Text Guidelines for AI-Written Ad Copy: Keeping Performance Max and AI Max On Brand in 2026

You know the scenario: text customization is on in Performance Max, the campaign cheerfully generates headlines, and the asset report shows “cheapest prices ever”, “guaranteed results” and a chummy tone from a brand that has spent eight years being deliberately formal. Until recently the choice was binary — tolerate it or switch automation off entirely. In 2026 there is a third option: text guidelines in Performance Max, a mechanism that constrains generation with words rather than a toggle.

Here is the practical version: the two types of guideline, their limits, where they live, the unpleasant side effect on assets already serving, and how to phrase rules so they actually filter instead of decorating the interface.

Read this before you switch anything on: when you add guidelines, existing Google-optimised text assets that do not adhere to them stop serving and get marked as removed. There is no single button to undo that, so guidelines are something you write deliberately and in advance — not something you “just try and see”.

What text guidelines in Performance Max and AI Max actually are

Text guidelines are campaign-level constraints applied to automatically created text assets. As of September 2026 the feature is in beta, is configured per campaign, and works in two places:

• Performance Max — through the brand settings (Brand Guidelines) and asset optimisation section;
• Search campaigns with AI Max enabled — alongside the other text customization controls.

There is one prerequisite: text customization must be turned on. The logic is obvious — guidelines constrain generation, and with no generation there is nothing to constrain. All Google Ads interface languages are supported.

If you have not yet worked through how AI Max changes search campaigns, start with the article on AI Max in search campaigns — text guidelines govern exactly that automation layer, and they matter most for accounts that went through the Dynamic Search Ads to AI Max migration.

The two guideline types and their limits

Guideline type	Limit per campaign	What it does	Notes
Term exclusions	up to 25	Blocks specific words and phrases in generated copy	Case-insensitive but language-specific: excluding “cheap” does not exclude its Spanish or Japanese equivalent
Messaging restrictions	up to 40	Specifies which concepts, associations and styles to avoid	Acts as instruction to the model, not a hard filter — a strong steer rather than a guarantee

The distinction matters. A term exclusion is a literal stop-list: the word is either present or it is not. A messaging restriction is an instruction along the lines of “no superlatives”, “no question-style headlines”, “never state a delivery timeframe” — and how well it is respected depends on how unambiguously you wrote it.

How this differs from brand exclusions

Text guidelines get confused with brand exclusions, but they operate on different layers. Brand lists control which brands and brand queries you serve on at all — covered in the piece on brand lists and brand exclusions. Text guidelines control the wording inside the ad. If the goal is “never name a competitor in copy”, that is a term exclusion; if it is “never show on their brand query”, that is an entirely different tool.

The side effect that rules out casual testing

The critical technical detail: once guidelines are added, existing Google-optimised text assets stop serving and are marked as removed if they do not adhere to the new rules.

Practical consequences:

1. Export the asset report first, with performance data. If you lose a high-performing generated headline, you will at least know which message was working and can add it back as your own asset. How to read that report is in the breakdown of asset performance and Ad Strength.
2. Do not add guidelines in bulk “just in case”. Every unnecessary constraint can knock out an asset that was earning.
3. Expect a short-term drop in asset volume: the stricter the rules, the fewer options generation has — which makes your own assets more important, not less.

How to write guidelines that work

Google’s own advice comes down to four principles: be specific, include examples inside the guideline, do not bundle several requirements into one, and do not issue instructions that conflict across asset types. In practice:

Weak	Why it fails	Better
“Write professionally”	No testable criterion — the model cannot tell what violates it	“Do not use slang or contractions; keep a formal register”
“No promises, no prices, no competitors”	Three requirements in one — partially obeyed at best	Three separate guidelines, one requirement each
“Avoid exaggeration”	Too vague to act on	“Do not use superlatives: best, cheapest, number one, only”
“Be persuasive but cautious”	Internally contradictory	“Do not use time-limited urgency unless a promotion exists in the feed”

A useful test: phrase the guideline so that you could check someone else’s copy against it in five seconds. If enforcing it would need a discussion, the model will not manage it either.

Scenarios where guidelines pay off immediately

• Regulated categories. Finance, health, education: no promises of outcomes, returns or “guaranteed approval”. A single term exclusion can close off a whole class of legal exposure.
• Comparative claims. Blocking “better than”, “cheaper than”, superlatives and competitor names.
• Tone and address. Consistent register, no emoji, no exclamation-driven phrasing.
• Prices and timeframes. Forbid specific figures unless they come from the feed. If prices must appear in the ad, that is a job for ad customizers on business data feeds, not for generation.
• Landing page alignment. When generated copy promises something the page does not deliver, you lose conversion rate and risk eligibility problems; requirements are collected in the article on Google Ads destination requirements.

A six-step rollout

1. Gather evidence. Export every automatically created text asset from the last 60–90 days with its performance. Write down the specific phrasings you object to — guidelines should be built from real examples, not imagined fears.
2. Sort problems into two buckets. What a stop-word fixes (specific words) versus what needs a style instruction (message, tone, promises).
3. Start minimal. Five to eight term exclusions and three to five messaging restrictions. Not 25 and 40: the wider the filter, the more working assets you remove.
4. Pilot on one campaign. Guidelines are campaign-level, so pick one — and not your flagship. Running it through Experiment Center gives a cleaner read than before/after on a live campaign.
5. Fill the slots yourself. The stricter your rules, the more the campaign depends on your own headlines and descriptions; testing method in the piece on RSA ad A/B testing.
6. Review after two or three weeks. Look beyond CTR and CPA: count active assets and check Ad Strength. If asset volume collapsed and performance followed, your guidelines are over-tuned.

What text guidelines do not do

They are not policy review. Guidelines filter style and vocabulary; they do not guarantee an ad clears Google’s policies. Category restrictions remain a separate matter.

Messaging restrictions are not hard blocks. They steer the model. You still verify the output manually in the asset report — this is not “set and forget”.

They do not cover other languages automatically. Term exclusions are language-bound. A multilingual account needs the list rebuilt per campaign language.

They do not touch images or video. These are text rules. The visual side is governed elsewhere — see the breakdown of Asset Studio and creative generation.

They do not replace decent signals. Guidelines decide how automation speaks, not what the campaign is about. Topical boundaries come from search themes in Performance Max and from your feed structure.

Common mistakes

Switching on the maximum set at once. Twenty-five exclusions and forty restrictions on day one is a reliable way to empty the generated asset pool and lose impression volume.

Writing guidelines like a copywriting brief. A paragraph with three conditions and a caveat performs worse than three unambiguous lines.

Not capturing the before state. Without the asset export you will never know what you lost, nor be able to re-add a winning message as your own asset.

Expecting a CTR lift from guidelines alone. This is a risk and tone management tool, not a performance lever. The return is fewer embarrassing lines, less legal exposure and a more predictable brand; the metric impact is usually neutral.

How guidelines interact with your own assets

Text guidelines are usually discussed as “a limiter on the AI”, but in practice their effect depends on what you already have of your own. The logic: the campaign has to fill every format slot. If generation offers fewer options after guidelines, the slot is filled either by your asset or by nothing.

Three consequences visible on live accounts:

• Strict guidelines plus a weak asset set = lost volume. The most common way to damage a campaign with good intentions.
• Strict guidelines plus a strong asset set = exactly what you wanted. Generation supplements rather than sets the tone.
• Loose guidelines plus a strong set is also a valid strategy. With twelve solid headlines, generation plays a small role and strict rules barely change the picture.

The prioritisation lesson: if your campaign has three of your own headlines and ten generated ones, the fix is not guidelines — it is writing your own assets. Guidelines are a tool for accounts that already have a base.

How to explain this to a client or a manager

The framing that usually settles the discussion: “We are not turning automation off, and we are not writing every ad by hand. We are stating in words what must never appear in our ads, and automation works inside those boundaries.”

What to set expectations about up front:

• Guidelines reduce the risk of bad phrasing but do not eliminate it — some of them act as a steer to the model rather than a hard block.
• Performance metrics do not rise because of guidelines; the gain is predictability and reduced legal exposure.
• On activation, some generated assets that were already serving will stop — expected behaviour, not a bug.
• The tool needs a monthly review. Without it, guidelines either go stale or quietly shrink the asset pool.

How this fits the 2026 pattern

Text guidelines belong to the same shift: Google removes setting-level manual control and hands it back at the level of data and constraints. The same pattern includes migrating Display campaigns into Demand Gen and local inventory ads being on by default in Shopping. The levers did not disappear — they moved: out of checkboxes and into the feed, the audience signals, and the rules you use to describe what is acceptable.

If you need a stable account and setup help for this kind of configuration, see Google Ads agency account rental and the PPC Rebels services.

Starter guideline sets

Three compact sets to start from. Do not copy them wholesale: take what applies to your category and pilot it on one campaign.

Ecommerce

• Term exclusions: “cheapest”, “free” (when no free option exists), “money-back guarantee”, “wholesale”, competitor brand names.
• Messaging restrictions: do not state prices or discounts unless they come from the feed; do not promise delivery timeframes; do not use superlatives; do not manufacture urgency without a real promotion.

Services and B2B

• Term exclusions: “guaranteed results”, “overnight”, “fully hands-off”, “best in the market”.
• Messaging restrictions: do not promise specific growth figures; do not name clients; keep a formal register; avoid question-style headlines.

Regulated categories

• Term exclusions: “cure”, “approval guaranteed”, “no checks”, “returns”, “100%”.
• Messaging restrictions: do not assert medical or financial outcomes; do not compare against competitors; do not reference age or social groups; do not use emotional pressure.

Notice the pattern: almost every useful guideline is either a stop-word from a legal checklist or a ban on promising something you cannot verify. That is the tool’s real practical value.

Staying inside the 25 and 40 limits

The limits feel generous until you start listing variants. Three ways to stay within them:

• Exclude roots, not every inflection. One well-chosen stem beats five word forms — but verify how it behaves across a few generations.
• Know which problems need a stop-word and which need a restriction. Ten superlative variants are often replaced by one clear messaging restriction.
• Do not duplicate Google’s policies. Guidelines exist for your requirements, not for repeating what the platform already blocks.

A monthly review process

1. Export the month’s automatically created assets.
2. Read 20–30 new lines with your own eyes — ten minutes, and the only honest quality control available.
3. List what does not fit, then decide: new guideline, or a missing asset of your own.
4. Check active asset count and Ad Strength — make sure the constraints have not narrowed past usefulness.
5. Remove guidelines that never mattered: a shorter list means fewer accidental removals.

The cycle takes under an hour a month and removes the tool’s main risk — quietly shrinking the asset pool until the campaign has nothing left to fill its formats with.

Who needs this and who does not

Worth it for: brands with a documented tone of voice; companies in regulated categories; agencies accountable to a client for wording; accounts where text customization is already on and generating heavily.

Probably not for: advertisers who do not use text customization at all; small accounts running three to five of their own ads where generation barely participates; and anyone expecting guidelines to lift CTR — wrong tool for that job.

FAQ: text guidelines in Performance Max and AI Max

Where are text guidelines configured?

At campaign level: in Performance Max under brand settings and asset optimisation, and in search campaigns wherever AI Max and text customization live. Text customization must be enabled first.

How many guidelines can I set?

Up to 25 term exclusions and up to 40 messaging restrictions per campaign. Start with far fewer than that.

Are term exclusions case-sensitive?

No, case does not matter. Language does: a guideline applies in the language you wrote it in and does not automatically carry over to other campaign languages.

What happens to assets already serving?

Google-generated text assets that do not adhere to the new guidelines stop serving and are marked as removed. Export the asset report before switching guidelines on.

Are messaging restrictions a hard ban?

No — they are instructions to the model rather than absolute filters. Term exclusions are enforced more strictly because they are checked literally, word by word.

Can I use guidelines to stop competitor mentions?

In the copy, yes — via term exclusions containing brand names. Serving on competitors’ brand queries is governed by other tools: brand lists and brand exclusions.

Do guidelines affect my own headlines and descriptions?

Guidelines are aimed at automatically created assets. Your own assets stay under your control — and they are what should fill the slots if generation gets more conservative after the rules.

How do I verify the guidelines are working?

Through the asset report: check which generated assets are active now, which are marked removed, and read the new phrasings manually for the first few weeks. Do not expect a “rule triggered N times” report.

Should I use guidelines if I do not want AI copy at all?

No. If you do not want generated text, switch text customization off and run on your own assets. Guidelines are for advertisers who want to keep generation, but inside boundaries.

How do guidelines behave in multilingual accounts?

They need duplicating. Because term exclusions are language-dependent, build a stop-word list for every language your campaigns run in.

Will guidelines increase conversions?

Do not count on a direct effect. The payoff is brand predictability and reduced risk. Any metric improvement tends to be indirect — from copy matching the landing page and user expectations more closely.

What if impression volume drops after enabling guidelines?

Cut the set back to the minimum needed, starting with the broadest messaging restrictions, and add strong assets of your own so the campaign has something to fill each format with.