TrueClicks ads

⚔️ 𝐁𝐚𝐭𝐭𝐥𝐞 of the AI models: which AI is best for Google Ads? (today :) I had a good reason to find out this week becuase recently a wave of new models landed across Claude, Gemini and OpenAI, and we updated all of them in TrueClicks AI assistant: → every Claude and Gemini model we support, on the latest version → three new OpenAI models we didn't support before: GPT-6 Astra, GPT-6.1 Sol, GPT-6 Luna → nine in total, our users can pick the one they want Not many PPC tools let you choose your model, and I think you should. Here's why: I gave all nine models the same prompt: 𝐚 𝐡𝐲𝐠𝐢𝐞𝐧𝐞 𝐚𝐮𝐝𝐢𝐭, deliberately mid-difficulty, on one real account with 17 verified hygiene issues. And the result? ✅ 𝐅𝐚𝐛𝐥𝐞 5.1 and 𝐎𝐩𝐮𝐬 5.5: 13 of 17, zero errors. The only two that caught the finding that mattered most: 1000+ negative keywords sitting on a campaign that ended in 2024. ✅ 𝐆𝐏𝐓-6 𝐀𝐬𝐭𝐫𝐚 and 𝐆𝐏𝐓-6.1 𝐒𝐨𝐥: precise on everything they claimed, but both missed the negative keywords. ✅ 𝐇𝐚𝐢𝐤𝐮 5.5: the surprise. Better than both Geminis and GPT-6 Luna, at about a tenth of what Opus costs per run. ⚠️ 𝐆𝐞𝐦𝐢𝐧𝐢 3.1 𝐏𝐫𝐨: fastest run at 45 seconds, surprisingly for me, the worst result. Reported paused campaigns as live problems. Found only 4 of 17 issues. FYI Cost per audit: about $0.02 (GPT-6 Luna) to about $2.50 (Fable 5.1). Opus 5.5 lands around $0.70 with the same accuracy as Fable. Full write-up with the findings matrix and live-account verification: https://lnkd.in/eZXM4gPq ⬇️⬇️⬇️ If you’re at ADworld Experience in Bologna 🇮🇹 this week, come find me. I’m running a focus group on AI in PPC for agencies and advertisers. Sign up here: https://lnkd.in/e237xwnW

10,000 search terms reviewed in 2 minutes for $0.29, without writing a single line of code. Using Jev, the new AI model that's been all over my feed this week, hooked up to Google Ads through Claude. 𝘓𝘦𝘵 𝘮𝘦 𝘴𝘩𝘰𝘸 𝘺𝘰𝘶 𝘩𝘰𝘸. 🔍 First, what Jev does: you give it a question with a fixed set of possible answers, and it picks one. It doesn't write text. Ask whether a search term is brand or non-brand and you get back "brand" plus a confidence score. That's the whole product. Sounds limited until you point it at 50,000 search terms. Is this a competitor name, someone looking for a job, someone who wants to repair instead of buy? Brand or non-brand, including the typos. Answers come back in seconds, at a fraction of what the same job costs on an LLM, and with highly consistent results. 😱 The catch: no chat window, no MCP server to connect. 𝘐𝘵'𝘴 𝘢 𝘙𝘌𝘚𝘛 𝘈𝘗𝘐. You create an API key and call the endpoint yourself, which puts it out of reach for most PPC specialists. ▶️ The video above shows the way around that. TypeSafe publish an official plugin for Claude Code, we have a Google Ads plugin for Claude Code, and connecting the two takes a few minutes and no code. 📍 And if you're coming to ADworld Experience in Bologna on October 15 and 16, I'm running the AI Tools for PPC focus group, both days. Building AI for PPC is what I do most, including the free Google Ads MCP by TrueClicks, the Google Ads plugin for Claude Code, and the TrueClicks AI assistant. Bring your hard questions. Seats in the focus groups are limited, so sign up if you want one.

Budget-limited AND a target ROAS or CPA. That's a wrong combination to have in the first place. I put that out there quite strongly. Some very smart people I respect are still pushing back on me. So Ruben Runneboom had me on Taskforce Talks yesterday to hash it out for 45 minutes. Watch the full episode on YouTube: https://lnkd.in/ek_9maH8 All the free calculators we discuss are at trueclicks.com/calculators The YouTube description has the rest: the resources I lean on most for target setting, plus a few timeless PPC principles. Worth a look even if you skip the video. Agree, disagree, tell me in the comments! I'm still collecting counterarguments.

PPC LinkedIn is furious about Google's August 17 bidding change. I'd call it a money grab too, and I expect CPCs to rise for everyone as stale targets start bidding at full strength. 𝗕𝘂𝘁 𝘁𝗵𝗲 𝗮𝗻𝗴𝗲𝗿 𝗶𝘀 𝗱𝗿𝗼𝘄𝗻𝗶𝗻𝗴 𝗼𝘂𝘁 𝘁𝗵𝗲 𝘂𝗻𝗰𝗼𝗺𝗳𝗼𝗿𝘁𝗮𝗯𝗹𝗲 𝗽𝗮𝗿𝘁. The only campaigns hit by this change are the ones ignoring the oldest law in paid search: you can't control spend and efficiency at the same time. The more you fix one, the less you control the other. A hard budget cap AND a tCPA or tROAS is two incompatible demands in one campaign. Organizational psychology calls this 𝗶𝗻𝘁𝗿𝗮𝘀𝗲𝗻𝗱𝗲𝗿 𝗿𝗼𝗹𝗲 𝗰𝗼𝗻𝗳𝗹𝗶𝗰𝘁: one boss telling one employee "work faster AND better, with the same resources." Every employee in that spot quietly picks one instruction to ignore. Smart Bidding was that employee. It obeyed your budget and treated your target as a suggestion. On August 17, it switches allegiance: the target becomes the instruction it follows, and your budget just caps the day. Which is why I disagree with the most common advice right now: "𝘫𝘶𝘴𝘵 𝘴𝘦𝘵 𝘺𝘰𝘶𝘳 𝘵𝘢𝘳𝘨𝘦𝘵 𝘵𝘰 𝘺𝘰𝘶𝘳 𝘳𝘦𝘤𝘦𝘯𝘵 𝘢𝘤𝘵𝘶𝘢𝘭𝘴." If the campaign stays limited by budget, you haven't resolved the conflict. You've handed the algorithm a fresh pair of contradictory orders. The fix is choosing your real constraint: → Budget? Drop the efficiency target. Maximize Clicks, Conversions or Maximize Conversion Value. → Efficiency? Set your tCPA or tROAS from your unit economics, and give the budget room to breathe. (Yes, the loose-target trick worked for years. It was a hack. Google just deprecated it.) The flowchart below walks you through that choice, including the conversion tracking each path requires. August 17 isn't taking away your control. It's ending the illusion that you had both kinds. Some sharp people wrote about this change, and I'm curious where they land: pick one constraint, or keep trying to have both? Aaron Levy, Ameet Khabra, Ben Luong, Birgit De Vries, Chris Nightingale, Ginny Marvin, Hana Kobzová, Joey Bidner, Kirk Williams, Thomas Eccel, Tomasz Abbott-Wieczorek

Can you show me, on paper, what your Target ROAS or CPA is based on? If the honest answer is "𝘮𝘺 𝘤𝘭𝘪𝘦𝘯𝘵 𝘴𝘦𝘵 𝘪𝘵" or "𝘪𝘵 𝘤𝘢𝘮𝘦 𝘸𝘪𝘵𝘩 𝘵𝘩𝘦 𝘢𝘤𝘤𝘰𝘶𝘯𝘵," I wrote something for you. My latest piece for Search Engine Land just went live, and this one has the exact formulas behind a profitable and achievable ROAS or CPA target. If you haven't worked through numbers like these in the past year, or you've never seen all of them, drop what you're doing and go read it (link in the comments). It's more important than the latest Claude trick, I promise. Your target is the single biggest lever you still control. Get it wrong in either direction and you either leave growth on the table or quietly lose money on every sale. Yet most of us never actually calculate it. The article walks through all four steps: → your break-even floor → the target your margin can actually support (inside-out) → the target the auction will actually allow (outside-in) → where your last dollar stops making money Not just so you can set a sharper target. So you can push back on an unrealistic one with data instead of a shrug. Stop inheriting your targets. Start choosing them. Hat tip to George Michie for his timeless "Maximizing ROI: the wrong game" article and to fellow Dutchies Bob Meijer and Miles McNair for covering the topic in their PPC Mastery materials.

Back in 2018, Baby Shark started showing up in every placement report I opened. It never left. You exclude one kids' channel, and next month a different cartoon takes its spot. Then three more. New ones go live every day. They show up most in Video campaigns. There, you can see and exclude every channel, but doing it manually, one at a time across every account, takes hours of work and doesn't scale. The waste that hurts most is in PMax, Demand Gen, and App, where even that control is gone. PMax won't show you clicks per placement, so it hides in your totals. So we stopped playing whack-a-mole. We scraped nearly 4 million YouTube channels and classified each by what it publishes, not by a label the creators set themselves. We found 30,818 aimed at kids. 46% weren't in English, across 100 languages. As far as we know, it's the most extensive, accurate, and current list of YouTube kids' channels anywhere, and we refresh it every quarter to keep it that way. Free and ungated. Link plus four ways to use it in the comments. Know someone whose budget is funding Cocomelon? Tag them below!