AI is becoming an increasingly important part of PPC management, with advertisers exploring new tools and ways to use it in their daily work. But how are they checking AI-generated results for accuracy and completeness? What non-AI automation tools are they still using, and what challenges have they faced with AI? These questions and more were discussed during this week’s PPCChat, hosted by Robert Brady.
Q1: What is your current go-to AI tool in your PPC work? If you have a couple, what tasks do you give to which tool?
For PPC work I’ve found myself leaning into Claude more. Seems to be more balanced and less sycophantic than others. (I use Gemini for personal questions. Keep the data pools separate) @robert_brady
Claude and Copilot are what I’m permitted to use at work, so I start there. And we have a bunch of ai-powered proprietary tools for ad copy and feed optimisation. Not everything can just plug into an account like some of the market tools, but they’re cool. @JuliaVyse
I’ve got a custom-built AI tool that helps develop the rough draft of ad briefs and landing page copy. But, ChatGPT is my daily driver and I think it’s getting good enough for a lot of PPC tasks I don’t care to do on my own. @Pete_Bowen
We use Gemini because it’s built into our Google workspace, so I trust it from a privacy setting, etc. @revaminkoff
I tried to test which tools were best back when I wrote this article: https://www.wordstream.com/blog/how-accurate-is-ai-for-ppc But I’ve still just defaulted to ChatGPT since I’ve been working it with it the longest over time or I’ve found Gemini is my go-to for super Google Ads specific tasks @Susie Marino
Claude combined with either the Optmyzr MCP connector or one that I wrote (vibe coded) myself for some tasks. Doing “Claudits” is very effective and I have it setup to fit my workflow. The nice thing about using Optmyzr’s is that there is an approval process. When I use my own, I have Claude generate an interactive checklist of changes it is suggesting. We also have a custom skill that can generate all aspects of a campaign from a landing page (and hopefully a plan on a page) and then after approval those plans can be (very accurately) created as campaigns in Google Ads via the MCP connector. @EronCohen
@EronCohen I’ve got a question later about non-AI tools you’re layering with AI. Would love to hear more about how you’ve tied in Optmyzr when we get there @robert_brady
BTW, for those unfamiliar with what an MCP connector is, just think of it as a way to give your favorite LLM a way to directly access all aspects of your Google Ads campaign without having to export it and then re-import it. So, for example, you COULD use Google Ads editor to export your campaign setup and then analyze it with Claude or ChatGPT — that works great, but it’s a better prospect and overall experience to have the LLM directly connected to your account. @EronCohen
Bonus tip: If you want, you can get Claude or ChatGPT to generate a CSV file that can be imported into Google Ads editor. So you could have it create an entire campaign and then bring it into the Google Ads editor to preview it and approve. @EronCohen
Q2: How are you auditing your AI outputs to ensure accuracy and thoroughness?
I see these horror stories on the web about AI making up numbers, data, graphs, etc. and it makes me a little nervous. @robert_brady
Yeah I really still am not super happy with the majority of AI responses, so I only use it for ideas for how to approach something, like I’ll ask it the same task I’m trying to do, skim its answer, then put it away and approach it myself still – just for my own peace of mind and sanity that there’s no chance for any AI mistakes to slip through the cracks, not worth the risk to me @Susie Marino
by hand, and the time investment is REAL! I’m always mindful that while we can easily generate big plans and get help with large accounts (mccs with sub mccss, with SUB mccs), the client still has to review all this. Trying to make it as simple as possible @JuliaVyse
I’m using it only for the first pass so everthing gets reviewed by a human before going live @Pete_Bowen
I will say if you have an export to excel, your QA can become a template. like, pop the ai answer into a ‘data’ tab, and have your other tabs look for weird numbers, negative keywords, that kind of thing @JuliaVyse
@Pete_Bowen Same basically @Susie Marino
I definitely lean toward the human, “by hand” review. But it makes me feel like I’m not really saving any time using AI. I’m just turning myself into the editor when I want to be the strategist @robert_brady
I’m literally drowning, and as usual we’re in a hiring freeze. No choice but to get the robots to work for me, but trying to make them smart. @robert_brady that struggle is so real. I equate it to reviewing my team’s output which helps, but it’s not ideal @JuliaVyse
Anyone using AI to QA the work of the other AI? I’ve heard of people doing it. @robert_brady
Lol I’ve heard of this too. Honestly I mean yeah I have asked two different tools to do the same task and compare that way @Susie Marino
@robert_brady in the spirit of curiosity I did it once, and it didn’t save me much time. @JuliaVyse
@robert_brady I have heard of doing that too and the main reason I don’t is because of trying to keep the number of tokens I use down. @EronCohen
@robert_brady I’m not yet convinced that AI has saved meaningful time. Sure, ChatGPT just found me 300 competitors to exclude, and a whole bunch on negative keywords, but IMHO it hasn’t yet lived up to the fire-and-forget hype @Pete_Bowen
@robert_brady I’m strongly considering building that in, if I ever get to that point. Right now, most of my AI outputs haven’t been accurate and definitively need human checking. @revaminkoff
Right I think the time I spend crafting the right prompts and reviewing outputs hasn’t been any less time overall @Susie Marino
I have found mistakes in the AI outputs too. For sure it depends which model you chose too. The newer Claude (Opus 5.5) models and OpenAI (Astra/Sol 6.1) are generally much better at the job. @EronCohen
@Susie Marino If you’re not saving time, do you think the quality of work is improved by looping in AI? @robert_brady
@robert_brady good question! I would say it depends! I mean it has helped me consider other angles to approach something in an account I might not have thought of, because you know how easy it can be to get stuck in your own personal preference of how to approach PPC, but I wouldn’t say it’s, like, helping me create work that’s better than what I’d be able to output on my own anyway @Susie Marino
Thank you for sharing. I think as the tools continually evolve and we humans evolve in how we leverage them, the more often we’ll be able to do superior work as a team with AI tools. @robert_brady
Q3: Which AI tool have you been disappointed with? Why did it fail to meet your expectations?
Copilot has been probably the most underwhelming experience for me. @robert_brady
Copilot, but I wouldn’t say it failed against expectations…Overall I’ve been pretty happy with Claude and some personal projects I’ve used Perplexity for. I think if you assume it’s a stupid robot and treat it like it can do VERY SPECIFIC things quickly, you’ll be fine. If you treat it like ‘plug me in and go to lunch’ you will not do well @JuliaVyse
I have been very disappointed with Google’s Ad Bot or whatever they’re currently calling the new bot built into Google Ads. I wanted to use it to get insights on asset performance, and what it told me wasn’t accurate and changed every time I asked. @revaminkoff
Anything Google built into the UI. @Pete_Bowen
Ads Advisor is the current name for Google Ads built-in AI. And it’s beta at best. @robert_brady
@revaminkoff I particularly love it when the Google Ads bot tries to tell me what isn’t working with a 20-second old campaign with zero data. @ynotweb
The AMOUNT of times I’ve gotten advice from the ad bot that is automatically rejected in the ad editor, gang, make the bots aware of your own policies at LEAST @JuliaVyse
Ads Advisor is definitely improving though…and it will actually change things for you now if you want it to…it is getting there. @EronCohen
@EronCohen in a good way, or one more thing to worry about? @JuliaVyse
@EronCohen And I expect it to improve quickly. I just think it got pushed to full production too early to keep investors/Wall Street happy that Google was leveraging AI in the primary profit-generator of the business @robert_brady
@JuliaVyse Well you always have to be on your guard/check behind the AIs, right? I have used it only once to change things: I had it go through and turn off AI Max in the campaigns that Google Ads automatically turned it on. @EronCohen
Using Google AI to defeat auto-applied stuff. That’s genius @robert_brady
All google stuff seems to be awful for the first 2 years and then it starts to work. I suspect the AI will be the same @Pete_Bowen
@EronCohen is out here in 3026 leaving us behind! @JuliaVyse
Ask Advisor in Google Ads has given me some interesting thought starters for reporting & optimizations, but fails to deliver usable hard outputs like a list of suggested negative keywords. I also get a little frustrated with the inconsistent results I get from Gemini when I use a prompt that previously worked well for me. Obligatory disparaging comment about Copilot because it’s the equivalent of a bad coworker. @rebeccawarne
Q4: What AI tool is next on your list to try? Why do you want to check it out? Why does it excite you?
I’m going to give Muse a whirl so I understand how consumers might start using it. Meta is building adoption quickly. @robert_brady
This will sound weird to those who know me but here goes: Muse is REALLY good. I’m as surprised as you are. It’s not perfect, but the interface is very friendly, smooth, learns fast, outputs strong. @JuliaVyse
I really want to try Muse! @revaminkoff
The vibe I get from Muse is that it’s like early Apple (think iPod and iPhone v1). They’re taking something complex and potentially scary to tech laggards and making it “just work”. The marketing and framing is really compelling to a less sophisticated user (aka the fat middle of the tech adoption curve) so it has huge potential. @robert_brady
Anyone who has tried ChatGPT knows some of the headaches – the long scroll, the reprompting, several chats for one thing. Muse seems to have learned from the other launches and built something very friendly and focused on being intuitive. Trying to be the daily use AI, not the biz one just yet. @JuliaVyse
Q5: What non-AI automation systems are you currently using alongside AI solutions?
I still use quite a few Google Ads scripts and having AI has made it easier to create them as a non-coder. @robert_brady
I’ve got systems for uploading offline conversions, preventing spam etc. Funny thinking of them as non AI though @Pete_Bowen
I use alot of Google Sheets. The connectors for Claude and ChatGPT have gotten great at creating and updating Google Sheets. Part of my work process is to create a log of what I have done, so now I have a skill where if I use Claude to update some aspect of a campaign, it logs it itself using the format I need it to in a Google sheet. @EronCohen
@Pete_Bowen One of the biggest differentiators to me is that AIs are probabilistic while “non-AI” stuff is deterministic in nature. @robert_brady
@robert_brady That’s an important distinction. I like derministic a lot. @Pete_Bowen
I use a lot of process automations, think Power Automate + microsoft 365. The trick is, big giant holdco, so not all my team has the same tools or access levels. A lot of our process is captured and shared in docs, google sheets, excel. @JuliaVyse
@EronCohen And how are you connecting Optmyzr to your stack @robert_brady
@EronCohenI want to pick your brain on building artifacts and automations in Claude, if you’re willing to share! @JuliaVyse
@Pete_Bowen How are you going about spam prevention? @robert_brady
I’d have to kill you if I told you @Pete_Bowen
Probably true. Spam is a big industry @robert_brady
Meta One is a gift to the spam world. crafting an alert for my gov clients now @JuliaVyse
It’s been a 17-year slog. I’ve got something that works now using a 2-step form + some AI @Pete_Bowen
I do use artifacts in Claude to build my checklists @JuliaVyse I basically ask Claude to audit a given account, but instead of just writing out all of the changes it suggests, I have it build me an interactive checklist that I can then click Fix, Skip or discuss and fill in a field (see screenshot) — when I am done going through each item in the checklist, I let it know and it goes through and makes the updates in the campaigns for me, then logs them in my tracking sheet. @EronCohen
I agree with @robert_brady on Google Scripts. It’s a good way to get consistent, repeatable results…but you can use an LLM to help you build the script you need. They are really powerful–especially when you can have a world-class developer (ChatGPT/Claude) build it for you. @EronCohen
@EronCohen I do something similar with my claudits, with a scoring system for pitches. haven’t cracked account builds yet, too many areas for error and no great way to review, at least not yet. any advice? @JuliaVyse
@JuliaVyse Are you using an MCP connector for Google Ads yet? @EronCohen
@EronCohen no, I have clients with VERY heavy privacy policies, so not a lot of direct connects available to me. But I do like scripts and scheduled exports. @JuliaVyse
That makes sense @JuliaVyse Well maybe if you have any personal accounts to try it with, you can give it a try. @EronCohen
Q6: What pitfalls have you encountered in your AI journey that you would have warned your past self about?
I haven’t done this, but it seems like every day I see another agency turn their SOPs into an AI tool in the hope of some sweet recurring SAAS income. I suspect that this is a waste of time for most. I get the attraction, I’ve built a load of software, but I think this is like building a better calculator the day before Excel was invented. @Pete_Bowen
Be ready for challenge. If you are pressed for time and understaffed like me, it’s very useful to have a robot help with work. When you get an answer that you agree with and seems credible, make sure that answer can stand up to questioning from other teams. Just because you agree based on your expertise doesn’t mean you can explain it in detail. @JuliaVyse
My advice would be to remember that AIs are not actual artificial intelligence yet. Yes, it’s digesting immense amounts of information/data, but it’s output will only be the average of all that information. Said another way, AI will always be firmly inside the box. @robert_brady
@Pete_Bowen my SOPs are so specific! truly unsellable, but very helpful to my specific team. @robert_brady YES! AI can help you with a lot of things, but it can’t innovate the way you can. at very best, with a bit of luck, it can surface an insight you missed, and then the human brain needs to take over. @JuliaVyse
Another caution would be to push back on outputs. AI tools are programmed to be extra agreeable and don’t ever want to say something like “I don’t know” If you found the right huckster @JuliaVyse they could definitely sell your SOPs to someone. @robert_brady
I like to use projects where I can instruct the ai to disagree more, or at least keep the language plain. Not a perfect fit, but very helpful @robert_brady I could be convinced to put them in an ebook of some kind @JuliaVyse
Q7: If you’re willing, describe your full AI stack and how you’re using it. Which piece was the biggest unlock for you?
Claude Enterprise with shareable Projects & Artifacts for me. It’s a powerful tool, and being able to work together was the unlock for me. As usual, my superpower is based on working together! @JuliaVyse
I feel like I’m pretty early in my development of an AI “stack”, but Claude is my go-to LLM tool. I’m also a longtime Adalysis user and they have an MCP that I’m starting to get into. And I use Claude to help me create/edit scripts that I use in Google Ads. @robert_brady
I would say the biggest unlock is the quality of work I can get from Claude. It’s not perfect, still needs a human in the loop, but saves hours and hours…especially on campaign builds. Give it a landing page, get out a most-of-the-way-there Google Sheet with all the details. Have you tried just pointing ChatGPT or Claude at a website page and asking it how to improve it? @EronCohen
I am also using Claude/ChatGPT to build websites on the side. It is so very effective. As an experiment, I had it design and run a WordPress site for a local blog from scratch. I pointed it in the right direction by handing it a few text articles from a blog I took down long ago. It is pretty decent and may even be helpful although it is mainly “listicles.” Again, none of this is “perfect” but it saves me hours. @EronCohen
PPCChat Participants
- Robert Brady @robert_brady
- Julia Vyse @JuliaVyse
- Peter Bowen @Pete_Bowen
- Reva Minkoff @revaminkoff
- Susie Marino @Susie Marino
- Eron Cohen @EronCohen
- Kammy Caruss @ynotweb
- Rebecca Warne @rebeccawarne
Related Links
Stop the wasted ad spend. Get more conversions from the same ad budget.
Our customers save over $16 Million per year on Google and Amazon Ads.