TL;DR
- Roughly twenty phrases account for most of the cliche output AI tools default to producing. The list groups into generic adjectives, status-marker verbs, and stock announcement openers.
- These phrases were generic before AI. AI did not invent them. It learned them from the corpus where business writing actually lives.
- Banned-phrase lists outperform style descriptions because they operate at the layer where the corpus already lives — the actual words on the page.
- The Toolient before-after pattern is the operational move. Replace each abstract qualifier with a concrete outcome the buyer recognizes.
- A working brand-specific list begins from the universal twenty and grows to forty or fifty entries over the first three months of editorial review.
A drafting agent reaches for the same twenty-some phrases on every prompt that does not forbid them. The list has been documented across multiple practitioner sources — Toolient, Brandfolio, Search Engine Land, Pretty Fly. The exact entries vary by source. The shape of the list is stable.
Reading the list out loud is uncomfortable. Most of the phrases have appeared in copy you have shipped or reviewed in the last twelve months.
That is the point. The list is what you stopped noticing.
This piece names the entries and explains why the agent defaults to them. It also walks through how a working copywriter turns the universal list into a brand-specific banned-phrase list the next prompt can load.
What is the AI-writing cliche list, and why does this list exist?
The AI-writing cliche list is a documented cluster of phrases that AI drafting tools produce at higher rate than the underlying corpus would predict. The cluster groups into three families.
Generic adjectives. Status-marker verbs. Stock announcement openers.
The cluster is not new. Every entry on the list appeared in business writing for decades before AI tools arrived. Press releases, product pages, LinkedIn posts, sales letters that nobody bothered to write tightly.
The agent did not invent the cluster. The agent learned the cluster from the same corpus where the cluster already lived.
The reason the list matters in 2026 is volume. The same draft that one writer used to produce in a day now arrives in ninety seconds.
The cluster shows up at the same rate it always did. The output volume multiplied by ten. The cliches multiplied with it.
The Brandfolio analysis names the mechanism. The agent is a pattern-matching system. It defaults to the statistical middle of its training data. The middle is where the cliches live.
The agent does not produce cliches because cliches are easy. It produces them because cliches dominated the data it learned from. The list is the antidote.
Naming the cluster is the first step toward forbidding it. A banned-phrase list with the cluster on it pulls the agent off the middle. A draft without that list inherits the middle by default.
Which generic adjectives signal a draft was averaged into the middle?
Eight adjectives recur across the practitioner sources — innovative, next-level, seamless, cutting-edge, high-quality, robust, revolutionary, game-changer. A buyer skimming a homepage encounters at least one of these on most pages. Most of the time the word does no work. The word names a quality the brand is asserting without naming the evidence behind the assertion.
The Toolient before-after pattern shows the edit. Strike the adjective. Replace it with the concrete outcome the buyer would experience.
"Lightweight design" becomes a sentence about carrying the bag all day without shoulder fatigue. "High-quality materials" becomes a sentence about lasting three or more years under daily use. "Innovative dashboard" becomes a sentence about cutting the weekly report from four hours to forty minutes.
The edit is mechanical. Strike the abstract qualifier. Name the outcome the buyer felt the last time the buyer used something this product is competing with.
The edit also forces a question the abstract version dodged. What is the actual outcome? If the answer is "we are not sure," the homepage cannot honestly carry the claim.
The cliche was hiding the gap. Removing the cliche surfaces the gap. The brand either fills the gap with a real outcome or rewrites the line around what the product actually does.
For more on the underlying diagnostic, see diagnose copy tactics versus positioning.
Which status-marker verbs flag corporate-speak in AI output?
The verbs in the cluster do a different job than the adjectives. Adjectives assert quality. The verbs perform belonging. They signal that the writer knows the conventions of business writing.
The list is short — leverage, synergize, unlock, utilize, skyrocket, optimize, streamline, enable, empower, drive.
Each one has a plain-English replacement. Leverage and utilize mean use, unlock means open, skyrocket means rise sharply, synergize means cooperate, empower means let, and drive means move. The replacement is shorter, more specific, and easier for a buyer to picture.
The reason the agent reaches for the status-marker verbs is the same reason marketing teams reached for them before AI arrived. The verbs sound like business. They felt safe in a slide deck. They drifted into homepage copy because they sounded official.
The fix is the banned-phrase list. The list names the verbs. The next prompt forbids them. The agent reaches for plain English instead — use, open, rise, cooperate.
The replacements are not poetry. They are also not corporate-speak. They sit in the middle of the sentence and let the sentence do the work.
The Toolient practitioner rule sets a useful constraint here. One adjective per sentence. No status-marker verbs.
The agent obeys both rules when both rules are explicit. The output reads measurably tighter inside three drafts.
Which announcement openers should never reach a published page?
The openers are the easiest to spot and the most expensive to leave in. They appear at the top of emails, the lead of blog posts, the hero of landing pages. They cost the reader’s attention before any actual content arrives.
The cluster is small. We are excited to announce. We are thrilled to share. Pleased to announce. In a world where. Imagine if you could. The perfect solution. Today, more than ever. Now more than ever.
Each opener has a structural problem. It starts on the brand’s emotion rather than on the reader’s situation. It signals announcement-language rather than information. It teaches the reader that the message is going to be about the brand, which is the message the reader did not come for.
Replace the opener with the news itself. We are excited to announce that pricing is changing becomes Pricing is changing on June first. In a world where customer expectations keep rising becomes a specific customer expectation that has actually risen, with a number behind it. The perfect solution for distributed teams becomes the specific problem distributed teams have, named in the team’s own words.
The replacement is shorter. The replacement also leads with the reader’s situation. Both moves are what good news writing has done since the 1920s. The cluster represents writing that lost the lesson.
For more on the editing pass that replaces an opener-cliche with a real lead, see humanize the draft or write a new prompt.
How does the cliche list connect to conversion craft and the awareness spectrum?
The cliches are not just bad style. They miss the buyer at every level of the awareness spectrum.
Eugene Schwartz’s spectrum still anchors conversion craft sixty years on. Five levels — unaware, problem-aware, solution-aware, product-aware, most-aware. The right copy follows the level rather than the product.
A problem-aware buyer needs the pain named in the buyer’s own words. Innovative and seamless do not name a pain. The buyer scans past the words because the words point at no specific feeling.
A solution-aware buyer needs comparison. Cutting-edge and next-level compare against nothing. The buyer cannot tell what the product is better than because the words index against the abstract middle of the category.
A product-aware buyer needs proof. High-quality and robust assert quality without showing the evidence. The buyer who is comparing options skims past the assertion and looks for the outcome, the case study, the data point. When the outcome is not there, the page loses against the page that names it.
A most-aware buyer needs the offer, the guarantee, the bonus. The cliche-cluster page is doing the language of marketing without doing the work of marketing. The buyer who was ready to act before reading the page is still ready to act after reading it. The page added no information.
The cliche list is the diagnostic. A page that depends on the cluster is a page that did not name the buyer specifically enough at any level. The fix is the awareness audit — which level does this page serve, and does the copy speak in the words that level uses?
Why does deleting cliches improve AI citation quality?
A page scrubbed of cliches reads better to a human buyer. The same page reads better to the AI engines crawling for citations.
The 2026 evidence on AI citation quality points at the same edits the voice rules point at. Specific outcomes beat vague qualifiers. Named entities beat generic descriptions.
Dated facts signal recency. Lived-experience claims signal E-E-A-T to engines that weight Experience.
The Toolient before-after pattern produces all four signals at once. Lightweight design becomes a specific outcome (carrying it all day without shoulder fatigue). High-quality materials becomes a dated fact (lasts three or more years under daily use). Innovative dashboard becomes a lived-experience claim (cuts the report from four hours to forty minutes, with the named team and the named tool).
The same edit that beats the cliche cluster on voice grounds beats the cliche cluster on citation grounds. The wins compound. A page rewritten with the banned-phrase list and the concrete-outcome rule reads sharper to the buyer and extracts more cleanly to the engine.
The cmswire 2026 piece names the underlying claim in plainer language. "The first letter in AI stands for artificial," the piece argues. What customers reach for under stress is "another A: authenticity."
Generic copy reads as artificial because it was averaged. Specific copy reads as authentic because it was edited away from the middle. Buyers and engines reward authentic copy for almost the same reasons.
How do you build a brand-specific banned-phrase list in one afternoon?
The universal list is the starting point. The brand-specific list grows from there.
Pull the last six months of published copy. Open the agent’s recent drafts in a separate tab. Run a side-by-side read.
Mark every word that appears in both columns and that the brand never said in person. Empower, enable, innovative, seamless. The matches are the agent’s defaults that the editorial pass let through.
Add each match to the brand-specific list. Twenty entries by the end of the read.
Read the universal cluster against the brand’s actual voice axes. The brand placed itself somewhere on each axis — fun-serious, formal-casual, respectful-irreverent, matter-of-fact-enthusiastic. Some entries on the universal list will conflict with the brand’s chosen placement.
Flag them. Other entries on the universal list will not conflict but still leak into drafts. Keep them.
Add a third layer. Phrases the editor has had to strike from drafts more than once in the last month. Each one earns a spot on the list with a one-line note explaining why.
The list is now thirty to fifty entries. That is the working size.
Beyond fifty, the list becomes too long for the agent to handle in a single prompt. The priority entries get diluted. Below thirty, the list misses too many of the agent’s actual defaults.
Save the list inside the voice document’s AI-prompting addendum. Every drafting prompt loads it. The pattern stops.
Other questions worth answering
When does a project owner need to disclose machine-assisted authorship to readers?
When the reader would care about provenance. Healthcare, legal, journalism, and academic publishing all sit at the high end. Marketing and routine sales copy live at the low end. The cmswire February 2026 piece argued automation should handle the simple, leaving humans to own the complex when trust matters most.
How do you decide between hand-editing an LLM result versus rewriting the prompt?
The choice comes down to whether the argument arc holds. Reverse-outline the draft in under a minute. If the arc works, hand-edit. If the bones break apart, rewrite the prompt instead.
Robin Da Silva named this the read-as-argument test. Eugene Schwartz’s 1966 awareness framework still anchors the diagnostic. Hand-editing fixes voice. Rewriting fixes structural problems no sentence-level edit can repair.
Do tools like GPTZero reliably catch machine-assisted text in 2026?
No. GPTZero and similar detection tools in 2026 fail at error rates around 60 percent on news queries. False positives on human work and false negatives on edited machine work are both common. Treat any score as one weak hint, never as a verdict on a piece.
What part of the work stays human after an LLM produces a near-final piece?
Taste. Roughly the last 20 to 30 percent of any draft is human judgment work. What to keep, what to cut, where to add a felt detail, where to break the rhythm. Around 70 to 80 percent of any 2026 LLM result is the easy part.
Which phrase from the cliche list would you ban from your drafts first?
Pick one. Not the one you find most annoying. The one your last three drafts contained.
Open the most recent piece you shipped. Read every paragraph. Mark every entry from the universal cluster you find.
Most pages have at least three. Some have a dozen.
Pick the phrase that appeared most. That phrase is the agent’s default for your brand at the moment. That phrase is doing the most work to pull your output toward the middle.
Add the phrase to your banned-phrase list with a note. The note says where the phrase appeared, what it was replacing, and the concrete-outcome edit you would make instead.
The note matters. The next time the agent slips the phrase past the prompt, the editor needs the rationale. The rationale defends the deletion against a contributor who liked the line.
Write the next draft with the list active. The agent reaches for a different word. The line lands closer to the buyer.
If your team has been shipping AI-augmented copy and recent pieces are passing review while still feeling generic, you can contact me here. Send me one piece and the prompt the team is using to draft it. I will mark the cliche-cluster phrases that survived editorial review and explain the banned-phrase list entries that would have caught them. There is no charge and no follow-up sales call.