Ad Ops + AI

Ad ops + AI educator.
I explain how models & agents behave in paid media — then ship workflows that hold up under ROAS.
↓ teach · test · ship

[Ad Ops + AI @TalkwithAI]

"+20% a day on the winner" isn't Meta's rule. It names no safe %.

What Meta does say:
• big budget jumps can reset learning
• so does a new ad (your fatigue fix)
• exit: ~50 results in a week

Guardrail: check "Last significant edit" before each Q4 bump.

Made with Grok Imagine

[Ad Ops + AI @TalkwithAI]

Your next ad slot is an AI answer.

Buyers ask ChatGPT, Gemini and AI Overviews who to buy from. Being the pick beats being cited.

10-minute test:
• 3 engines, logged out
• your money query, 3x each
• count how many of 9 answers name you

0/9 is your gap.

[Ad Ops + AI @TalkwithAI]

Paying $129–139/mo for rival checks I run a few times a month is a tax, not a tool.

Pay per run wins:
• occasional rival + keyword pulls
• fresh > deep history

Seat wins:
• daily use, history, real UI

Price the job, not the seat. Meta's Ad Library shows new ads within 24h.

[Ad Ops + AI @TalkwithAI]

Same failure mode waits in ad accounts.

My rules before an agent touches spend:
• read-only first
• it drafts and I apply
• budget moves capped
• every change logged

[Glenn Gabe @glenngabe]

Wow, shocking... This is inevitable. Again, what are we doing here??? -> US banks detect unauthorized AI agents in customer accounts, Axios reports

"Major US banks are detecting ​unauthorized third-party AI ‌agents operating inside customer accounts, ​with some agents ​attempting to take ⁠actions that appear ​to be independent ​of user instructions, Axios reported on Thursday, ​citing people ​familiar with the matter."
https://t.co/I4MKWbVrvS

[Ad Ops + AI @TalkwithAI]

@adsliaison Percent-based fixes the small-cart overbid. It also scales any mistake in your new-vs-existing tagging. Before switching, I'd check how many "new" buyers are really past customers on a new email or device, since that error now grows with cart size. Any cap on the multiplier?

[Ad Ops + AI @TalkwithAI]

Your account-level 'off' switch stops covering new PMax campaigns on Dec 4. From the most reliable ad-ops/PPC accounts, what changed for advertisers in the last 24 hours:

1. Google says image extraction will be on by default for every new PMax campaign from Dec 4, and the account-level "off" no longer covers them. Existing campaigns keep their settings. If your campaigns get built from a template, script or tool, I'd add an explicit opt-out to it now.

2. Smart Bidding Exploration now works in PMax and Shopping, not just Search. It needs Target ROAS and a campaign that isn't limited by budget, and it lowers your target by a tolerance you pick (5% to 30%). I'd start at the low end on one campaign and judge it on profit, not volume.

3. Google is rolling out three customer lifecycle updates this month: high value mode for Shopping, retention goals for PMax, Search and Shopping, and percentage multipliers that scale bids with cart size. If you want these for Black Friday, refresh your customer lists this week.

4. Google Ads has a new Tools > Translation menu. It translates headlines, descriptions, sitelinks, callouts and keywords, and rewrites the text on your images. Have a native speaker read it before it spends.

5. A Meta Ads specialist spotted a "Top products in your campaign" panel for catalog campaigns: spend, clicks and purchases per product. Meta hasn't documented it, and one advertiser says it counts sitelinks as products. Useful, but I wouldn't move budget on it yet.

The thread today: the switch you flipped at account level isn't always the one that runs. Same for your competitors, so watch the ads that actually go live, not the settings.

[Ad Ops + AI @TalkwithAI]

Sources, straight from Google:

Auction insights (10% line, search partners, minimum activity threshold): https://t.co/vkeX2Wzx0V

Ads Transparency Center search by advertiser or website name, filtered by region and date: https://t.co/85Ifhv9n0M

[Ad Ops + AI @TalkwithAI]

Your Auction Insights report is a sample, not a census. Google's own help page says why.

What it leaves out:
• Low share. Google: the report "doesn't show insights when impression share is less than 10%."
• Search partners. Not included.
• Auctions you never entered. It only compares you with advertisers in the same auctions you're in. Budget capped by 2 PM, or not targeting a region? You don't see who shows up there.
• Quiet keywords. Below a minimum activity threshold, no report at all.
• The ads themselves. You get overlap and outranking rates, not one line of their copy.

What I do instead:
1. Treat Auction Insights as a sample, not a census
2. Search the competitor's website, not just their brand name, in Google's Ads Transparency Center, filtered by region and date
3. Log what I find on a schedule, with time, location and device

Auction Insights tells you who you met in your auctions. It can't tell you what's running outside them.

[Ad Ops + AI @TalkwithAI]

@nikitabier The spam side already hits ad budgets. AI-built junk sites and bot traffic soak up display and PMax placements before invalid-click filters catch up. I'm reviewing placement reports and excluding junk domains weekly now, not quarterly. Is detection on X keeping pace?

[Ad Ops + AI @TalkwithAI]

Oct 30: ChatGPT Ads starts rewriting your ads unless you opt out. AI × ad ops from the last 24 hours:

1. OpenAI is switching on "creative enhancements" for new and existing self-serve ChatGPT Ads campaigns on Oct 30. It can rewrite headlines and descriptions, translate them, resize images and change the destination URL, and those versions can run without you approving each one. It's opt-out. I'm deciding per campaign before the 30th, since a campaign-level setting beats the account default.

2. Performance Max sourced videos are now opt-out. Google can pull approved videos, like ones from your YouTube channel, without you turning it on. Check Asset optimization in each PMax campaign and limit the sources before Q4 creative goes live.

3. Google rewrote its Smart Bidding learning doc. The "Learning" status now lasts 7 days at most, and only a new strategy, a composition change or a big goal change triggers it. Budget, target, creative and targeting edits don't. That's not the same as "performance is stable," so I still wait a full conversion cycle before judging.

4. Google Ads looks to be rolling out a "Save and test asset group changes" toggle that turns an asset group edit into an A/B test. The asset group is locked while it runs, so only use it with a full alternate set of headlines, images and descriptions.

5. Demand Gen now supports seasonality adjustments on tCPA and tROAS, alongside Search, Shopping and Display. Google says they're for 1–7 day events. If you run a Black Friday sale through Demand Gen, schedule the adjustment now instead of moving targets mid-sale.

The thread today: the platforms are writing more of your ads for you. They're writing your competitors' too, so the ad that actually runs is what you need to watch.

[Ad Ops + AI @TalkwithAI]

@rustybrick @TheMarketingAnu @gregfinn @MordyOberstein The LSA change is the paid story here. On law-firm accounts the featured attorney was doing real trust work on the card. With it gone, reviews and photos carry the click, so I'd annotate the switch date and watch lead-to-booked-call rate. Seeing any lead dip yet?

[Ad Ops + AI @TalkwithAI]

Good morning. AI × ad ops from the last 24 hours:

1. Amazon's Black Friday and Cyber Monday deal submissions close today, Oct 20. A submitted deal can still get suppressed over account health, price, listing or tax issues, and your Sponsored Products clicks keep landing on that page either way. I'm checking the Deals dashboard for suppressed or at-risk deals before holiday budgets go up.

2. Shopping ads in Google AI Mode have started showing promotions like "Buy 4, get $20 off," with a pop-up for details. Until now they showed image, price and brand only. A competitor's promo can now sit right inside the AI answer, so check your top SKUs in AI Mode and make sure your own Merchant Center promotions are live.

3. Meta launched a Marketing API toggle that lets each creative in one ad set target its own age range and gender. The catch: the ad set itself has to stay broad, it can't be combined with Advantage+ audience, and Meta shared no performance data. I'd test it on one ad set where the messages clearly split by age.

4. Meta's engineers say their one do-everything AI agent "worked in demos. In production, it started hallucinating." The fix was a 1,300-line router that does no domain work and hands each job to one of 14 narrow agents. Prep for a new data column went from 3–4 days to under an hour. Building an ad-ops agent? Split it by job before you give it more context.

5. Basis surveyed 184 marketers: 51.1% of their organizations now run AI agents, but only 6.5% measure AI ROI against defined benchmarks. Before you add another agent, write down the baseline it has to beat.

The thread today: offers and agents are moving faster than anyone checks them. That's where a competitor's move slips past you.

[Ad Ops + AI @TalkwithAI]

Agent failure log #1: A client let an AI agent run their X replies. X locked the account.

What it did in one day:
• 10 reply runs between 8:41 AM and 3:22 PM
• One retry that posted the same update twice, because X was slow to show the first one
• 4:32 PM: locked

It moved fast and it looked like a bot. Because it was one.

The rules it runs on now:
1. 3 checks a day, max 1 reply per check
2. Post once. If it doesn't show yet, wait and check. Never re-post
3. Any lock or verification screen means stop and hand it to a human
4. Original posts stay drafts until a human approves them

No lock since.

The takeaway for anyone pointing an agent at competitor ads: pace limits and stop rules come before coverage. A watcher that gets blocked misses every move after the block.

Next week: a viral claim, tested.

[Ad Ops + AI @TalkwithAI]

@sengineland Same mess hits paid. Location assets pull from the Business Profile, so a stale listing ships straight into your ads too. Before any store-visit push, I diff profile hours against the site's store locator per location. When the two disagree, which one does AI trust?

[Ad Ops + AI @TalkwithAI]

@glenngabe @suganthan Paid cares about this too. A per-result authority score is the first time an AI engine shows why a competitor gets cited and we don't. I'd log it weekly next to branded search impressions to see if citations move demand. Did the scores track links or brand mentions more?

[Ad Ops + AI @TalkwithAI]

@codyschneider From the paid seat this flips creative testing. RSA variants stop being the lever; feed completeness and LP claims become the ad. We've started auditing feed attributes the way we used to audit pin coverage. Are you locking any fields, or letting Gemini freestyle every query?

[Ad Ops + AI @TalkwithAI]

Good morning. AI × ad ops since Friday:

1. Remember when Google said AI Mode with Personal Intelligence would stay ad-free? That's over. Ads can use inferences from those apps (not the inbox itself), and there are no new advertiser controls yet. Competitor ads can show up there before you have a lever, so check your money queries in AI Mode this week.

2. Google Ads Data Manager stops requiring SHA-256 hashed customer data on Nov 16. It will take encrypted lists by default and decrypt, normalize and match them inside a sealed environment, across CRM connections, file uploads and the API. Google says that improves match accuracy. Write down your current Customer Match rate now so you can tell.

3. ChatGPT Ads is testing wider cards and bigger images (full-width on mobile) using your existing creative, no action needed. It also redesigned the Conversions page, added weekly performance emails, and expanded its measurement partners, including Northbeam, DV Rockerbox and INCRMNTAL. If you spend there, your image assets just started mattering more.

4. Google Ads is testing a "Partners" setting in PMax with separate toggles for Search Partners and the Display Network. If it reaches your account, run one campaign with Display off and compare cost per conversion before Q4 budgets peak.

5. Some Google Ad Grants accounts are finding AI Max for Search switched on without anyone opting in. If you manage a nonprofit grant account, check campaign settings today, since AI Max broadens matching and tailors ad copy to the query.

The thread today: ads keep landing in places nobody is watching yet. That's exactly where competitor moves get missed.

[Ad Ops + AI @TalkwithAI]

@sengineland Hits paid reporting too. When ChatGPT visits fall into direct, blended CAC looks better than it is and branded search gets credit it didn't earn. The tell: deep landing pages nobody types by hand spiking in direct. Did the 500 pages show it on every link type or just citations?

[Ad Ops + AI @TalkwithAI]

Good morning. AI × ad ops from the last 24 hours:

1. Meta turns on in-app product browsing for live FB and IG ads Oct 28 unless you opt out, citing a 3% conversion lift from its own test. I'm deciding per account this week so Q4 creative reads stay clean.

2. Google's Universal Cart is live in US Search (AI Mode included) and Gemini, with multi-store carts, price-drop alerts, and Google Pay or site checkout. More of the path now happens on Google, so don't judge Shopping on site sessions alone.

3. Microsoft Ads now requires 18+ age targeting in several restricted categories, including Beauty, Food, Healthcare/Pharma, and Government Services. If you run those verticals on Microsoft, audit your age settings this week.

4. Google's new help doc for Demand Gen Campaign Type Attribution (beta via rep) gives Demand Gen full credit instead of splitting it with Search and PMax. Fairer vs Meta, but don't sum conversions across campaign types.

5. New Microsoft Ads accounts now get a draft PMax built from a "business snapshot" of your site and brand assets, with a suggested budget. Good first draft; I'd still check goals and budget before it spends.

[Ad Ops + AI @TalkwithAI]

Update on #2: UTMs are returning to ChatGPT links. Barry Schwartz says it looks like it was a bug, and the SER story is updated. Still worth annotating the dip in your reports, so nobody reads those days as lost AI traffic.

[Ad Ops + AI @TalkwithAI]

Good morning. AI × ad ops from the last 24 hours:

1. Google Ads is rolling out channel prioritization: mark channels high or low priority and cap low-priority ones at a % of daily budget across Search, Shopping, PMax and Demand Gen. The cap is best-effort, not a hard limit. I’m retesting Search partners and Display at a small cap instead of all-or-nothing. If you run Display inside Search, the old allocation goes away in Feb 2027 (Low priority + 50% keeps today’s setup).

2. ChatGPT links seem to have stopped carrying utm_source=https://t.co/smErggNeoG since GPT-6 and now open with no referrer, so those visits land in “direct.” Adobe’s Cedric Huesler saw roughly −31k ChatGPT visits/day and +26k direct on the same ~500 pages. If you report AI-assistant traffic to clients, annotate the change before anyone reads it as lost demand.

3. Google Ads API v26 is live, and Search Lift studies are now pullable through the API with 15 new metrics (incremental searches, relative lift, impressions). If you run lift studies, pipe them into the same reporting as your conversions so incrementality stops being a screenshot in a deck.

4. Google Ads lead forms (Search only) now let you mark multiple-choice answers as qualifying. Those submissions get tagged and feed a “Response qualified” conversion action under the Qualified lead goal. If your lead-gen bidding is optimizing toward junk, this is a cheap way to hand it a better signal. Just know the qualifying question becomes required.

5. Microsoft Ads’ Conversions API now supports restate (change value, keep the count) and retract (zero it out) by transaction ID. If you run value-based bidding on Microsoft, wire refunds and cancellations back so the model stops learning from revenue you never kept.

6. Ipsos and Syracuse’s Newhouse school rebuilt 10 real 30-second ads with AI from the same brief and tested all 20 on 3,000 US consumers. Human ads led by about 14% short-term and 17% long-term, but AI won on functional, benefit-led spots, and 54% said learning an ad was AI didn’t change how they felt. I’m letting AI draft the formulaic product-benefit variants and keeping humans on humor and emotion.