Because a language model writes the most likely words for the prompt it's given, and a short prompt with no specifics gets the average of every similar piece of marketing it has seen. Fix it by feeding it what only your business has (real details, numbers, customer phrases, proof), showing it examples of your own writing, banning stock phrases and editing each draft for specifics.
Asking the AI to "make it less generic" or "more human" rarely works, because the problem is what went in, not how it's worded. The model can't know that your instructor used to be terrified of driving herself, that most of your members come straight from work, or that couples always ask you about seating plans. Without those details it fills the gaps with the phrases that appear most often in copy like yours, which is exactly why it sounds like everyone else's.
The six fingerprints of generic AI copy
Readers spot these without being able to name them. Once you can name them, you can prompt against them and edit them out.
- The stock opener. "Are you looking for…?", "In today's busy world…", "Ready to take the next step?" A question nobody asked, or a statement nobody needed.
- Abstract adjectives. "Professional, friendly and reliable." Every business claims these, so they carry no information.
- The rule of three. Lists of three benefits, three adjectives, three short sentences in a row. Once is fine; every paragraph is a tell.
- Balanced, symmetrical sentences. "Whether you're a beginner or an experienced driver, we've got you covered." Smooth, and says nothing.
- Claims without proof. "Our proven approach gets results." Proven how? Which results?
- The upbeat sign-off. "Book today and start your journey!" with an emoji or two. Every piece ends the same way.
A quick exercise: take your last three AI-drafted posts and underline every instance. Many owners find most sentences contain at least one.
Why models drift to the middle
A language model is trained on a vast amount of text, including a great deal of marketing copy, and learns which words tend to follow which. When you ask for "an ad for a driving school", it produces the kind of text that most often follows a request like that. The most common driving school ad copy is, by definition, the most generic. The model is doing precisely what it was built to do.
Two more things push it towards the middle. Assistants are tuned to be helpful, positive and inoffensive, so they avoid strong opinions, odd details and anything that might put a reader off, which are often the things that make copy memorable. And they know nothing about your business beyond what you tell them. Where the prompt is silent, the model fills in the typical answer, sometimes as a plausible-sounding invention ("over 20 years' experience") rather than a gap.
This also explains why vague style instructions backfire. "Make it engaging" or "make it punchy" nudge the model towards what engaging marketing usually looks like: more exclamation marks, more rhetorical questions, more "reach your full potential". You get more of the fingerprints, not fewer. Turning up a tool's "creativity" setting, where one exists, mainly makes word choice less predictable; it doesn't supply the facts that were missing.
The same driving school ad, before and after
An illustrative independent driving school, run by two instructors, wants a short Facebook ad for learners who are nervous about starting. The first prompt:
Write a Facebook ad for my driving school targeting nervous learners.What came back (illustrative):
Nervous about learning to drive? You're not alone! At our driving school,
our friendly and professional instructors are here to help you build
confidence behind the wheel. Whether you're a complete beginner or
need a refresher, we'll guide you every step of the way.
Book your first lesson today and start your journey to freedom! 🚗✨Every fingerprint is there: stock opener, abstract adjectives, symmetrical "whether you're…" sentence, a claim without proof, the upbeat sign-off. Now the second prompt, with specifics:
Write a Facebook ad (under 70 words) for nervous learners.
Specifics to use:
- First lessons are on quiet roads and an empty car park; no busy
roads until the learner says they're ready.
- Lessons can be 60 or 90 minutes; nervous learners usually prefer 90
because the first 20 minutes are for settling in.
- Real review phrase: "she never once made me feel stupid".
- One of our two instructors was a nervous learner herself and failed
her first test.
- Offer: first 90-minute lesson at the standard 60-minute price.
Rules: no question as the opener; no adjectives like friendly,
professional, reliable; no emojis; end with the offer, not a slogan.The result (illustrative):
Your first lesson starts in an empty car park. No busy roads until you
say so. One of our instructors failed her own first test through nerves,
so she knows what the first 20 minutes feel like. As one learner put it:
"she never once made me feel stupid." First 90-minute lesson at the
60-minute price, so you've got time to settle in.Same model, same length. The difference is that the second version contains five facts no other school can copy. That is the whole fix in miniature: specifics in, specifics out.
Build a specifics bank the AI can't invent
You won't write a detailed prompt from scratch every time, so keep the raw material in one document and paste the relevant parts in (or save it as the instructions of a shared project in your AI tool). Here's the driving school's bank, filled in:
| Category | Entries |
|---|---|
| Facts about the service | 60- and 90-minute lessons; manual and automatic cars; pick-up from home, school or work; mock test before every real test |
| Numbers you can stand behind | Two instructors; lessons 7am to 8pm; learners typically book 1-2 lessons a week |
| Customer phrases (from reviews) | "never made me feel stupid", "explained it until it clicked", "calm", "patient with my parallel parking" |
| Stories (with permission) | Instructor who failed her first test through nerves; learner who went from panicking at junctions to passing in four months |
| Objections and honest answers | "Too old to start?" No; plenty of learners start in their thirties and forties. "Can I go at my own pace?" Yes; no busy roads until you're ready |
| Proof | Number of Google reviews and the rating (update monthly); quotes above |
| Never say | Pass-rate claims, "guaranteed pass", "cheapest", "best in the area" |
The "customer phrases" row is the most valuable and the easiest to fill: paste your reviews into an assistant and ask for the phrases customers repeat, with counts. Turning reviews into marketing copy covers that step. The "never say" row matters just as much, because it stops the model filling a gap with a claim you can't back up.
Let the AI interview you before it writes
The quickest way to fill a thin prompt is to have the assistant ask for what's missing. Instead of "write a post about our flash day", try: "Before writing anything, ask me five questions whose answers would make this post specific to my studio. Ask them one at a time." It turns a two-line prompt into a two-minute conversation, and the answers are exactly the specifics a draft needs.
An illustrative exchange for a tattoo studio's flash-day post. The assistant asked: what's on the flash sheet this time; what's the price range; is there anything different from last time; what do first-timers usually worry about; and what should people bring. The owner's answers, typed quickly: "botanical and small script pieces, $80 to $150, first time we've done a colour sheet, people worry it'll hurt and that they'll be judged for picking something small, bring ID and eat beforehand."
The resulting draft opened with "Our first colour flash sheet: botanicals and small script, $80 to $150" and ended with "Bring ID, eat something first, and nobody here judges a small first tattoo. Most of ours were small too." That last line came straight from the owner's answer about first-timers' worries, and it's the line people commented on. The model didn't get more creative; it got better material.
Save the instruction as a reusable prompt. Once it's a habit, the interview takes less time than editing a generic draft into shape, and your answers can go into the specifics bank so the next post needs fewer questions.
Teach your voice with examples, not adjectives
"Write in a friendly but professional tone" describes almost every business on earth, so the model does what it would have done anyway. Examples work far better. Give three to five short pieces you wrote and liked, of the same type you're asking for, plus a short list of what you do and don't say:
| We say | We never say |
|---|---|
| "lessons", "learners", "your test" | "journey", "freedom awaits", "hit the road" |
| Short sentences; the occasional dry joke | Exclamation marks in every line |
| "You'll", "we'll" (contractions) | "We are pleased to offer" |
| Specific times and places | "Flexible scheduling to suit your needs" |
The pairing of positive and negative examples is what does the work. If you want this written up as a proper document the whole team can use, building a brand voice guide that AI can follow walks through it.
A banned-phrase list and a five-minute edit pass
Some phrases turn up so often in AI drafts that it's worth banning them in every prompt. A starter list for small business marketing:
- "Are you looking for…", "Ready to…?", "Your search ends here"
- "Take it to the next level", "reach your full potential", "to new heights"
- "Journey", "passion", "tailored to your needs", "we've got you covered"
- "Whether you're a… or a…"
- "Friendly and professional", "high-quality", "second to none"
- "Don't miss out!", "Book today and start your journey!"
Add the ones you notice in your own drafts; a banned-phrase list for your brand has a longer list grouped by type. Bans help, but a model told not to write "journey" sometimes reaches for "adventure", so the list doesn't replace editing. Five minutes per piece:
- Delete the first sentence. It's often throat-clearing; the second sentence is usually the real start.
- Circle every adjective. Replace each with a fact or cut it.
- Find one claim and add its proof, or remove the claim.
- Check the ending: does it tell the reader exactly what to do next?
- Read it aloud. Anything you wouldn't say to a customer's face goes.
Fixes by format, with before and after
An Instagram caption for a yoga studio
Before: "Find your inner peace with our amazing classes! Whether you're a beginner or experienced yogi, there's something for everyone. 🧘♀️✨ #yoga #wellness". After, with specifics from the owner's note: "Thursday 6.30pm, 14 people, most of them straight from work. By the end of the class the room was so quiet you could hear the rain. Slow flow runs every Thursday; beginners always welcome." The second works because it's a scene, not a claim.
Email subject lines for a photography studio
Before, the AI's five options included "Capture your precious memories" and "Don't miss our special offer!". After adding the facts (autumn mini-sessions, 20 minutes, a park setting, eight slots), the options became "Eight autumn slots, 20 minutes each" and "Mini-sessions are back, in the park this year". Specific numbers in subject lines give people a reason to open that "precious memories" doesn't.
A website headline for a wedding planner
Before: "Creating unforgettable weddings tailored to you." After mining enquiry emails for what couples actually worried about: "You enjoy the day. We deal with the suppliers." The second headline answers the worry couples raise most in first calls, supplier coordination, which is exactly what they're paying a planner for.
A Google Business Profile post for a personal trainer
Before: "Transform your body and mind with personalised training sessions designed to help you achieve your goals!" After: "New 6.30am small-group sessions from Monday: four people, one trainer, 45 minutes. Two places left. Tap Book to grab one." Updates on Google work best as plain news with a next step, and AI will write them well once you give it the news.
An ad headline for a tattoo studio
Before: "Express yourself with stunning custom tattoos." After: "Fine-line flash from $80. Walk-ins Saturday." Short, factual, with a price and a day. A headline that could only come from one studio.
When every post has the same shape
Even with good specifics, AI copy can feel samey because of structure: hook question, three short benefit lines, call to action, every time. Readers notice the pattern across a feed long before they notice any single phrase.
The fix is to ask for different shapes explicitly. Keep a list of five or six structures you like and name one in each prompt: a single short scene; a customer question and your answer; a number and what it means; a before and after; a behind-the-scenes detail; a straight announcement with no hook at all. Or ask for three versions of the same post in three different structures and pick one. A month of posts that rotate through several shapes reads like a person wrote them, even when AI did the first drafts. For Instagram in particular, writing captions that sound like you covers this with more examples.
A 60-second check for genericness
Before anything goes out, run four quick tests:
- The swap test. Replace your business name with a competitor's. If the copy still works, it isn't yours yet.
- The specifics count. Count numbers, names, times, places and quotes. Under two in a short post is a warning sign.
- The "so what" test. After each sentence, ask "so what?" If the next sentence doesn't answer it, one of them can probably go.
- The read-aloud test. If you'd feel silly saying it to a customer at the counter, rewrite it.
These take less time than rewriting a prompt, and they catch what bans miss. Some owners reach for AI "humaniser" tools at this stage; they mostly reword text rather than add substance, and whether a humaniser is worth paying for explains why they rarely fix generic marketing.
When the problem isn't the AI
Sometimes the copy is generic because the business hasn't decided what makes it different. If you can't say in one sentence why a customer should choose you over the three competitors nearest them, no prompt will produce that sentence for you, and the AI will politely fill the silence with "friendly and professional".
The fix sits upstream of any tool. Read your last 20 reviews and note what customers praise that competitors' reviews don't mention. Ask your last five customers why they chose you. Look at the questions people ask before booking. The answers usually reveal two or three things only you do, or do noticeably better, and those go straight into the specifics bank. From then on, the AI has something real to work with, and the generic copy problem mostly takes care of itself.
More on fixing generic AI copy
Will a better AI model fix generic copy?
A stronger model follows instructions and examples more closely, which helps, but it can't invent what makes your business different. Given the same thin prompt, most current models produce similarly safe copy. The biggest improvements come from better inputs: real details, customer language and samples of your own writing. A mid-range model with those usually beats a top model without them.
Should I pay for an AI humaniser tool?
Usually not for marketing copy. Humanisers mainly reword text to change its statistical fingerprint, which doesn't add the specifics that make copy persuasive, and can introduce errors or awkward phrasing. If your copy sounds generic, the problem is what went into it, not how it's worded. Spend the time on a specifics bank and an edit pass instead.
How many examples of my writing does the AI need?
Three to five short pieces you're happy with is usually enough to shift the tone, ideally of the same type as what you're asking for: captions for captions, emails for emails. More examples help less than better ones. Refresh them every few months with recent writing, or the AI keeps imitating how you sounded a year ago.
Further reads
- How to Write LinkedIn Posts With AI Without Sounding Like AI — The same problem on LinkedIn, with platform-specific fixes.
- How to Write a House Style Guide Your Team and AI Both Follow — One style guide for staff and AI tools alike.
- How to Build a Product Glossary AI Must Use in Every Draft — Make AI use your real product and service names.
- Brand Voice Tools Compared: Jasper, Writer and Custom Instructions — Whether a dedicated brand-voice tool is worth paying for.
- How Agencies Keep Brand Voice Consistent in AI-Written Content — How agencies keep many clients' voices distinct.
- 12 AI Marketing Mistakes Small Businesses Make (and the Fixes) — Generic copy is one of twelve common AI marketing mistakes.
- How Freelance Writers Use AI Without Losing Their Voice — Keep control of the angle and first draft, then use targeted prompts to find weak passages without flattening your writing.
- Is It Cheaper to Use AI or a Freelancer for Marketing Content? — A per-piece cost comparison of AI and freelance writers for marketing content, counting your own editing time, with three worked cases.
- How to Fact-Check AI Marketing Copy Before It Goes Live — A claim-by-claim checklist for AI-written ads, emails and product pages, with a fact bank template, red-flag words and a sign-off log.
- How to Write Facebook and Instagram Ad Copy With AI — A creative brief, angle-based prompts, Meta's recommended text lengths and the policy traps AI walks into, worked through for a real-looking cookery class ad.
- How to Write Cold Emails With AI That Stay Out of Spam — The sending set-up, DNS records, warm-up pace and AI prompt that keep cold emails in the inbox, with a before-and-after rewrite and warning signs to watch.
- How to Create Lead Magnets With AI: Checklists, Guides, Calculators — Turn the question customers ask most into a checklist, short guide or working calculator with AI, then deliver it through a sign-up form, with a bakery example.
- How to Write a Press Release With AI That Journalists Will Read — Test the angle, draft from a fact sheet, strip out the tells that get AI releases binned, and write the short personal pitch that actually gets coverage.
- Does Google Penalise AI-Written Content on a Business Website? — What Google's rules actually say about AI-written pages, how to tell a penalty from a page that never ranked, and a value check for every draft.
- Are AI-Generated Customer Personas Accurate? How to Check — Why AI personas read convincingly and still miss, plus a two-hour audit with source tagging, data checks, recognition calls and prompts that show their working.
- Tone of Voice Examples: AI Copy Before and After Editing — Before-and-after edits of real-looking AI copy for a handmade jewellery seller, with the voice rules, prompt and checks that made the new tone stick.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.