AI Brief in Google Ads: How to Steer AI Max Before It Reaches Your Account

By Lee Moran, Founder of Veritas Digital

Google has added a natural language control layer to AI Max. It is a genuine improvement, and it is not the same thing as getting full manual control back. Here is what an AI Brief does, where its guarantees start and stop, and how to write one before it lands in your account.

If you run Google Ads, you have spent two years watching various controls quietly disappear. Match types became suggestions. Ad copy became assembled rather than written. Landing page selection became something the system decided for you. For most advertisers that trade was ok. For anyone in a regulated sector, it was closer to a slow-motion compliance problem.

AI Brief is Google’s answer. Rolling out for Search campaigns in English before extending to Shopping and Performance Max, it lets you describe your business, audience and brand rules in plain English. It arrives alongside text disclaimers, which solve a problem regulated advertisers have been stuck with for years to avoid compliance headaches.

I have not used AI Brief yet – it is not live in the accounts we manage at the time of writing, so this is a preparation guide rather than a hands-on review. What I have done is run AI Max across a number of client accounts. The honest summary: this is the most meaningful control Google has handed back since AI Max launched, and it is more limited than the announcement suggests. That gap matters enormously if the wrong sentence in your ad is a regulatory issue rather than a marketing one.

What AI Brief is steering

AI Max is not a campaign type – it is an optimisation layer inside an existing Search campaign, with three components you can toggle independently. If you have been following the DSA to AI Max migration, these are the levers that migration eventually hands over.

• Search term matching. Keywordless expansion beyond your keywords. The benefit is real, because search behaviour has moved towards longer conversational queries that keyword syntax struggles to capture.
• Text customisation. Generates headlines and descriptions from your ads, landing page copy and the query. It reliably improves click-through rate but can reduce conversion quality if left unconstrained.
• Final URL expansion. Routes each click to the page the algorithm judges most relevant. Often an improvement for a deep e-commerce catalogue; for a lead generation business with one carefully built landing page, it can undo months of conversion rate work. It only functions when text customisation is active, and dynamic URLs can produce 404s where tracking templates are not configured compatibly.

The brand results are strong: Ikea US used all three features and reported a 127% increase in physical store visits, a 65% lift in unbranded clicks and a 28% rise in incremental ROAS. ClickUp reported conversion value up 161% and ROAS up 25% within a month of adopting the same combination. Royal Canin and L’Oreal are the other names Google leads with, and in each case all three levers were switched on together.

Read Google’s own footnotes, though, and the picture gets more conditional. The headline claim of 14% more conversions at a similar CPA explicitly excludes retail advertisers, and it comes from campaigns where more than 70% of conversions were already coming from exact or phrase match. In other words, the advertisers who gained most were those with tightly built keyword accounts and genuinely untapped query volume – not those already running broad match. My personal read is that AI Max is not bad, but highly conditional. It rewards clean conversion data, adequate budget and disciplined negative keyword hygiene, and punishes accounts without them. It is also worth remembering that AI Max sits on top of your bidding, so the Smart Bidding changes landing on 17th August will shape what it can deliver. Which is exactly why a control layer is the right thing to be building.

What is the feature AI Brief?

A Gemini powered interface that lets you guide a campaign in natural language rather than through structured settings alone. Brandon Ervin, Director of Product Management at Google Ads, framed it as describing your business and brand guidelines in your own words, the way you would brief a partner. Three input categories:

• Messaging guidelines. What ads should and should not say. Google’s own example: “never mention prices or discount percentages”.
• Matching guidelines. Which intent to reach and which to avoid – a strategic layer sitting above negative keywords and brand exclusions.
• Audience guidelines. Who you are reaching and how messaging should be adapted for them.

There is also a live previewer. Once guidelines are submitted, AI Brief plays back its interpretation so you can audit how your direction translates into proposed copy, queries and audiences before anything serves. That is the most underrated part of the announcement as you get to check draft ads before anything goes live.

The distinction that matters most: guaranteed versus steered

This is the part I would want any client to understand before writing a line of brief. Google is explicit that two mechanisms are at work. Term exclusions function as negative keywords for creative – excluding competitor names or off-limits phrases is described as guaranteed to omit those terms across every generated asset. Strategic guidance works differently: when you prioritise a value proposition, the system honours the theme while leaving the AI room to contextualise in real time.

The gap between “guaranteed to omit” and “honoured with room to contextualise” is where risk lives. If you are a dental practice and need certainty that “pain free” never appears in an ad, that belongs in term exclusions, not in a guidance sentence about your gentle approach.

To be fair to Google, publishing that distinction at all is a more honest architecture than most automated systems offer. But AI Brief is not a compliance guarantee. It is a well-designed steering mechanism with one hard constraint lever inside it and knowing which lever you are pulling is the whole job.

Text disclaimers: the quiet win for regulated advertisers

Previously, if you needed specific legal wording in your ads you faced a choice: keep tight control of copy, or use final URL expansion. Not both, because with expansion active you could not pin assets or guarantee disclosures appeared.

Text disclaimers ensure mandatory text appears regardless of how the rest of the ad is generated, with a global multi-language rollout indicated. For finance, healthcare and insurance advertisers this genuinely changes the calculation. The risk it addresses is not theoretical: text customisation pulls from your website, including blog posts and secondary pages, so if your site mentions clinical results anywhere the AI can pull those claims into ad copy. There is nothing inherently preventing generated copy along the lines of “clinically proven to reduce symptoms by 40%” – a sentence no healthcare advertiser would knowingly approve.

One caveat though – nothing here is compliance advice. UK ad copy sits under the CAP Code and ASA scrutiny, with sector rules from bodies such as the GDC, GMC or FCA on top. Anything you write into an AI Brief should be reviewed by your compliance lead before it goes live and so should a sample of what the previewer generates back at you.

How to write an AI Brief that works

Google’s mental model is the usual one: richer prompt, better output. That is right as far as it goes, but a campaign brief is not a chat prompt. It is a control document. Four principles I’d personally recommend:

• Put hard rules in exclusions, not prose. If a phrase must never appear, list it in term exclusions where omission is guaranteed. Do not write “avoid clinical claims” in a guidance paragraph and assume you are covered.
• Be concrete about intent you do not want. The most expensive AI Max failures we see are about matching, not copy. Competitor terms, job seekers, students, people seeking free or NHS alternatives all convert badly and all are plausible semantic matches. Write them all done.
• Write for a reader with no commercial context. The system does not know your margins or capacity. If you only want self-pay patients, say so. If a treatment has a six-month waiting list, exclude it.
• Treat the previewer as a test. Push adversarial cases through it. See whether it will generate an outcome claim when a query invites one.

An AI Brief template

Adapt this to your business. It is deliberately generic and maps to Google’s three categories plus exclusions.

Business context. We are a [type of business] in [location], serving [customer type]. Our commercial priority is [service line]. We compete on [differentiator], not price. Our typical customer is [description] and comes to us because [trigger].

Messaging – always include. Always reflect [core value proposition]. Always use [terminology your brand or regulator requires]. Reference [credential, registration or accreditation] where relevant.

Messaging – never state or imply guaranteed outcomes, success rates or clinical results. Never mention prices, discounts or offers. Never use urgency or scarcity language. Never make comparative claims about named competitors.

Matching – intent to reach. Match users actively researching or ready to purchase [service]. Match longer question-based queries about [problem area] where our service is a genuine answer.

Matching – intent to avoid. Avoid job seekers and recruitment queries. Avoid free, NHS, charitable or DIY alternatives. Avoid academic and student research. Avoid competitor brand names. Avoid [services you do not offer or cannot fulfil].

Audience guidelines. For [segment A], emphasise [benefit]. For [segment B], emphasise [different benefit]. Maintain a [tone] tone – [what that means in practice].

Term exclusions (hard constraints). Every word and phrase that must never appear in a generated asset. Be exhaustive rather than elegant. This is the only category where omission is guaranteed.

What to do before it reaches your account

• Audit what your website says. Text customisation draws from your landing pages and secondary content, so any outcome, claim or statistic you would not put in an ad is already a live risk today.
• Build your exclusion list now and have compliance review it. That list becomes your term exclusions when the feature arrives.
• Check tracking templates against final URL expansion. Test before enabling, not after.
• Segment out the Search Partner Network and assess it separately.
• Review negative keyword lists. They remain the most reliable control available, respected at the account, campaign and ad group level. AI Brief adds as an additional layer – it does not replace it.
• If you have not tested AI Max at all, use campaign experiments rather than switching it on wholesale. Enable search term matching first, in isolation, for at least four weeks and measure performance changes from there.

One timing note: The September auto-upgrade window still applies to campaigns using automatically created assets and the campaign-level broad match setting, even though the Dynamic Search Ads migration was pushed back. Plenty of advertisers have filed AI Max under “delayed until 2027” and stopped reading. If either configuration is live in your account, that is worth correcting now.

The wider pattern is one I wrote about last month on agentic AI in PPC. The direction of travel is not less automation – it is better instructions given to automation, with a human operator who understands what those instructions do. The skill AI Brief rewards is the ability to articulate a business precisely enough that a machine can act on it without embarrassing you with rogue ad copy that breaches compliance or wasted spend in irrelevant search terms. That is closer to writing a scope for a contractor than to building a campaign.

References

• Think with Google – How AI Brief unlocks asset optimization at the scale of Search (interview with Brandon Ervin)
• Search Engine Land – Google AI Max gets new controls, Shopping rollout and travel consolidation
• Search Engine Land – Google AI Brief may be the replacement keywords never had, by Frederick Vallaeys, co-founder of Optmyzr and former Google AdWords Evangelist
• Search Engine Journal – Google AI Max Adds New Brief And Text Disclaimer Controls
• Search Engine Land – AI Max for Search: Everything you need to know
• Search Engine Land – Google appears to be testing new branded search controls in AI Max campaigns
• Think with Google – Why AI Max is the new growth engine for Search (Ikea US, Royal Canin)
• Think with Google – How brands use AI Max for Search campaigns (ClickUp)
• Google Ads Help – How AI Max for Search campaigns works
• Google Ads Developers – Get started with AI Max for Search campaigns