Microsoft Advertising’s October roundup, posted October 28, buried the most useful item for brand and compliance people in third place. Messaging guidelines, a way to give Microsoft’s AI up to 40 written rules about what its generated ad copy should and should not say, moved into open pilot for Performance Max campaigns and for Search campaigns with AI Max switched on. Microsoft’s example is a luxury retailer writing “Never imply that products are low cost.” Four weeks before Black Friday, every advertiser letting a machine write headlines has a list like that somewhere. Now there is a box to put it in.
This is a docs-based review. I have not had the pilot in an account, written a guideline or watched what the AI did with one. Everything below comes from Microsoft’s October 28 post, its August AI Max launch post, the Microsoft Learn developer guide for AI Max (updated October 27), the Performance Max help page, and Google’s help page for its own version of the feature, checked against trade coverage from PPC Land, Search Engine Watch and PPC News Feed. Where the docs don’t answer something, I say so.
My read: switch it on wherever you run AI written copy on Microsoft, because a written brief beats no brief. But treat it as a brief to a copywriter, not a legal review. Microsoft’s own words are that it “helps reduce the risk,” and its documentation is much thinner than Google’s on the questions that decide whether you can trust it in a holiday week.
What it is
Messaging guidelines sit inside what Microsoft now calls text guardrails, a campaign level setting that shapes the copy its AI writes for you. There are two parts. Term exclusions are the old, blunt part: exact words or phrases that should never appear in generated copy, up to 25 per campaign and up to 30 characters each, according to the AI Max developer guide. Messaging guidelines are the new part: plain sentences, up to 40 per campaign and up to 300 characters each, describing claims, themes, tone or language to avoid, or wording and themes you want.
The difference matters more than it sounds. Excluding the word “cheap” does nothing about “priced for every budget.” A sentence that says never imply the products are low cost is aimed at the idea, not the word. As Search Engine Watch put it, an ad can suggest a bargain without using that word.
The copy these rules steer comes from two places. In Performance Max, Microsoft recommends headlines and descriptions during setup and, with autogenerated text switched on, adds text assets of its own. In Search, the same job is done by text customization, one of the three AI Max features. The August AI Max post says the AI “uses your existing assets and website content to generate and test additional messaging variations” and picks combinations at auction time.
What changed
Until this autumn, Microsoft’s only control over generated text was the exclusion list. The August post promised “prompt-based messaging guidelines” as coming soon. On October 11, PPC News Feed spotted them on Microsoft’s help pages as a closed pilot. On October 28 Microsoft announced the open pilot, and the screenshot in its post shows the guidelines being entered in the campaign creation flow.
The timing is not an accident. AI Max for Search became available in late August, and the August post says it is switched on by default for new Search campaigns. If you already used auto-generated assets in responsive search ads, that setting now shows up inside AI Max as text customization. In other words, a lot of accounts have a machine writing copy whether or not anyone made that call recently. Guidelines are the first control that lets the brand team, rather than the person holding the exclusion list, talk to it.
What works
The scope is right. Guidelines are aimed at the things that actually get brands in trouble: claims, implied pricing, tone, how the brand name is written. Microsoft’s own advice is to use them “primarily” for “broader rules around claims, themes, concepts, tone, or language to avoid,” and the screenshot in its post, as Search Engine Watch describes it, shows separate rules for descriptive language, headline capitalization and how to write the brand name. That is the list most review rounds already produce by hand.
It takes positive guidance too. You can ask for preferred wording, themes or a tone of voice, not only bans. That makes it useful for the boring wins, like always writing a product name the same way or keeping a seasonal theme across every generated line.
The numbers match Google. Google’s text guidelines allow 25 term exclusions and 40 messaging restrictions of up to 300 characters per campaign, the same shape as Microsoft’s. If you already wrote a rule set for Google’s AI Max or Performance Max, you have a first draft for Microsoft. You will still have to enter it, because nothing I found in Microsoft’s docs says Google Import carries guidelines across.
There is a bulk route. The developer guide says campaign level messaging restrictions can be set through the Bulk Campaign record now. For agencies running dozens of accounts, that is the difference between a policy and a wish.
And there is a way to check the output. The August post says AI generated assets show up next to yours in the asset report, and the developer guide describes a search term landing page report that maps searches to the delivered headline and the landing page, with rows that got no clicks left out. Microsoft says search term landing page reporting rolls out to advertisers starting mid November.
What breaks or is missing
Steering is not a guarantee. Microsoft says guidelines help “reduce the risk” of off brand copy and “reduce the time spent reviewing and editing generated copy.” It does not say generated copy will always follow them, and publishes no figure for how often it doesn’t. Microsoft’s September guidance on AI generated ads, as Search Engine Land reported, keeps the advertiser responsible for what runs. So the guideline field changes who writes the first draft. It does not change who signs the ad.
The docs don’t say what happens to copy that already exists. This is the big one. Google’s help page is explicit: guidelines also apply to existing generated assets, and anything that breaks them “will stop serving and be marked as removed.” Google also warns that “inefficient or counterintuitive guidelines may cause removal of a large number of good” assets and a drop in performance. Microsoft’s post and help pages say nothing either way. If its version works like Google’s, adding 40 rules to a live campaign in the week before Black Friday could quietly thin out the copy that is converting. If it doesn’t, your old off brand lines keep running. Either answer changes how you roll it out, and you currently have to ask.
Conflicts and languages are undocumented too. Google tells you restrictions apply across languages while exclusions don’t, that its model “can only take hard requirements” so “if possible” phrasing fails, and that one rule per guideline works best. Microsoft gives one example sentence and no rules of use. PPC Land also notes the post says nothing about conflicting guidelines, and doesn’t confirm whether Search campaigns need text customization switched on for guidelines to apply, which is a requirement on Google.
Order of operations matters, and the docs are behind. The Performance Max help page, last updated in July, still calls the setting text guidelines, mentions only term exclusions, and warns you to enter exclusions before the Final URL or “our asset recommendations won’t reflect your exclusions.” It is reasonable to assume guidelines behave the same way, but nothing says so. Enter them first, every time.
The API is half built. Bulk works; the main Campaign Management API support for messaging restrictions is “coming soon,” per the developer guide. Teams that build campaigns through their own tooling can set exclusions in code but not guidelines yet. And the vocabulary drifts: the blog says messaging guidelines, the developer guide says messaging restrictions, the help page still says text guidelines. Expect your team to search for the wrong thing.
Finally, deleting a bad line is permanent. The Performance Max help page says a deleted autogenerated asset “cannot be restored,” with reports kept for 30 days. Pull the off brand ones, but pull them on purpose.
Who it’s for
Anyone running Performance Max on Microsoft, or Search with AI Max on, where generated text is serving. That is most new Search campaigns, given the default. It matters most in categories where implied claims carry real cost: regulated products, finance, health adjacent retail, premium brands that live on never looking discounted, and retailers whose holiday offers change daily so stale pricing language is a real risk.
It matters less if you have generated text switched off and write every asset yourself. In that case the exclusion list was already doing nothing, and so will this.
Agencies get the most from it, because one written rule set can travel across clients through bulk. In-house teams with a legal reviewer get a cleaner first draft, not a reason to skip the review.
Pricing and limits
Microsoft’s post mentions no fee; this is a campaign setting, and you pay for media the way you already do. The limits are 40 guidelines of up to 300 characters each, plus 25 term exclusions of up to 30 characters each, per campaign. Scope is Performance Max and AI Max Search campaigns. Status is open pilot with no published end date, no published eligibility rules and no market or language list. In the same post Microsoft tells advertisers who want its other open pilot, audience exclusions, to contact their account team or Support. If the guidelines field isn’t in your campaign setup, start there.
What to do this week
Start with the corrections you already make. Pull the last three months of edits your team or legal made to generated or agency copy and turn each recurring fix into one sentence. One rule per guideline, written as a hard requirement, with an example, following Google’s advice since Microsoft offers none. “Never imply our products are discounted, for example ‘great value’ or ‘save big'” beats “keep it premium.”
Write the dull rules too: brand name styling, capitalization, product names, a ban on specific prices if your prices move during the sale. Those are the lines that cost you the most embarrassment per word.
Test before you touch live holiday campaigns. Add the set to one campaign that is not carrying Black Friday spend, check the asset report daily for a week, and count what disappears as well as what changes. If generated assets drop out, you have your answer about retroactivity.
Ask your Microsoft rep four questions in writing: do guidelines apply to existing generated assets, and are breaking ones removed; must text customization be on in Search; how are conflicting guidelines resolved; and do guidelines carry through Google Import. Keep the answers with the campaign.
Enter guidelines and exclusions before the Final URL in every new build, and keep one human on final review for anything with a price, a claim or a comparison through November 30. If you’re also running ChatGPT ads, the same discipline applies to its default-on rewrites. The machines now take a brief. Nobody has promised they read it.
