ChatGPT Ads & Google AI Mode: Inputs Beat Keywords

Welcome back to the front row.

Google changed the keyboard shortcuts this week and forgot to send the memo. The operators who relearn them are about to eat very well.

There is a story going around that Google, Meta, and OpenAI are quietly taking your job. Cropping your images, picking your queries, trimming their own promises. It is a good story. It is also the story of someone who is about to lose to the person sitting next to them. Here is the other reading.

The keyword did not lose the auction. It clocked out.

Have you ever opened an account you have run for years and felt like a stranger in your own house. The furniture moved overnight. That was this week, for a lot of people.

Open OpenAI's advertiser documentation and read how a ChatGPT ad gets picked. No exact-match keyword. No phrase match to argue about at 2 a.m. The system reads the conversation, your landing page, your copy, and a few context hints you hand it. Those hints, in OpenAI's own words, are not keywords and do not guarantee delivery. (OpenAI, Ads in ChatGPT)

Sit with that. The one input you spent a decade mastering is not a lever on this platform. Your landing page is.

Now the sideline starts shouting. "Google is doing the same thing, and it just wants more of my money." Half true, and the wrong half. On Google, ads now show inside AI Mode on 29.45% of commercial queries, nearly one in three, from SE Ranking's test of 50,032 keywords. On the expensive stuff above $10 CPC it climbs to 53.56%. And only 11.53% of the advertisers sitting in those slots also appear as a cited source. Read that twice. Almost nine in ten ad slots go to a brand the model did not pick on merit.

The complaint writes itself: pay to play. Look one layer down and it turns into a different sentence. The slot is not being taken from you. It is being handed to you, in a room you could not walk into by ranking. That is not a tax. That is a door.

Then there is the number everyone is angry about. AI Max, they say, quietly walked its promise back from 14% to 7%. That is not what happened. Google's 7% is the full AI Max feature suite measured against matching alone, a narrow and honest comparison (Google Ads & Commerce). The 14% was a different test. Nobody downgraded anything. Someone read two numbers and picked the outrage. Independent accounts, meanwhile, land everywhere: SMEC logged 13% more revenue, while Monks watched 99% of AI Max terms drive zero conversions in one account. Both are true. That spread is not Google failing. That is your inputs deciding your outcome.

That is the pattern, said plainly. Three platforms, one fortnight, all reading the same things before they act. ChatGPT reads your page and your copy. Google's AI answer reads your relevance and your offer. Meta's model reads your creative and your signals. Not your bid ceiling. Not your match type. The inputs a machine cannot fake for you, page quality, creative, brand clarity, first-party data, are the whole game now.

Google did not take the craft. It automated the part most accounts were doing badly and handed back the part that compounds.

So here is your Monday, dry version. Pull the landing pages behind your top five campaigns and read each one the way a model does. First screen: does it say exactly what it sells and to whom. If a stranger cannot answer in three seconds, no bid strategy saves you. Next, open your Meta creative toggles and write down which enhancements are on, because if the platform is editing your ads you should know what it changed. Then run your brand through an AI answer for your category and see if the model says your name. In a world where the machine picks who gets mentioned, being unsayable is the new page two.

The people panicking about lost control are guarding levers that already left the building. The ones who win are upstream, making the inputs so clear the machine has no reason to choose anyone else.

Google handed you a shorter path to the work that was always the actual work. The 2016 crowd is still filing a complaint about it.

Prompt of the Week

Paste your landing page copy in, then run this to read your page the way a ranker does:

You are an ad-ranking model deciding whether to surface this landing page for a commercial query. Using only the text below, tell me: (1) in one sentence, what this page sells and to whom; (2) the exact query intent it best matches; (3) three reasons a ranking model might skip it; (4) the single edit that would most improve its clarity. Be blunt. Here is the page: [paste].

What this gives you: a cold read on your most important input, the page the machine actually judges, before you spend another riyal sending traffic to it.

This Week

Google AI Mode runs ads on nearly 1 in 3 commercial queries. SE Ranking's 50K-keyword test (via Search Engine Land) puts it at 29.45%, and 53.56% once CPCs pass $10. Paid is the entry ticket to answers you cannot rank your way into.

ChatGPT Ads is self-serve, CPC, no minimum. In about ten weeks OpenAI went from a $60 flat CPM with a $200,000 floor to a no-minimum CPC auction inside the OpenAI Ads Manager. Early blended CTR sits near 3.8 to 4%. It is testable this week.

DSA is officially becoming AI Max. Google confirmed Dynamic Search Ads upgrade to AI Max, with the 7% figure measured against matching alone, not a walk-back of the old 14%. Know what the number means before a rep or a rival tells you what it means.

Meta's Advantage+ creative enhancements keep switching themselves on. If you have not checked the toggles lately, the platform may be recutting your images and rewriting your copy by default. Audit before your next launch.

The Discourse

Digiday caught the whole fight in two quotes this week. One agency lead loved the transparency: "With AI Max we can jump in and review the exact queries that triggered or the assets that were generated." Another practitioner shrugged: "It's not delivering what it's supposed to deliver."

Here is the thing. Both are describing the same tool. One used the reporting to fix the inputs. The other waited for the machine to fix them for free. Feeling let down by AI Max for the first time is usually a good sign that your feed, your copy, and your landing pages were carrying more weight than you thought. The platform did not get worse. It got honest.

Tool of the Week

Semrush AI Search Visibility Checker (free, no sign-up). Enter a domain and it scans ChatGPT, Google AI Overviews, and Gemini to show whether your brand appears in AI answers, which prompts trigger a mention, which sources get cited, and where a competitor shows up and you do not.

Use it now: run your brand and your two closest rivals for your category's buying query. If the model names them and skips you, that is your input problem in one screenshot. It measures the exact thing the platforms reward, whether the machine can clearly say who you are.

My Take

I have spent years telling clients the account is the craft. The match types, the negatives, the bid rules. This week made that a harder sell, and I am fine with it.

Every platform is offering the same deal. Give us the outcome, we run the execution. I do not flinch at that, because I have watched too many sharp accounts lose with a beautiful bid strategy and a landing page that felt like a waiting room.

Here is the part the angry threads miss. When the machine takes the levers, skill does not die. It moves upstream, into the page, the offer, the creative, the clarity of who you are. That work is harder to fake and impossible to automate away from you.

Google did not come for your job. It came for your busywork. The marketers still cursing the update are polishing a steering wheel that stopped being connected to anything.

Optimize the inputs. That is the only auction left.

That's Issue #13 of NVM. If this gave you one thing to act on this week, share it with someone who needs it. Reply to this if you have a question. I read every one.

See you next week. Mostafa. NVM · Front row seats to AI and digital marketing

• Subscribe → nvm.marketing/newsletter
• Join the community → nvm.marketing/community
• Write for NVM → nvm.marketing/blog