resize your latest campaign and uploading your entire confidential client strategy because typing out the relevant paragraph seemed like a faff.
That distinction is increasingly important because AI adoption is racing considerably faster than our collective understanding of what constitutes safe use of AI.
Creativepool has already looked at why AI literacy is becoming a leadership skill rather than simply a software skill, and this is really where that literacy earns its keep. Knowing how to produce a clever prompt is useful. Knowing whether the task should have been handed to an AI system in the first place is considerably more useful.
The technology itself doesn’t come with a neat moral border around it. AI can be brilliant at helping somebody make a decision and terrible when asked to make the decision for them. It can save hours of tedious work and create hours of entirely new work when somebody has to untangle a confident mistake. It can help us think or allow us to stop thinking, depending largely on how much responsibility we surrender along the way.
So, rather than another grand declaration about whether AI is good or bad for creativity, perhaps we need something more practical.
Here are 10 things you should absolutely be letting AI help with, and 10 things you probably shouldn’t be surrendering to the machine just yet.
10 Things You Should Let AI Do
1. Summarise Things You’ve Already Got
If you want one of the least controversial gen AI use cases, start here.
Most creative businesses are drowning in information. Research documents, meeting transcripts, client decks, strategy papers, audience studies, workshop notes, feedback documents and email chains have an unfortunate tendency to reproduce when nobody is looking. Somewhere inside them is usually the thing you actually need, but extracting it can involve spending your afternoon rereading 78 pages because somebody buried the useful bit beneath three diagrams about “customer centricity”.
AI is exceptionally useful at compression.
Give it material you already trust and it can summarise themes, pull out actions, identify repeated concerns, compare documents or turn a rambling meeting transcript into something another human might conceivably read.
This is particularly valuable in research-heavy creative work, where AI can help turn large amounts of fragmented information into more usable strategic material without pretending the machine itself has discovered the insight.
The distinction matters. Summarising evidence and interpreting what that evidence means are different activities. Let AI help with the first. Keep humans firmly involved in the second.
2. Deal With the First Draft Nobody Wants to Start
The blank page has acquired an almost mystical reputation in creative culture, but sometimes a blank page is just irritating.
Not every first draft is an act of artistic expression. Sometimes you need an agenda. A rough project description. A structure for a presentation. An initial version of an internal document. Ten possible ways of organising the same piece of information.
These are precisely the kinds of tasks where AI earns its subscription fee.
A first draft gives you something to react against. You can dislike it, cut it apart, move sections around, rewrite half of it and wonder aloud why anybody would ever use the phrase “in today’s fast-paced digital landscape”. The important thing is that you’re no longer staring at an empty screen waiting for inspiration to arrive with a coffee.
That’s also where hybrid human-AI workflows make considerably more sense than fantasies of fully automated creativity. AI can supply material quickly. Humans can supply judgement, direction, taste and, ideally, the instinct to delete all the bits that sound like AI supplied them quickly.
The AI best practice here is straightforward: treat the first output as raw material rather than permission to stop.
3. Generate Variations Once the Idea Is Already Good
There’s an important difference between asking AI to invent your campaign and asking it to help extend one.
Once you have a strong idea, generative systems are exceptionally useful at variation. Different formats. Different lengths. Alternative layouts. Social crops. Subject lines. Calls to action. Adaptations for different audiences. Variations designed for testing. Rough localisation drafts.
This is exactly the sort of repeatable execution that AI is beginning to commoditise across the creative industries, and there’s little virtue in pretending a talented creative needs to spend half a Tuesday manually creating 38 near-identical versions of something purely because that’s how it was done in 2019.
The human value sits further upstream. Somebody needs to establish the idea, brand system, quality threshold and boundaries first. Once those exist, allowing AI to accelerate repetitive execution can give people more time for the work where they actually add something distinctive.
Using AI to multiply a good idea is useful.
Using AI to multiply an average one simply means you now have a much larger average idea.
4. Organise Messy Information
There’s an entire category of work that isn’t difficult so much as horribly untidy.
Hundreds of survey responses. Customer reviews. Workshop Post-its. Interview transcripts. Product feedback. Competitor claims. Search queries. Notes collected by six people who all apparently developed their own taxonomy halfway through the project.
AI can be extraordinarily useful for creating a first-pass structure around that mess. It can cluster similar comments, identify repeated language, categorise themes and expose patterns that would otherwise take hours to surface manually.
This is one of the more valuable ways creative businesses are already using AI for strategy and insight because it frees experienced people to spend less time sorting information and more time deciding what matters.
But “first-pass” is doing important work there.
A machine might tell you that 142 customer comments relate to price. A human strategist still needs to work out whether those people genuinely think the product is expensive, whether they simply don’t understand its value or whether the actual frustration lies somewhere entirely different.
Organisation is not insight.
It merely makes insight easier to find.
5. Be Your Slightly Annoying Devil’s Advocate
One of my favourite uses of AI is not asking it for an answer at all.
Ask it to disagree with you.
Give it your proposed strategy and ask where the assumptions are weakest. Give it your campaign idea and ask what a hostile audience might dislike. Ask what you’ve overlooked, which stakeholders might object or what evidence would undermine your conclusion.
This can be remarkably useful because creative teams, like all groups of humans, get attached to things. Once three people have spent a week building a proposition, their ability to identify reasons why that proposition might be rubbish becomes understandably impaired.
AI has no such emotional baggage.
That doesn’t mean its criticism is correct. It may produce objections that are irrelevant, naïve or simply wrong. But as a sparring partner it can force you to articulate why you believe something rather than allowing a room full of agreeable people to nod it through.
That is a much healthier application than treating AI as some kind of synthetic oracle. One of the emerging skills creative agencies increasingly need is the ability to use AI to challenge thinking rather than replace it.
Sometimes the best response an AI can give you is a question you hadn’t considered.
6. Handle Low-Risk Repetitive Admin
The AI revolution doesn’t all need to be synthetic cinema and existential debates about authorship.
Sometimes it can just rename things properly.
Creative businesses contain a breathtaking quantity of administrative work that contributes nothing to the brilliance of the finished product. Formatting meeting notes, generating action lists, organising schedules, drafting routine documentation, restructuring spreadsheets, categorising files and converting information from one predictable format into another are hardly sacred expressions of human imagination.
Automating some of this is an entirely sensible use of the technology.
Indeed, one large field experiment involving thousands of knowledge workers found that access to generative AI tools reduced time spent on email and helped people complete documents faster.
The key phrase, again, is low risk. If getting the task slightly wrong creates minor inconvenience and a human can quickly catch the problem, automation makes sense. If getting it wrong ruins a client relationship, creates a financial loss or exposes confidential information, you’ve moved into a very different category.
A useful rule emerges here: the easier the mistake is to reverse, the more comfortable you can probably be delegating the task.
7. Prototype Before You Spend Proper Money
Generative AI is extraordinarily good at creating things that are good enough to discuss.
That’s different from things that are good enough to publish.
A rough storyboard can help sell a film idea. A synthetic layout can test a visual direction. Temporary imagery can help a team understand a proposed experience. A crude prototype can expose whether a concept makes any sense before developers, directors, photographers, illustrators or production teams start invoicing.
That makes prototyping one of the best uses of AI in creative work.
It lowers the cost of asking “What if?” Instead of arguing about what somebody imagines a concept might look like, teams can create enough of it to have a more useful conversation.
The danger arrives when temporary becomes final because the temporary version looks suspiciously polished.
That problem is increasingly common. Generative systems create material with enough surface finish that people can mistake visual resolution for conceptual resolution. Something can look complete long before the thinking underneath it is.
AI should make experimentation cheaper.
It shouldn’t make standards cheaper.
8. Help You Translate and Localise a First Pass
Translation is one of those areas where AI capability can feel almost magical until you encounter a sentence where the literal meaning is entirely different from the useful meaning.
For straightforward internal communication or an initial working translation, AI can save considerable time. It can also help compare phrasing, highlight ambiguities and adapt the same basic information across multiple languages very quickly.
But localisation is more than replacing words.
Tone, humour, idiom, cultural associations, legal claims, product conventions and even what feels polite can shift dramatically between markets. That’s why professional human review remains crucial for anything important enough to represent a brand publicly.
The right workflow isn’t necessarily “human or AI”. It’s often AI first pass, human judgement afterwards.
This distinction crops up repeatedly in sensible AI best practices. Let automation handle the scalable middle where it’s competent. Put expertise back in where nuance, consequences and cultural meaning enter the picture.
9. Help You Search for Patterns, Not Declare Truth
AI can be useful during research, but only if we stop confusing research assistance with research evidence.
Ask a generative system to suggest avenues to investigate, explain unfamiliar terminology, propose search terms, identify possible connections or tell you which questions you should be asking next and it can be enormously helpful.
Ask it to provide a definitive set of obscure facts with citations you never intend to open and things become considerably more exciting for all the wrong reasons.
Even the organisations developing frontier models acknowledge that hallucinations, where a system confidently produces plausible but false information, remain a fundamental challenge. The sensible response isn’t to stop using AI for research. It’s to use it as a navigator rather than a destination.
This answers one of the more important questions around safe use of AI: when should you fact-check AI?
Whenever the truth of the output actually matters.
If a model says Paddington Bear enjoys marmalade sandwiches and you’re brainstorming a joke, the consequences of a mistake are relatively small. If it gives you a statistic for a client presentation, a quotation for publication or information that could affect somebody’s money, reputation or health, check the original source.
AI can help you find the path.
It shouldn’t become the evidence that the path exists.
10. Remove Drudgery So Humans Can Spend More Time Being Useful
This, ultimately, should be the point.
The most convincing case for creative AI was never “Look, it can also make an advert.” It was that technology might absorb enough repetitive work that talented people could spend more time on the things they’re actually talented at.
Research. Thinking. Talking to customers. Arguing about strategy. Developing ideas. Making difficult judgement calls. Learning new skills. Crafting the parts of work audiences genuinely notice.
That remains the optimistic version of the story.
As Creativepool has explored in looking at why human creativity becomes more valuable as average execution becomes easier to automate, abundance changes where value sits. AI can generate more. People still need to decide what deserves to exist.
The challenge is making sure the productivity benefit actually flows in that direction. There’s not much point saving your creative team five hours of repetitive production if management immediately rewards them by demanding another 400 pieces of repetitive production.
Efficiency should buy thinking time.
Otherwise we haven’t freed anybody from the machine. We’ve merely increased the speed of the conveyor belt.
The Problem Starts When We Stop Treating AI Like a Tool
Most of the risks of using AI don’t arise because somebody used it to transcribe a meeting.
They arise when we quietly promote it.
Assistant becomes expert. Expert becomes authority. Authority becomes decision-maker. Somewhere along the line, the human who was supposed to be checking everything becomes a ceremonial presence clicking “approve” because the output looks confident and the deadline is Thursday.
That is where safe use of AI stops being primarily a technical question and becomes an organisational one.
Current generative systems can be extraordinarily capable while remaining capable of confidently inventing facts. Formal risk frameworks explicitly warn that AI “confabulation” can become particularly dangerous when outputs influence consequential decisions.
So what should stay on the human side of the line?
Quite a lot, actually.
10 Things You DEFINITELY Shouldn’t Let AI Handle
1. Don’t Let AI Be Your Final Source of Truth
Can you trust AI-generated information?
Sometimes.
Which is precisely the problem.
An information source that’s always wrong is surprisingly easy to manage. You stop using it. An information source that’s impressively right most of the time but occasionally invents something with complete confidence requires considerably more attention.
AI hallucinations aren’t necessarily wild nonsense. That would be convenient. They can be tiny alterations to names, dates, quotations, studies, statistics or citations, presented with exactly the same linguistic confidence as something completely accurate.
That makes unverified AI particularly dangerous in professional content, journalism, pitches, strategy and research. A fabricated source placed inside a beautifully written paragraph doesn’t become less fabricated because the prose flows nicely.
One of the clearest AI best practices should therefore be embarrassingly traditional: check your sources.
Not by asking the same chatbot whether it’s sure.
Actually check them.
2. Don’t Let AI Make High-Stakes Human Judgements for You
AI can help structure a decision.
It shouldn’t necessarily own one.
Hiring is an obvious example. Automated systems can help process large application volumes, identify patterns or support administrative stages, but handing meaningful employment decisions entirely to an opaque system raises questions around bias, transparency, data rights and accountability. Current UK guidance specifically warns about unfair or discriminatory outcomes and emphasises safeguards and meaningful human review around automated recruitment decisions.
The principle extends far beyond recruitment.
Promotion. Redundancy. Discipline. Credit. Insurance. Education. Healthcare. Legal disputes. Any decision with significant consequences for an actual person deserves considerably more scrutiny than “the model gave them a 73”.
AI can surface information humans overlook. Humans can also be biased, inconsistent and terrible at decisions, obviously. The answer isn’t pretending people are perfect.
It’s recognising that outsourcing a judgement doesn’t outsource responsibility for it.
If somebody’s life changes because of your decision, there needs to be a person capable of explaining why that decision was made and taking responsibility for it.
“The algorithm said so” has never been an especially inspiring leadership philosophy.
3. Don’t Feed It Sensitive Information Just Because the Box Is Convenient
Is it safe to give personal information to AI?
The sensible answer is: not unless you understand exactly what system you’re using, what contractual and privacy protections apply and whether your organisation has approved the workflow.
That might sound obvious.
Then remember how many AI workflows begin with somebody copying an entire email chain into a prompt because they only wanted the machine to “tidy this up”.
Personal data, unreleased client plans, commercially sensitive information, private correspondence, employee records, customer information, passwords, legal documents and proprietary creative material should not simply be dropped into whichever consumer AI product happens to be open in another tab.
UK data-protection guidance treats AI processing of personal data as a genuine governance issue requiring attention to lawfulness, fairness, security and data minimisation. Cybersecurity guidance likewise warns that AI systems can create routes through which sensitive information is exposed, including through attacks designed to manipulate model behaviour.
This is one of the simplest safe use of AI rules imaginable: before uploading something, ask whether you would happily hand the same information to an unknown external supplier without checking the contract.
If the answer is no, perhaps don’t paste it into a chatbot because the interface has friendly rounded corners.
4. Don’t Let AI Decide Whether the Creative Work Is Actually Good
AI can assess work against criteria. It can identify potential problems, compare versions and act as a useful critic.
What it can’t do is relieve creative professionals of the obligation to develop taste.
That matters because AI is increasingly brilliant at producing work that is plausible. Clean. Competent. Familiar. The stuff that immediately looks like the sort of thing it was supposed to look like.
Which is not the same thing as being good.
The more AI commoditises competent creative execution, the more valuable the human ability to reject competent-but-forgettable work becomes. Somebody still has to know when the technically correct answer is culturally dead, when a campaign is polished but anonymous or when something uncomfortable is worth defending because the discomfort is precisely the point.
Taste isn’t mystical human fairy dust. It’s accumulated judgement shaped by culture, craft, experience, curiosity and exposure to great work.
If creatives surrender that part to the machine, there won’t be much creative profession left to defend.
5. Don’t Let It Reproduce Real People Without Thinking About Consent
We’ve made synthetic humans extraordinarily easy to create.
Our ethical thinking hasn’t accelerated quite as efficiently.
An AI system can replicate or approximate faces, voices, gestures and identities with astonishing fidelity. That can be useful in production. It can also become profoundly uncomfortable when a real individual’s likeness, voice or creative identity is being used without meaningful consent.
The fact that something is technically possible doesn’t resolve whether you have the right to do it.
This is particularly important in commercial creative work involving actors, models, performers, influencers, employees or members of the public. If you’re creating a synthetic version of somebody, they should understand what’s being produced, how it will be used, for how long and whether that synthetic version can subsequently appear somewhere they never agreed to.
As interest grows in “Made by Humans” and provenance as signals of creative trust, brands should be moving towards more clarity about synthetic participation, not looking for cleverer ways to conceal it.
Consent isn’t an inconvenient production hurdle.
It’s part of the work.
6. Don’t Treat It as Your Lawyer
AI is wonderful at sounding legal.
This is not the same as practising law.
Ask a model to explain a contractual clause in plain English and it may provide an extremely useful starting point. Ask it to identify questions you might want to raise with your lawyer and you’ve probably saved everybody some time.
Ask it to make the final call on whether your campaign infringes somebody’s copyright, whether a generated image is commercially safe to use or whether an employment practice complies with current legislation and you’re taking a rather different gamble.
Copyright and AI remain particularly complicated in the UK. The government’s 2026 review makes clear that the legal and economic questions around the use of copyrighted material in AI development remain active policy territory rather than a beautifully settled rulebook.
AI can help you understand the language of a legal problem.
It cannot assume the professional responsibility that comes with advising you what to do about it.
When the answer has legal consequences, use a lawyer.
Revolutionary, I know.
7. Don’t Let AI Replace Real Research With Synthetic Guesswork
There’s a worrying temptation emerging in creative strategy.
Why speak to actual customers when an AI can pretend to be one?
You can certainly ask a model to simulate different audience perspectives. It can help develop hypotheses, expose possible objections or create scenarios worth investigating. Those can all be useful exercises.
They are not primary research.
A simulated 23-year-old woman in Birmingham doesn’t actually live in Birmingham. She didn’t get the bus to work this morning. She doesn’t have an embarrassing overdraft, an argument happening in a WhatsApp group, a favourite takeaway, a mother she’s avoiding calling or an opinion about your product shaped by something her friend said six months ago.
She is a probabilistic construction derived from patterns in data.
That distinction becomes especially important when creative businesses use AI for strategy and insight development. Synthetic audiences can help pressure-test assumptions. They cannot replace the experience of listening to the people you’re supposedly trying to understand.
Real people are frustrating.
They contradict themselves. They misunderstand questions. They make irrational decisions and care deeply about details your segmentation model says shouldn’t matter.
That’s why they’re useful.
8. Don’t Publish Important AI-Generated Work Without Human Review
The promise of automated content systems is intoxicating.
Generate. Approve. Publish. Repeat.
Ideally remove the middle word.
That’s also how brands end up distributing an error at industrial scale.
The rise of synthetic content production at scale means organisations can create more copy, images, video, localisation and variations than any human team could realistically have produced only a few years ago. That creates enormous efficiency.
It also makes quality control considerably more important because mistakes now scale too.
An incorrect price across one page is a problem. An incorrect price automatically propagated across 300 pieces of content in 12 markets is a project.
Human review should be proportionate rather than ceremonial. Nobody needs the CMO personally inspecting whether an approved product image has been resized correctly. But claims, messaging, culturally sensitive material, regulated content, major campaign assets and anything capable of damaging trust should pass through somebody who knows what they’re looking at.
The cheaper creation becomes, the more dangerous it is to assume creation and publication should be the same process.
9. Don’t Let AI Make Ethical Decisions You Don’t Want to Own
This is related to judgement, but it goes deeper.
Businesses constantly make decisions where there isn’t a clean mathematical answer. Should we use this image? Is this targeting approach manipulative? Is this campaign punching down? Should a synthetic person be disclosed? Is it acceptable to automate this conversation? Are we exploiting somebody’s work, identity or vulnerability simply because the technology allows us to?
AI can discuss those questions endlessly.
It cannot own the answer.
It has no reputation, career, client, community or conscience on the line. It won’t wake up at 3am wondering whether something crossed a line. It doesn’t stand in front of the people affected and explain itself.
That’s one reason trust is becoming a more valuable marketing asset as synthetic content expands. Audiences aren’t merely judging whether output looks professional. They’re increasingly judging whether they believe the people and organisations behind it.
Ethics cannot become another prompt template.
The person benefiting from the decision also needs to be prepared to take responsibility for it.
10. Don’t Let AI Do All Your Thinking
This might ultimately be the biggest one.
Does AI reduce critical thinking?
Potentially, yes, particularly when confidence in the machine becomes greater than confidence in our own ability to question it.
Research involving 319 knowledge workers and 936 real-world examples found that higher confidence in generative AI was associated with less critical-thinking effort, while users with greater confidence in their own ability were more likely to engage critically with AI-assisted tasks.
That doesn’t mean using ChatGPT three times makes your brain fall out.
It means cognitive convenience has a cost if we stop noticing it.
Humans have always outsourced mental work. Calculators reduced the need for manual arithmetic. Search engines reduced the need to remember where every piece of information lives. GPS ensured an entire generation now requires satellite assistance to visit an aunt who has occupied the same house since 1998.
That isn’t necessarily bad. Tools free cognitive capacity for other things.
The danger is outsourcing the wrong layer.
Let AI recall, rearrange, summarise, calculate, compare and generate. But keep asking whether the problem has been framed correctly. Keep interrogating assumptions. Keep developing your own opinion before requesting one from a machine. Keep reading full sources occasionally rather than living entirely inside summaries of summaries.
Creativepool’s argument that human judgement becomes more valuable as AI makes production more abundant applies just as much to thought as it does to design.
If AI does the boring thinking, brilliant.
If it does all the thinking, we’ve misunderstood the assignment.
Why the Difference Matters
The most useful way to decide what not to use AI for isn’t to maintain a static list of forbidden activities.
Technology will move too quickly for that.
Today’s dangerous use case may become tomorrow’s boring utility once reliability, governance and safeguards improve. Tasks that currently require extensive human intervention will become increasingly automated. New risks will appear in areas nobody has thought particularly hard about yet.
What we need instead is a way of judging the task.
One useful test is consequence. If the AI gets this wrong, what happens? Do you spend three minutes correcting a paragraph, or does somebody lose their job? Does the designer reopen Photoshop, or does confidential client information leave the organisation?
Another is reversibility. Can the output be easily reviewed and changed before it causes harm? AI-generated workshop notes are highly reversible. A fully automated hiring rejection isn’t quite so forgiving.
Then there’s accountability. Who owns the outcome? If the only available explanation when something goes wrong is “AI did it”, your process probably has a rather large human-shaped hole in it.
Finally, ask whether the task depends primarily on pattern recognition or human judgement. AI thrives on the former. The latter is precisely where creative agencies are being forced to develop stronger judgement, governance and AI literacyrather than simply adding more tools to the stack.
Those four questions are a far better guide to safe use of AI than simply dividing work into “things humans do” and “things machines do”.
Because that boundary will keep moving.
Responsibility shouldn’t.
So, What Should You Actually Give AI?
Give AI work.
Don’t automatically give it authority.
That’s probably the cleanest version of the whole argument.
Let it chew through the transcript. Let it summarise the enormous report. Let it produce 40 variations of an already approved asset, tidy your notes, suggest alternative questions, create the ugly first prototype and tell you why your strategy might be wrong.
Those are useful jobs. In many cases they’re among the best uses of AI because they exploit precisely what the technology is good at: speed, scale, pattern recognition, transformation and generation.
But maintain a firmer grip when the task involves truth, people, privacy, rights, accountability, taste or significant consequences.
That isn’t anti-AI. Quite the opposite. It’s how AI becomes genuinely useful rather than merely ubiquitous.
The creative industry has already moved beyond the question of whether AI belongs inside our workflows. It does. The more important question now is what role we give it once it gets there.
An intern? Sometimes.
A researcher? With supervision.
A production assistant? Absolutely.
A provocateur? Surprisingly good.
A lawyer, HR director, creative director, therapist, compliance officer and omniscient source of objective truth rolled into one?
Perhaps not.
The risks of using AI tend to increase at precisely the point where convenience persuades us to stop checking, stop questioning and stop taking responsibility. Conversely, the smartest AI best practices mostly involve keeping people engaged at the points where being human actually matters.
That’s why the future probably isn’t about deciding what AI can do.
The list of things it can do will keep expanding faster than any article can document them.
The useful question is what we want it to do.
And, occasionally, having enough confidence to tell the machine that this particular one is still ours.