← All posts

September 8, 2026

Restricted Terms in Marketing Copy: The Words That Get Flagged and What to Say Instead

restricted termscompliancebrand rulesai content

A restricted term is a word or phrase a brand has decided can't appear in its content without a specific condition being met: a source, a disclosure, a legal review, or never at all. Restricted-term lists exist because the same handful of words cause most of the trouble, and because a machine can check for them on every piece while a human can't.

This is the list we see most brands start from, grouped by why each family is a problem. The "say instead" column is the part people actually use.

Why a list, and not just a careful writer

A careful writer knows not to write "guaranteed". A careful writer also writes three hundred pieces a year and is careful on two hundred and ninety of them. An AI generator is neither careful nor careless; it produces whatever the prompt makes likely, and persuasive words are likely.

A list turns "be careful" into a check that runs every time. The list doesn't need to be clever. It needs to be complete for the brand, carry a reason for each entry, and run before a person reviews. We covered where that check sits in the workflow in How to Build an AI Content Approval Workflow That Holds at Volume.

The families

Outcome promises

Flagged Why Say instead
guaranteed, guarantee Promises a result the brand can't ensure for every customer "designed to", "built to", or describe the actual mechanism
risk-free Almost nothing is; implies a refund or protection that may not exist State the actual policy: "30-day returns"
results in X days/weeks A timed outcome claim that needs substantiation for the typical customer Describe the process and the typical timeline as a range, with a source
you will, you'll see Second-person certainty about an outcome "many customers", "designed so that"

Outcome promises are the most common flag in most brands' content because they're the most natural way to sell. The fix is rarely to delete the sentence; it's to replace the promise with the mechanism.

Superlatives and rankings

Flagged Why Say instead
best, #1, number one Comparative claim; needs evidence against named competitors The specific attribute: "the only one with X" if true, or a real award with its year
fastest-growing, leading, top-rated Same; "top-rated" also implies reviews that may not be independent Cite the rating and its source, or cut
award-winning Fine if the award is real and recent; flag to check Name the award and year

Superlatives are worth flagging for substantiation rather than deletion. Sometimes the evidence exists. The rule should say "attach the source" and the reviewer decides.

Health, safety and efficacy

Flagged Why Say instead
cure, treat, heal, prevent Medical claims; regulated in most markets regardless of product "supports", "designed for", and only what the evidence covers
clinically proven, scientifically proven Needs the study; "proven" is a high bar "in a [year] study of [n] participants, [result]" with the citation
safe, non-toxic, hypoallergenic Absolute safety claims; each has a regulatory meaning The specific certification or test, named
flame-resistant, waterproof, certified Performance claims that need the supplier's certificate Name the standard and the certificate

For most brands this family is small but the consequences are large. A single unsupported health claim can be a regulator's whole case.

Money and earnings

Flagged Why Say instead
earn, income, profit, ROI, six figures Earnings claims; in franchise content these are financial performance representations that belong in FDD Item 19 Route to legal, or use only the approved Item 19 language
free Legally specific; conditions must be stated clearly and prominently "free with X", with the condition in the same sentence
lowest price, cheapest Price comparison; needs to be true against named competitors at the time "from $X", with conditions
save X% Needs a reference price that was genuinely offered State the reference price and the period

Franchise brands have the strictest version of this family; we wrote about it separately in Franchise Social Media Compliance When AI Writes the Posts.

Endorsements and testimonials

Flagged Why Say instead
a quoted customer with no name or source Fabricated or unverifiable endorsement A real, named, consented quote, or none
"our customers say", "people love" Implied testimonials without a basis A specific, sourced review
influencer or partner praise with no disclosure The FTC Endorsement Guides (16 CFR Part 255) require material connections to be disclosed Add the disclosure in the same piece

A generator asked for "social proof" will invent it. This family exists to catch that.

Competitors

Flagged Why Say instead
any competitor name Comparative claims need substantiation; misstatements are actionable Describe your own attribute without the comparison, or substantiate it
"unlike other", "the alternative to" Implied comparison Same

Some brands allow competitor mentions in specific contexts (a comparison page with sourced data, for instance). The rule can be "flag and route", not "forbid".

The brand's own list

Every brand has words of its own: a product name that must be spelled one way, a tagline that must not be altered, a topic leadership has decided the brand doesn't comment on, a disclaimer that must accompany any mention of a regulated product. These are the entries nobody else can write for you, and they're usually the ones that catch the most.

How to encode the list

Three fields per entry, and the third is the one people skip:

  1. The term, and its obvious variants. "Guarantee" should catch "guaranteed" and "guarantees".
  2. The reason. One sentence, in plain language, naming the rule or the risk. This is what the reviewer and the writer see when it fires. A flag with no reason is a red highlight; a flag with a reason is a lesson.
  3. The remediation. What to do instead. Often the "say instead" column above.

Then attach the list to the brand, not to a person or a prompt document. If you run several brands, each gets its own; a term that's restricted for a supplement brand is fine for a software brand. And put the check before human review, so the reviewer sees the verdict and spends their attention on the flags rather than hunting for the words.

Frequently asked questions

Should a flagged term block publishing, or just warn?

Neither, on its own. A flag should stop the piece from being approved until a person resolves it, either by editing or by recording why the rule doesn't apply here. Blocking outright makes people route around the check; warning only makes them ignore it.

How many terms should the list have?

Fewer than you think, with better reasons. A list of forty entries that each carry a reason and a remediation will be used. A list of four hundred bare words will be muted within a month. Start with the families above, add the brand's own, and grow it from real flags.

Does the list replace legal review?

No. It replaces the part of legal review that is pattern matching, so that the lawyer's time goes to the judgment calls. The entries that need "route to legal" as their remediation are exactly the ones the list can't resolve on its own.

Where the list lives in practice

In Azimuth, each brand's restricted terms and banned topics live in its profile, every draft is screened against them before review, and each flag carries its reason and remediation on the piece and in the audit trail. What we do with your rules and your data is set out on the Trust Center.

Run governed AI content across every brand you manage

Azimuth grounds generation in each brand's own knowledge, screens every piece against its rules, and keeps a human signature on everything that ships.

Talk to us