Most marketing teams I speak to have the same complaint about AI. The output is fine. It's grammatical, it's on topic and it arrives in ten seconds. It also reads like every other brand in the category, and customers notice that faster than managers do.
Banning the tools won't fix it. Changing what goes in, what you use it for and who signs off on what comes out will. This is how I'd set up AI in a marketing team so it saves real time without sanding the personality off everything you publish.
Why AI output defaults to generic
A language model predicts the most likely next word. Give it a thin brief and it hands back the most likely version of the thing you asked for, which is roughly the average of everything ever written on that topic. That average is exactly what your brand is trying not to sound like.
Anyone who has briefed an agency will recognise the pattern. At Fields Luxury I worked through two content agencies and three ad agencies before landing the right ones. Writers without enough context produce the safe, middle-of-the-road version of your brand. A model with no context does the same thing, just faster and in greater volume.
So the problem is mostly upstream. Generic in, generic out. The fix is to treat the model like a capable new freelancer who knows nothing about your business yet.
Feed it context, not a blank box
The single most useful thing a team can build is a brand context pack. One document, kept up to date, that goes into every piece of AI-assisted work. Mine usually covers:
- The customer in their own words. Lines lifted from reviews, support tickets, sales call notes and social comments. Real phrasing beats a persona slide every time.
- Voice rules with examples. Not "friendly but professional". Show a sentence that sounds like you and one that doesn't, side by side.
- A banned list. Every category has its clichés. Write yours down so the model stops reaching for them.
- Your best work. Five to ten pieces the team is proud of: emails, product pages, ads, posts.
- Product facts. Materials, sizing, pricing logic, delivery promises. Specifics are what make copy sound like it came from someone who knows the product.
The rule of thumb is simple. If a new freelancer would need it to write well for you, the model needs it too.
Store it somewhere central, load it into whatever project or custom instructions feature your tools offer, and give one person ownership. A context pack nobody maintains drifts out of date within a quarter.
Use AI for the thinking before the writing
The biggest gains I see from AI in marketing teams don't come from drafting copy at all. They come from the work around the copy: research, summarising, reporting, spotting what changed and why. None of that has a voice problem, because nobody outside the team ever reads it.
I build my own AI tools and agents for marketing work, and two of the most useful are deliberately unglamorous. One is a weekly account pulse. It looks at how the ad accounts performed over the week and writes a short summary of what moved and what's worth a closer look. The other is a budget review that checks spend against plan and flags where money might be better placed.
Neither writes a single customer-facing word. The point is that the team starts the week with questions to answer instead of spreadsheets to build. A human still decides what to do. They just get to the decision quicker.
If your team is only using AI to write social captions, you're starting at the riskiest end. Start where a slightly bland output costs nothing.
Build small tools instead of a prompt library
Most teams try a shared prompt library first. It usually dies quietly. People copy a prompt, tweak it, forget to save the better version, and within a few months there are forty prompts and nobody trusts any of them.
Small, specific tools last longer. A tool here can be as simple as a saved assistant with fixed instructions, the context pack baked in and a fixed output format. The difference is that the thinking is done once, by someone who knows the task well, and everyone else gets the benefit.
This is how I'd build the first one:
- Pick a task someone on the team does every week, the same way, and finds tedious.
- Ask the person who does it best to write down exactly how they do it, including the judgement calls.
- Turn that into instructions. Add the context pack and two or three examples of a good finished output.
- Fix the output format so results are easy to compare week to week.
- Run it alongside the human version for three or four weeks. Where they differ, work out whether the tool or the person was right, and update the instructions.
Recurring, structured work suits this far better than one-off creative work. Reporting, briefs, competitor round-ups, product description first drafts and email QA checks are all good candidates.
Keep a named human with the final say
The rule I'd put on the wall: AI can draft, but a named person owns every piece that goes out. Not "the team". One person, whose name is on it.
That person needs a short edit checklist, because reading AI copy for long enough makes everything look acceptable. Mine is roughly this:
- Could a competitor publish this unchanged? If yes, it isn't finished.
- Is there at least one thing only we could say? A product detail, a customer phrase, a number, an opinion.
- Does the first paragraph earn its place? Models tend to warm up slowly, so the opening often goes.
- Does it sound right read aloud?
There's a second rule of thumb that saves a lot of frustration. If editing a draft takes longer than writing it from scratch would have, don't keep fixing drafts. Fix the inputs. Nine times out of ten the context pack is thin or the brief was vague.
Rolling it out across the team
Tools are the easy part. Habits are harder, and they're set by whoever leads the team.
Don't measure AI adoption by volume. A team producing three times as much content that all sounds the same has made things worse, not better. Measure time saved on recurring tasks and whether the quality of the work held up.
Share the failures as openly as the wins. When someone gets a bad output, look at what went in. That's how the context pack and the tools improve, and it stops people quietly deciding AI "doesn't work for us".
And be honest about what stays human. Strategy, positioning, the big creative idea, the call on where budget goes. AI can inform all of those. It shouldn't make any of them.
If you want something to do this week, keep it small. Start the context pack with your best ten pieces of work and twenty lines of real customer language. Pick one recurring task and build a single tool for it. Name an owner for every AI-assisted piece that goes out. That's enough to see whether the approach suits your team before you invest more time in it.
FAQ
Will using AI in a marketing team hurt our SEO?
Not by itself. Search engines care about whether content is useful and original, not which tool helped write it. The risk is publishing large volumes of thin, interchangeable pages, which is exactly what unedited AI output tends to produce. Add real expertise, specifics and a clear point of view, and the tool used matters far less.
Which tasks should a marketing team hand to AI first?
Start with internal, recurring work where bland output costs nothing: performance summaries, budget checks, research round-ups and first drafts of briefs. These save real time and let the team learn how the tools behave. Move to customer-facing copy once your context pack and review process are in place.
How do you keep brand voice consistent when everyone uses AI differently?
Give everyone the same inputs. A shared context pack, a banned list and a handful of approved examples do more for consistency than any style guide paragraph. Then make sure a named person signs off on anything that goes out, using the same short checklist every time.
If you're working out how to bring AI into your own marketing team, I'm happy to talk it through. I'm open to Head of Marketing and Head of Growth roles, particularly with DTC and consumer brands, and I take on consulting work alongside that. You can get in touch via the contact link on tomosborne.me.