A note before the split. This is general guidance, tools change constantly, and your organisation's policies on data and sign-off govern. The tension the piece is built on, volume against voice, is the durable part, and it applies whichever tool your team drafts in.
Marketing managers face the AI question in its most concentrated form, because the job is words at industrial volume: the campaign copy, the emails, the social calendar, the landing pages, the case studies. The tools produce words at industrial volume too, which looks like a perfect match right up until you notice what kind of words.
The internet is currently drowning in competent, interchangeable, faintly familiar marketing content, all of it produced the same way, and audiences have responded by developing an ear for it and scrolling past. So the marketer's version of the trust question is not really about accuracy.
It is about a trade nobody prices out loud: every hour the tool saves you is purchased with a small withdrawal from your brand's distinctiveness, unless you deliberately buy the distinctiveness back. The good news is that buying it back is cheap once you know where the till is.
The short answer: hand ChatGPT, or the AI of your choice, the volume. That means variant generation, format repurposing, first drafts from proper briefs, the metadata drudge, and campaign synthesis. Protect the voice by feeding the tool your best existing copy as the model, rather than letting it default to the middle, and by giving everything customer-facing a human pass that restores the phrases only your brand would use. Keep three things out entirely: performance numbers it invents, competitor claims you have not verified, and customer data on consumer versions. Do that, and the evenings come back without the brand dissolving into the beige.
Where the hours are
| The job | The hours it returns |
|---|---|
| Variants at volume | Ten subject lines, five hooks, three lengths of the same message, generated for your judgement rather than your composition, which is the single biggest weekly saving in the function |
| Repurposing | One webinar or case study reshaped into the post, the email, the one-pager and the social cuts, the multiplication that used to eat Thursdays |
| First drafts from real briefs | Campaign copy started from a brief with audience, goal and constraints in it, arriving at reviewable rather than blank |
| The metadata drudge | Descriptions, alt text, tags and the other invisible copy nobody should hand-write in bulk |
| Campaign synthesis | The post-mortem structured from your notes and numbers, with the numbers themselves supplied by you, for reasons the next section makes plain |
The brief is the hinge in that list. A one-line request gets you a one-line answer's worth of thinking, which is why the setup that stops generic answers matters more here than in almost any other function.
The voice problem, which is the actual problem
Left to its defaults, ChatGPT writes the statistical centre of all marketing copy ever written, which is precisely the register your audience has learned to skim past. A brand that ships enough of it becomes indistinguishable on purpose.
The fix has two halves and takes minutes once habitual:
- Upstream, never let the tool start from nothing. Give it three examples of your best-performing copy and tell it to match their register, because the tool holds a voice far better than it originates one.
- Downstream, give everything customer-facing the pass that restores what only your brand would say. The specific claim, the house phrase, the joke the middle of the internet would never make.
The draft is allowed to be generic. The published version is not, and the distance between them is your actual job now. The same principle applies to your outbound, and the piece on AI emails that do not read as AI covers the tells worth hunting for before anything sends.
What stays out
Three things, each for a different reason:
- Performance numbers. The tool predicts plausible figures rather than reading your dashboard, and an invented conversion rate in a report is a small career event.
- Competitor claims. It will confidently describe what rivals offer and charge from stale or imagined knowledge, and a wrong public claim about a competitor is a legal letter waiting to happen. Everything comparative gets verified against their actual current materials.
- Customer data. Lists, behaviour, anything identifiable, because the everyday consumer versions can learn from what enters them, and audience trust is the one asset marketing cannot buy back on sale.
The line that holds
The function's uncomfortable gift from AI is clarity about what marketing judgement actually is, because the tool has commoditised everything that was not it. Generating copy is now free, which means choosing the right copy, knowing the audience well enough to brief precisely, and protecting the voice that makes your brand recognisable are the whole remaining job.
Conveniently, they were always the important part. Run the volume through the machine, run the judgement through yourself, and the output goes up while the beige goes down, which is the only version of AI-powered marketing your audience will thank you for.
If your commercial colleagues are working through the same split, the version of this question for consultants makes a useful companion read, since their trust line sits in a different place to yours.
Not sure where to start?
Volume from the tool. Voice from you.
The free finder tells you which AI to open for the task in front of you, in about 60 seconds, and the full "AI, sorted." reference lands in your inbox.
Try the free finder →One question in, one answer out.
Clair helps non-technical professionals know when to trust their AI, when to check it, and when to skip it.