> ## Content Index
> Fetch the complete content index at: https://thebettermarketerproject.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# The Worst Thing You Can Give AI Is a Vague Brief and Your Credit Card
- URL: https://thebettermarketerproject.com/ai-campaign-brief/
- Published: 2026-10-15T10:01:45.000Z
- Updated: 2026-10-15T10:01:45.000Z
- Description: Google’s testing campaign creation from a plain-English goal. Meta has removed manual placement controls from several objectives. If your whole strategy fits inside “more sales,” please step away from the launch button.
- Author: Janelle Cardozo
- Tags: AI, Tools, Growth

*Google’s testing campaign creation from a plain-English goal. Meta has removed manual placement controls from several objectives. If your whole strategy fits inside “more sales,” please step away from the launch button.*

---

The brief says we need more sales.

That sentence has already survived a kickoff call, two Slack threads and a meeting where everybody nodded. Nobody’s defined which customers, at what margin, within what payback period, or which products the company can actually ship. Nobody’s said what counts as a sale worth paying for.

And now somebody is about to paste it into an ad platform attached to a credit card.

EXCELLENT.

On 12 October, [Google Ads was spotted testing a new campaign creation screen](https://searchengineland.com/google-ads-tests-ai-powered-campaign-creation-using-natural-language-495256?ref=thebettermarketerproject.com). Instead of asking you to pick a campaign type and grind through the settings, it opens with one question: what’s your goal for this campaign? You describe it in your own words and Google’s AI takes it from there.

The same day, [Meta made news going the other way on the interface](https://www.socialmediatoday.com/news/meta-offers-updated-ad-placement-control-guidance/832791/?ref=thebettermarketerproject.com), taking away some of the choices advertisers used to make by hand. Google handed everyone a text box, Meta quietly removed some checkboxes, and it sounds like two different stories right up until you notice they’re walking to the same place. Ad platforms are slowly turning campaign setup into a briefing exercise.

Which is easier, and it genuinely IS easier. It also means the quality of your campaign now depends less on how well you know the dashboard and more on whether your business has managed to finish a thought.

## First, what actually happened, because the headlines are running ahead of the products

Search Engine Land reported the Google test, and Google Ads Liaison Ginny Marvin confirmed it’s an experiment they’re gathering feedback on. Google hasn’t said how widely it’s available, which campaign types it supports, or whether it’ll stick around. The screenshot even includes a “create campaign manually” option. So no, Google hasn’t quietly handed every ad account to Gemini overnight.

It didn’t come out of nowhere either. Google introduced a conversational way to build Search campaigns back in 2024, one that suggests keywords, headlines, descriptions, images and sitelinks for you to review before anything goes live. Then came AI Brief for AI Max, where you explain your business, audience and messaging rules in ordinary language, and Google’s own example includes something as specific as “never mention prices.” In September 2026 they said the closed beta had expanded into seven more languages. The new screen is an experiment. The shift behind it already happened.

Meta’s getting to the same destination down a different road. It removed manual placement controls for campaigns using the engagement, leads, app promotion and sales objectives, while awareness and traffic kept theirs at the time of reporting. Advertisers haven’t lost every way to influence delivery, because you can still customise media per placement, use placement value rules, and apply certain exclusions and partnership controls. The accurate version is duller than “Meta took away all control,” which is probably why you saw the dramatic version first.

Put the two updates side by side though and the direction is obvious. Marketers define the outcome and the boundaries. The platform makes more of the execution calls.

## The bank account part is almost literal

In 2025, Mark Zuckerberg described Meta’s advertising endgame at Stripe Sessions, and his version went roughly like this: a business tells Meta what result it wants, says what that result is worth, connects its bank account, and Meta handles creative, targeting, measurement and delivery.

That’s a description of where they’re heading rather than a button you can press today. It does explain why individual controls keep getting less central to the product.

And honestly? I get the appeal. A local business owner doesn’t dream of becoming excellent at campaign architecture. A founder shouldn’t need seventeen browser tabs and a minor identity crisis to promote a product. If the platforms can delete repetitive setup work, lovely, delete it. I’m not emotionally attached to dropdown menus.

My problem is narrower. A friendly text box makes a very expensive decision look casual.

## Congratulations, the CPL is down. Sales still hates the leads.

Tell an ad platform you want more leads and it’ll get extremely good at finding people who complete forms. Those people won’t necessarily have the authority, the budget, the need, or the faintest intention of buying anything from you. Ask for purchases instead and it might find you an audience that converts beautifully, as long as the discount is big enough to eat your margin. Ask it to lower acquisition cost and it’ll quite reasonably chase whichever conversion you made easiest to count.

Nothing has to malfunction here. The machine can follow your instruction perfectly and still lose you the quarter.

Google’s own guidance says as much. Its documentation tells you to use qualified and converted leads and to import later CRM outcomes so automated bidding can learn which leads turn into money. Its Performance Max guidance for lead gen pushes you toward deeper-funnel goals, lead scores and predicted lifetime value, and Google supports value-based bidding on signals like revenue, profit, offline conversion value and customer lifetime value. So this isn’t me arguing against automation. I’m basically reading you the product manual.

If you tell the system a completed form is the win, it will go get you completed forms. When the real win happens three weeks later inside your CRM, that information has to travel back, otherwise the algorithm is learning from a half-finished story.

“More leads” isn’t an objective, it’s an unfinished sentence. More leads from whom, worth how much, qualified by what? And what happens on the day your dashboard says the campaign is thriving while your sales team quietly bins half of them?

## Prompt engineering is the least interesting part of this

The prompt will get all the attention because the prompt is the visible bit. People will publish templates, somebody will sell a course, and we’ll all be told the future belongs to marketers who know the perfect combination of verbs and punctuation.

Please.

You can write “get more leads” in iambic pentameter and you still haven’t defined a qualified lead. The hard part happens before the cursor starts blinking: which customer is worth acquiring, which action creates commercial value, what you can truthfully promise, what the business will actually pay, and which data source gets the final vote.

A well-written prompt can’t rescue a bad assumption. It just communicates the bad assumption more clearly.

The prompt is the last line of the brief. The brief is everything your company had to decide first.

## Seven things I’d answer before letting AI spend anything

This is an internal decision sheet rather than a claim that you can paste every answer straight into Google’s current experiment. The test’s later steps haven’t been disclosed. But before anybody types a goal into an AI-assisted builder, I’d want these answers somewhere marketing, sales and whoever owns the budget can all see them.

**1\. What business result are we actually buying?** Name the thing that creates value outside the ad dashboard. For ecommerce that might be a profitable first purchase from a new customer, or a repeat purchase within 60 days, or a sale that’s still standing after the refund window. For B2B it’s probably a sales-qualified opportunity rather than a downloaded guide. “Conversions” is a menu option, not a business result, at least not until you say what converted into what.

**2\. Who’s allowed to count?** Describe the customer you want and the ones you don’t: industry, company size, geography, use case, buying authority, minimum order value, whatever eligibility rules apply. If 400 university students download your enterprise software guide, you’ve technically generated leads, and also a small administrative crisis.

**3\. What’s a good result worth?** A target cost per lead means almost nothing without a qualification rate. A cheap lead that never becomes an opportunity is expensive, and a pricey customer who stays five years might be a bargain. Give the platform a value wherever the product lets you, and give your team the logic behind it, because the system can optimise toward your number but it can’t tell you whether your unit economics make sense.

**4\. What true thing are we allowed to say?** Write down the offer, the strongest claim you can credibly make, and the evidence holding it up: a case study, product data, customer research, a guarantee, a comparison you’d defend in a meeting. AI will produce fifty versions of a claim before lunch. It still needs one true thing worth saying.

**5\. What must never happen?** List the boundaries before the campaign finds them for you. Prohibited claims, sensitive audiences, excluded locations, unsuitable contexts, existing customers who shouldn’t be getting the acquisition offer, legal requirements, creative rules. “Use your judgment” is not a brand safety policy, and this only matters more as placement, targeting and asset creation get automated. If a boundary matters, write it down.

**6\. Which system gets the final vote?** The ad platform reports what it can see and attribute. Your business has to decide where the truth actually lives, whether that’s the CRM, paid subscriptions after the trial ends, revenue after returns, or closed deals in the finance system. Even a gorgeous ROAS doesn’t tell you whether the advertising created those customers, which is why Google itself separates attributed conversions from causal ones through Conversion Lift. The dashboard is evidence. It shouldn’t be grading its own homework.

**7\. What makes us stop?** Set the point where a human reviews, pauses or changes something: a spend threshold with no qualified opportunities, a sudden drop in lead quality, an unsupported claim turning up in generated creative, a placement pattern worth investigating. “Let the algorithm learn” is reasonable for a defined period. It’s not an open tab at the bar.

## What the difference actually looks like

Say you market an AI expense management platform. Here’s the vague version:

> Create a campaign that gets more leads from US finance teams. Keep the tone professional and focus on saving time.

Sounds completely reasonable. It’s also carrying a suspicious amount of business ambiguity in two sentences.

Now this one:

> **Business result:** CRM-qualified demo opportunities.  
> **Customer:** Finance leaders at US companies with 50 to 500 employees processing at least 1,000 transactions a month.  
> **Problem:** Their teams lose hours reviewing expenses manually and keep missing the close.  
> **Optimisation signal:** A lead only counts once the CRM confirms company fit, transaction volume and buying authority. Completed forms don’t count.  
> **Offer and proof:** Lead with “close the books faster” only where the existing customer case study supports it.  
> **Exclusions:** Students, job seekers, unsupported industries, existing customers.  
> **Economics:** Cost per qualified opportunity below \[target\].  
> **Review rule:** Review after \[spend or conversion threshold\] if fewer than \[minimum percentage\] of submitted leads qualify.

The second one isn’t better because it’s longer. It’s better because it tells the humans what has to be true before anyone’s allowed to call the campaign a success.

Those brackets are doing real work, by the way. If your team can’t fill them in, you’ve just found the meeting you need to have before any money moves.

### Fewer controls make the brief more important

Automation will probably run parts of your campaigns better than you can. It processes more signals, tests more combinations and reacts faster than someone checking performance twice a day between meetings. Use it.

The trouble starts when execution gets automated before the business has agreed on the strategy. The ad platform knows its own auction better than you ever will. It also wasn’t on the call where your perfect lead admitted they’ve got no budget until next year, it doesn’t know your highest-volume product has your worst margin, and it won’t stop a campaign because sales can’t deliver what the generated ad just promised. That knowledge has to get into the system somehow, and right now the only route in is you.

So as the interfaces get simpler, the valuable marketer becomes the one who notices sales is rejecting every cheap lead and goes and fixes the feedback loop. The one who knows which claim the evidence can actually hold up. The one who can tell a platform conversion from profitable growth.

Clicking the right checkbox used to look like expertise, and soon there’ll be fewer checkboxes to click.

Good. Now we find out who can write a brief.

If Google asks what your campaign goal is, don’t give it a slogan. Give it a decision. Because hand AI a vague brief and your credit card, and it will spend your money with absolutely magnificent confidence.

*If your team has ever celebrated a low CPL while sales quietly rejected half the leads, send this to whoever owns the CRM integration. And tell me: what does your ad platform count as a win that your business absolutely does not?*