The Briefing
Google Ads starts force-upgrading Search campaigns to AI Max on September 1. If you run Automatically Created Assets or campaign-level broad match, your campaigns get migrated over the month of September. Search term matching and text customisation come on by default. Google already blocked new setups of the old settings on August 3, so there is no opting out by standing still. Check your campaigns before the 1st or the machine picks your settings for you.
Meta made pixel and Conversions API setup a one-click job. AI-guided install, no developer, no maintenance. That sounds small. It is not. Tracking is the single most common silent failure in a Meta account, and I have the receipts on that below. Meta also opened its Ads Manager AI assistant to every advertiser and agency worldwide, and Advantage+ is now the default for Sales, App and Leads campaigns.
EU AI transparency rules are live. Article 50 of the EU AI Act is now in force, and AI content has to be labelled and machine-detectable. Anthropic has already shipped machine-readable watermarks to comply. If you sell into Europe, your AI-generated ad creative and voiceovers now need disclosure. Build the habit now rather than after a complaint.
ChatGPT is getting local search teeth. OpenAI and Yelp signed an official partnership to power local results inside ChatGPT. If you run anything with a physical footprint or a service area, your Yelp listing just became an AI ranking factor. Go look at yours today.
The Build
This week I rebuilt a Meta ad account from the ground up.
Then I went hunting for creative. I pulled about 200 video clips out of an event archive, ran them through an AI transcription tool, and catalogued every single one by what it actually says. Not by title. By claim. The good ones went into teaching ad sets, seven ads each.
Separately I wrote 30 teaching emails straight from the event transcripts. Then rewrote them three times. The first two passes sounded like a robot with a thesaurus. The third pass finally hit a 4th-grade reading level and passed my ad copy formula.
The result was real. August is the best month this account has ever had. 19.4 days in, pacing to add another 20%, on $355 a day. July finished as the best month ever. August should chase it down.
But the moment of the week was not the win. It was Monday, when I typed this to my AI:
Why do you keep changing your approach? You have full access to the ad account, the CRM, the assessment data, plus the booking calendar. You know the numbers. Yet you repeatedly fail and reinvent your approach. Stop looking for the data to back just your thinking. Pull ALL the data and then create from there.
Here is why I wrote it. Three separate errors surfaced this week. A human caught all three by asking one skeptical question.
Error one. A divide-by-100 bug reported one product as a $2,904 line item. It's a $95,850 line item across 42 buyers. The table looked perfect. It was off by over $600.
Error two. A double-counted conversion pixel had manufactured the entire conclusion that one offer had hit its ceiling. Luckily, I had been building strategy on top of my own conclusions. The ceiling did not exist.
Error three. 11 of 13 newly built ads shipped with no pixel in their tracking spec. I only found it because I asked "did you add website tracking?" That one question triggered an audit of all 1,012 ads in the account.
So here is the prompt I now run before I trust any number my AI hands me:
Before you give me the summary, do this: 1. Pull ALL the data for this date range, not the slice that supports the point you were about to make. Show me the row count and the date range you actually queried. 2. For every number in your table, show me the raw source figure and the maths you did to get there. Units and currency included. 3. Sanity check each total against a second source (the platform UI, the payment processor, the CRM). Flag any gap bigger than 2%. 4. List every assumption you made to produce this. Then tell me which one, if wrong, would change my decision. 5. Tell me what data you could NOT get, and what that makes uncertain. Do not give me a conclusion until steps 1 to 5 are done.
What worked: Almost every gain this week came from subtraction, not addition. I killed one ad angle after it burned $237 for a single book. I stripped a teaching ad that had never sold anything. And I found that the awareness layer was buying traffic with no link to anything. Money was being spent driving people to a dead end. Fixing that cost nothing.
What didn't work: I spent real hours on creative volume when the constraint was never creative volume. It was structure and tracking, upstream of the ads. The second thing that did not work was my own trust. A confident wrong answer and a confident right answer look identical on screen. Every real correction this week started with a human (me) refusing to accept a clean-looking table.
The Play
If you are a business owner in an adding season when you are actually in a subtracting one, this is for you. More content. More ads. More clips. Another tool. And the numbers do not move, because the leak is upstream. A broken link. A page that goes nowhere. A spend line nobody culled.
And more pointedly, this is for the owner who has handed real decisions to AI and is quietly letting it mark its own homework.
Here is the play. Run a subtraction audit before you build one more asset. Take your ad account, your funnel, your email list, whatever is spending money. Then ask three questions in order.
One: does every paid click land somewhere that can take money? Click every destination yourself. Not the AI. You. Two: is tracking actually firing on every ad, or just on the ones you set up first? Three: what is the single worst-performing spend line, and why is it still on?
Copy this into your AI to start:
You are auditing my paid account for leaks, not for ideas. Do not suggest anything new. List, in a table: - Every active ad, its destination URL, and whether tracking is present in its spec. Mark missing tracking in ALL CAPS. - Every destination page and whether it contains a link to the thing I am selling. If it does not, say DEAD END. - Every spend line from the last 30 days, sorted by cost per conversion, worst first. Then tell me the three things I should turn OFF this week, and the money that frees up. Show me the source figures for every number. If you cannot verify something, say "unverified" instead of estimating.
Then, and this is the whole point, check its answer yourself. Open a few of the destination pages by hand. If the AI said they were fine and one of them is a dead end, you have just learned what the rest of the table is worth.
The skill that mattered this week was not prompting. It was knowing the business well enough to smell when a number was off, and saying so out loud. This is the human skill you must own.
Try this today. Open your ad account. Click the destination URL of your three highest-spend ads. If any of them cannot take money in two clicks, turn that ad off before you write another word of copy.
Download social-media-advertising-1.md
Growth Partner for founders stuck between $1M and $20M | I find the money you're sitting on and build the machine that collects it | $1B+ in client results.