A small set of words gives AI prose away: delve, tapestry, seamless, robust, leverage, elevate, landscape, plus stock openers like In today's fast-paced world. Sentence habits matter as much: the not just X, it's Y contrast, adjectives in threes, and paragraphs starting Moreover,. Ban them, and replace each with something specific.
No single word proves a text was machine-written; plenty of people write robust. What readers notice is density: five of these in a paragraph and nothing concrete to hold on to. That is why deleting words isn't enough. The fix is to replace each one with a fact, a number or a step, which is also what makes a page worth reading.
Why these words give AI writing away
Language models pick words that are likely in context, and in business writing the likely words are the safe, abstract ones: phrases that fit any company in any sector. They also carry habits from the polished marketing and corporate text they learned from, which is why the same verbs and adjectives turn up whether the draft is a job advert or a service page. The underlying reasons, and the fixes beyond word choice, are covered in why your AI marketing copy sounds generic. This tutorial is the practical list.
A useful test for any word on it: could a competitor paste this sentence onto their own website unchanged? If yes, the sentence is doing no work for you, whatever the vocabulary.
The banned-phrase checklist
Seven groups, each with why the item reads as generic and what to write instead. The "search for" stems catch every form (leverag finds leverage, leveraging and leveraged), which matters because models slide between forms.
1. Inflated verbs
| Search for | Why it reads as generic | Write instead |
|---|---|---|
delv | Almost never used in everyday business writing | look at, explain, cover |
leverag | Consultancy jargon for a plain idea | use (or name what you use) |
utilis, utiliz | A longer word for the same meaning | use |
facilitat | Hides who does what | help, run, arrange |
empower, elevat, unlock, harness | Promise a feeling, not an outcome | the outcome: "cut the month-end close from five days to three" |
streamlin, revolutionis | Claim change without saying what changed | the step removed: "no more re-keying invoices" |
navigat, embark, spearhead, underscore | Metaphors doing a plain verb's job | deal with, start, lead, show |
2. Abstract nouns and scene-setters
| Search for | Why it reads as generic | Write instead |
|---|---|---|
landscape, realm, ecosystem | Grand words for "market" or "area" | market, sector, the specific field |
tapestry, testament | Literary flourishes nobody uses at work | cut, or "shows" |
journey (figurative) | Turns a process into a story | process, the steps, the timeline |
synergy, paradigm, game-changer | Worn-out management vocabulary | what actually changes, in numbers |
plethora, myriad, treasure trove | Vague quantity dressed up | the actual number |
3. Empty adjectives
| Search for | Why it reads as generic | Write instead |
|---|---|---|
seamless, hassle-free | Every service claims it; nobody believes it | what the customer doesn't have to do |
robust, comprehensive | Asserts quality without evidence | what is included, how it's tested |
cutting-edge, innovative, dynamic | Self-description readers discount | the specific new thing |
tailored, bespoke, holistic | Claimed by everyone, explained by no one | how the work changes per client |
pivotal, crucial, transformative, vibrant | Emphasis without content | cut, or give the reason it matters |
4. Openers and closers
The first and last sentences of a piece are where AI defaults cluster, because they are where the model has least to go on. Ban these outright:
In today's fast-paced world,In an ever-evolving,In the digital age: scene-setting nobody needs. Start with the point.Whether you're a...or a...: tries to address everyone, so speaks to no one. Name the reader you mean.Are you looking for...?followed byLook no further: an advert cliché that wastes the opening line.I hope this email finds you well: the most skipped sentence in business email. Open with why you're writing.In conclusion,Ultimately,,At the end of the day: signposts announcing a summary. End on the next step instead.Don't hesitate to reach out,Feel free to contact us: replace with the actual route: "Reply to this email or call [name] on [number]".
5. Sentence patterns
These don't show up in a word search, so train your eye for them:
- The contrast frame:
It's not just a job, it's a career,This isn't about X, it's about Y. It sets up a claim nobody made in order to knock it down. - The note-taking opener:
It's important to note that,It's worth noting. Delete the phrase; keep the note. - Linking-word starters:
Moreover,,Furthermore,,Additionally,at the start of sentences. Usually removable with no loss. - Triplets: "fast, reliable and affordable". Three adjectives in a row signal filler; keep the one you can prove.
- The question reveal:
The result?followed by the answer. Just state the result. - The sweep:
From small startups to large enterprises. Name who you actually serve. - Dashed asides: several long dashes per paragraph, each adding a clause. Use commas, or two sentences.
6. Hedges and intensifiers
truly, incredibly, arguably, a wide range of, a variety of, various, numerous, plays a crucial role, stands out as. Each either inflates a claim or hides a number. "A wide range of services" becomes "payroll, bookkeeping and year-end accounts".
7. Your own sector's clichés
Every sector has its own layer on top of the general list, and AI reproduces it faithfully because it learned from the sector's worst examples. For recruitment: rockstar, ninja, self-starter, fast-paced environment, exciting opportunity, competitive salary, work hard, play hard, hit the ground running. For mortgage and financial advice: dream home, peace of mind, financial freedom, best rates. Spend ten minutes listing yours from your competitors' pages; that list is often more useful than the general one.
8. Customer-reply clichés
Support and client emails drafted by AI have a vocabulary of their own: Thank you for reaching out!, I completely understand your frustration, Rest assured, I apologise for any inconvenience, Thank you for your patience, Great question!. Each one sounds sympathetic and tells the customer nothing. An illustrative example from a law firm, where a client had emailed about a delay:
Thank you so much for reaching out! I completely understand
your frustration regarding the delay. Rest assured, we are
committed to providing you with the highest level of service.
Please don't hesitate to reach out if you have any further
questions. Thank you for your patience!Five clichés, zero facts: the client still doesn't know why it's late or when it will be done. The rewrite:
Thank you for your email. The delay is on our side: we were waiting for the other party's lawyer to return the signed documents, which arrived this morning. I will send you the completed paperwork by 5pm tomorrow. If you need anything before then, call me directly on [number].
The cause, the fix and a deadline, in four sentences. That is the pattern for every reply: say what happened, what you are doing, and by when. Sympathy is better shown by a date than by a phrase.
Hard bans, watch words and context exceptions
A flat list treats delve and tailored the same, but they aren't. Sort your list into three tiers so the rules are strict where they should be and flexible where the word is sometimes right.
A filled-in list for an illustrative three-adviser mortgage brokerage:
| Tier | Rule | Items |
|---|---|---|
| Hard ban | Never appears in anything client-facing | guarantee(d), best rates, hassle-free, dream home, peace of mind, In today's market, Whether you're a first-time buyer or, plus groups 1 to 5 above |
| Watch | At most one per piece, and only with a specific next to it | tailored, expert, trusted, straightforward, support |
| Context exception | Allowed only in the stated meaning | leverage only in its financial sense (borrowing to invest); navigate never figuratively |
The guarantee ban is doing more than style work there. In regulated sectors a promised outcome can be a compliance problem, so the banned list and the compliance rules overlap. If you keep a house style guide, the hard-ban tier belongs in its required-and-banned-wording section, so there is one source of truth.
Building your brand's own list in 30 minutes
The general list catches what every AI draft does. Your brand's list should also catch what your drafts do, which is often different. Collect ten recent AI-assisted pieces (emails, posts, a web page or two), remove client names, and ask a chat assistant to find the repeats:
Below are ten pieces of writing from our business. List every
phrase of two to six words that appears in three or more of
them, with a count. Then list any sentence that could appear
unchanged on a competitor's website. Don't rewrite anything.
[paste the ten pieces]For an illustrative bookkeeping firm, the answer included:
REPEATED PHRASES
"every step of the way" (5) "take the stress out of" (4)
"our clients" (9) "so you can focus on" (6)
COULD APPEAR ON A COMPETITOR'S SITE
"We make bookkeeping simple, so you can focus on what you
do best."Sorting that list takes judgement: "our clients" is a normal phrase the firm is right to use, so it stays off the list; a repeat isn't automatically a cliché. "Every step of the way", "take the stress out of" and "so you can focus on" go straight onto the hard-ban tier, because they appeared across emails, posts and pages alike and say nothing specific. And the assistant missed "a quick and easy", which the owner spotted in four pieces on a second read, so treat the output as a first pass. Thirty minutes of this usually yields five to ten phrases that no general list would contain and that your readers have been skimming past for months.
Loading the list into your AI instructions
Put the hard bans and watch words into whatever your team drafts in: a shared project's instructions, a Gemini skill (the successor to Gems), or your custom instructions. Two phrasing details help. List every form of each word (guarantee, guaranteed, guarantees), and state the tier rules in one line each. Then expect the problem every banned list meets: synonyms.
An illustrative test. The instructions ban the general list; the prompt asks a bookkeeping firm's assistant to "write a 60-word introduction for our payroll service page". The output:
Payroll shouldn't keep you up at night. Our smooth, all-in-one
service takes care of every detail, so you can dig into what
matters most: growing your business. With our reliable,
friendly team behind you, you'll have complete confidence that
your people are paid right, every time.Not one banned word, and it is still generic. "Smooth, all-in-one" has replaced seamless, comprehensive; "dig into" has replaced delve into; "reliable, friendly" is the same triplet habit; "what matters most" and "every time" are claims with nothing behind them. Banning words moves the model to their nearest neighbours. The fix is to ban the move, not only the vocabulary, by adding a positive rule and the facts it needs:
Instead of describing our service with adjectives, state
facts: what we do, how often, by when, for how many clients.
Facts: we run payroll for 46 small employers, weekly or
monthly. Hours due by Tuesday, staff paid Friday, one-page
summary emailed the same day.The same prompt then produced: "We run payroll for 46 small employers, weekly or monthly. Send hours by Tuesday; your staff are paid on Friday, and you get a one-page summary the same day. Changes that miss the cut-off go into the next run, not a lost week." Nothing on the banned list, and more importantly, nothing a competitor could paste onto their own site.
A job advert before and after
Job adverts are the densest source of these phrases in most small businesses, so they make the clearest worked example. An illustrative recruitment agency drafted an advert for a client, an accounts assistant role at a twelve-person construction supplier, with a single-line prompt. The opening of the 260-word draft:
Are you a dynamic self-starter looking to elevate your career?
We're a fast-growing, innovative business where every day is
different. In this exciting opportunity, you'll leverage your
skills in a fast-paced environment, seamlessly managing a
comprehensive portfolio of accounts. We offer a competitive
salary and a vibrant, supportive team. It's not just a job,
it's a career. Don't hesitate to apply!Against the checklist, those 72 words contain 15 hits: the question opener, dynamic, self-starter, elevate, innovative, every day is different, exciting opportunity, leverage, fast-paced environment, seamlessly, comprehensive, competitive salary, vibrant, the contrast frame and the Don't hesitate close. The full draft had 21 hits in 260 words, about 81 per 1,000 words. The consultant replaced every hit with a fact from the client's brief:
Accounts assistant at a twelve-person construction supplier. Office-based, 37.5 hours a week, $34,000 to $38,000. You'll process around 400 supplier invoices a month, chase overdue customer accounts and prepare the weekly payment run for the finance manager to approve. You'll need a year's experience in accounts payable or receivable. The finance team is four people, and you'll report to the finance manager, who has been with the business for nine years.
The rewrite is 180 words in full, with zero hard-ban hits and one watch word, and it contains nine facts the original lacked: job title, employer size, working pattern, hours, salary band, invoice volume, main tasks, experience needed and reporting line. Candidates can decide from it whether to apply, which is the entire job of an advert. Wording that screens people out unfairly is a separate check from generic wording; writing job adverts with ChatGPT without biased language covers that one.
Checking a draft against the list in five minutes
Instructions reduce hits; they never remove them all. Before anything client-facing goes out, one person runs this check:
- Search the stems of your ten most frequent offenders with your editor's find tool. Ten searches take about two minutes.
- Read only the first and last sentence of each paragraph. That is where openers, closers and linking-word starters sit.
- Look for threes and contrasts: any list of three adjectives, any "not just" or "isn't about" frame.
- Count watch words. If there are more than two per 1,000 words, rewrite the densest paragraph.
- Apply the real test: does every paragraph contain at least one specific, such as a number, a name, a date or a step? A paragraph without one needs rewriting whatever words it uses.
Step 5 is the one that matters. A draft can pass steps 1 to 4 and still say nothing; a draft that passes step 5 rarely fails the others badly.
When a banned word is the right word
Blanket bans misfire, and the usual way they misfire is a find-and-replace. An illustrative financial planning firm told an assistant to replace every leverage on its website with "use". A guide on property investment went from the risks of leverage when buying a second property to "the risks of use when buying a second property", which is nonsense, and nobody noticed for a month. In finance, leverage is a precise term for investing with borrowed money, not jargon.
The same trap exists in other sectors: a garden design firm can't ban "landscape"; an IT support firm may need robust in a technical specification where it has a defined meaning; a marine business uses "navigate" literally. Three rules prevent the damage:
- Never replace all. Review every hit in context.
- Keep the context-exception tier from the table above, and write down the allowed meaning.
- Apply the list to marketing and correspondence, not to technical documents where terms are defined.
Keeping the list current
AI tools' favourite words shift as models are updated, and your own team develops habits too. Once a month, the person who edits most of your client-facing copy notes the phrases they deleted. Any phrase deleted three or more times goes onto the watch list; any watch word that keeps turning up moves to the hard bans. Once a quarter, run the list against your last ten published pieces and retire items that no longer appear: a list that only grows becomes one nobody reads.
Your existing pages need the same treatment as new drafts. Start with the ten pages customers see most (home, services, about, your busiest tutorials or guides, current job adverts), run the stem search on each, and fix the three with the most hits first. On most small-business websites that is an afternoon's work, and it tends to improve the pages that matter most for enquiries.
One last point on purpose. Avoiding these words won't make AI content undetectable, and that shouldn't be the aim; whether customers can tell a reply was written by AI depends far more on whether it answered their question. Tools sold as "humanisers" chase the wrong target for the same reason, as the tutorial on AI humaniser tools explains. The list earns its place because generic phrasing wastes your reader's attention, and specific writing is what keeps it.
Banned-phrase lists: common questions
Will removing these words stop AI detectors flagging our content?
Don't aim for that. Detectors are unreliable: OpenAI withdrew its own AI-text classifier in July 2023 after it identified only 26% of AI-written text and wrongly flagged 9% of human writing. Swapping words to dodge detectors produces odd prose without fixing what readers notice: vague claims and missing specifics. The list is useful because readers and clients skim past generic phrasing, not because software looks for it. Write so a customer learns something concrete in every paragraph.
Should our human-written content follow the same list?
Yes. The list describes generic writing, and people produce plenty of it too, especially in job adverts, proposals and web copy. Applying one list to everything is simpler than tracking who or what drafted each piece, and it keeps your pages consistent. Staff usually adopt it quickly once they see a before-and-after example from your own material.
How long should a brand's banned-phrase list be?
Short enough to remember: around 20 to 40 hard bans plus a watch list of 10 to 20 words. A list of 200 terms gets ignored by people and applied patchily by AI tools, which tend to follow short instruction lists better. Add a word only when it keeps turning up in drafts, and retire words that stop appearing.
Further reads
- Tone of Voice Examples: AI Copy Before and After Editing — More before-and-after edits showing specifics replacing filler.
- How to Check AI Content for Plain English and Readability — Readability checks that pair well with a banned list.
- How to Build a Brand Voice Guide That AI Can Follow — Describe the voice you want, not just the words you don't.
- How to Write Job Descriptions With AI That Attract Good Hires — Job descriptions with AI that attract good hires.
- How to Draft Customer Email Replies With AI That Sound Like You — Customer email replies with AI that sound like you.
- How Much Does Human Editing of AI Content Cost? — What a human line edit costs when the list isn't enough.
- Using AI in a Yoga or Pilates Studio While Keeping Your Voice — Why AI makes every studio sound the same, and the voice sheet, before-and-after edits and teacher-bio method that keep yours recognisably yours.
- How Coaches Use AI to Build Workbooks, Exercises and Programmes — Turn your coaching method into a programme and workbook with AI doing first drafts, while the method, the edits and the claims stay firmly yours.
- How Agencies Keep Brand Voice Consistent in AI-Written Content — A filled-in voice card, two client voices side by side, a sample-led prompt and a five-line scoring rubric for keeping AI copy on-brand.
- How Tutoring Centres Use AI to Write Parent Progress Reports — Short session notes in, specific parent reports out: the note fields, prompt and checks that cut a term of reports from 20 hours to about 7.
- Donor Thank-You Letters With AI That Still Feel Personal — Personal means specific, not a first-name merge: feed AI each gift's details and true impact stories, tier the response, and keep a real signature.
- How to Write Website Articles With AI That Still Rank on Google — A step-by-step method for AI-assisted website articles that rank: evidence packs, outline and draft prompts, the edit pass, and what to check in Search Console.
- How to Write Instagram Captions With AI That Sound Like You — Build a voice file from your own captions, feed the AI raw notes instead of topics, and use a read-aloud edit so every caption sounds like you wrote it.
- How to Write LinkedIn Posts With AI Without Sounding Like AI — Start from a two-minute voice memo, let AI tighten without rewriting, and remove the tells readers spot in a scroll, with a before-and-after post and hooks.
- Brand Voice Tools Compared: Jasper, Writer and Custom Instructions — Prices, voice features and banned-word enforcement compared for Jasper, Writer and shared ChatGPT or Claude instructions, with a scored trial.
- Social Media Pre-Publish Checklist for AI-Drafted Posts — A grouped pre-publish checklist for AI-drafted social posts, with why and how to check each item, platform limits, AI labels and a worked example.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.