Kinase’s own Secret Digital Veteran returns, with a ‘respectful whistle’ and a review of three times Google played the magician with Google Ads

Christmas is a time for… lists! Top 10s! Song of the year! Newcomer of the decade!

But what are they for? Are some lists designed to provoke – either by elevating obvious rubbish or missing out obvious gems? This occurred to me when I asked ChatGPT to give me a Top 3 of ‘things that Google have done to Google Ads that really enraged advertisers and their agencies’.

The list was… bland. The “missing out obvious gems” approach maybe?

Given that ChatGPT is effectively a competitor of Google you’d expect it to put the boot in - but instead it cited “Introduction of Advanced Campaigns (2013)”. Even though I was doing actual work on client accounts in 2013 I had to look that one up. Tip: don’t bother. Also: Introduction of Responsive Search Ads as default (2021) was listed as unpopular. Yeah… I am still appalled that we are supposed to believe that knowing “Top Combinations” is better than actual CTR% data on actual ads we actually wrote ourselves rather than chucked into a collage engine… but automation is only going in one direction at least until the apocalypse. Plus it’s nice to be able to throw 15 headlines into one bucket and see what bobs to the surface (assuming anyone can actually think of 15 separate headlines). Also on Chat GPT’s list from 2021 were the “Changes To Match Types” which finally killed the long-running hack that was Broad Match Modifier by improving Phrase Match so that it understands word order. Well I quite liked that one. Didn’t we all, really? A massive reduction in negative keyword management agony in one hit. Also, I was so busy being grateful that Google is still allowing us to target using keywords at all rather than the “search themes” it’s clearly leading us towards that I forgot to wonder whether Google can really “understand” the intent of every keyword.

So what’s on my list of “Top 3 things that Google have done to Google Ads that really enraged advertisers and their agencies”? Can I do better than AI?

So many things! So much to choose from! But I want to focus on times when Google has acted the magician. Those times when by sleight of hand, by cunning misdirection, they have put up prices in Google Ads when they thought everyone was looking in a different direction. I am opening my “Big Story Book of Google Ads” which goes all the way back to 2002 (UK launch, not US launch in 2000) to pick three of my favourites. They are from 2013, from 2018 and the most recent and best-known from 2023.

1. Shopping launches, text ad CPCs leap up! No one cares!

When Shopping ads launched in 2013 they were… amazing! Credit where credit is due. We got on an early Beta for a large DIY sector client and we pretty much had the whole Shopping real estate to ourselves (mainly on the right hand side of the page on Desktop). So, CPCs were low compared with text ads, and CTR% was really high. People like to click on the new thing. And in spite of my early misgivings about being forced to send all the traffic to product pages rather than experiment with product pages vs category pages, Conv% was really high. Also Shopping Product Groups were monumentally complex to set up (still are, if you’re still playing it old school and are using “Standard Shopping Campaigns”, as you might be, to meet some use cases that still exist) and horribly error prone – and many advertisers and other agencies made an enormous hash of setting them up so we loved them, and within a couple of years we were spending 50%+ of our budgets on them. Also, “custom labels” were an early method of “leveraging first party data”, the Holy Grail of the mid 2020s, ten years early. So, we were happy and so were our clients.

““custom labels” were an early method of “leveraging first party data”, the Holy Grail of the mid 2020s, ten years early”

Until someone said, “But wait, what’s happened to our text ads CPCs? They’ve shot up!” It was true. It was immediately obvious what had happened. Impressions were flat, but CTR% took a massive hit when Shopping launched. Having multiple ads on a single SERP was great. Many opportunities for a click – but the fundamental law hadn’t changed that each SERP can only yield a maximum of one click. So, CTR% on text ads inevitably went down as clicks on Shopping increased. So QS on text ads went down, and CPCs went up. When most companies introduce a new product they start charging less for the old product. Not Google. They started charging more. Few people noticed and fewer people cared. Impressive really. I let out a respectful whistle. As in Joseph Heller’s Catch 22: “Yossarian was moved very deeply by the absolute simplicity of this clause of Catch-22 and let out a respectful whistle.”

It wasn’t the last time I was moved to let out a respectful whistle.

This effect contributed to an increase in CPCs after a relatively long period of flat or even declining CPCs. In many places, the decline was put down to the shift to mobile impressions, which conventional wisdom held to be “cheaper” (partly because unsophisticated cross-device measurement at the time tended to show mobile Conv% as lower, and the traffic therefore less valuable). Another factor was the introduction of “Enhanced Campaigns” (Nope! Still don’t remember them!) which apparently increased CPCs through the use of mysterious “signals” which raised / lowered bids according to mysterious, unknowable factors. Imagine that! Commentators did note that increased click share for Shopping increased CPCs overall in retail verticals… but I don’t remember anyone talking about the knock-on effect on text ads.

2. ‘Twas the night before Christmas and Shopping ads started appearing for totally irrelevant searches

Ok – Not exactly the night before Christmas. But at a time which, even in 2018, was well on its way to becoming even more important than Christmas: Black Friday. Or Cyber Weekend, or Cyber Week, or Black Friday Fortnight. Whatever.

Anyway, during Cyber Weekend we noticed that Shopping ads were appearing for increasingly irrelevant terms and generating wasted spends across almost all clients. This is one of the reasons why we get nervous every time Google degrades keyword- or search term-level data (relatedly, see our articles on Performance Max). We don’t trust what we can’t see and verify for ourselves. In fairness, Shopping targeting had become pretty good by 2018 and we hadn’t had any problems – at least not on the scale we encountered in Q4 2018. Of course, what we experienced, along with many other agencies and advertisers, had absolutely nothing at all to do with Alphabet (not Google Inc. since 2015) experiencing a drop in advertising prices (aka our friend the CPC) that would somewhat un-nerve the markets in the following February when its Q4 earnings were announced. Nothing at all to do with that, no.

Here’s some example search terms which began triggering search times for clients – which in the previous week had not been triggering our clients’ ads: Jewellery client – [motorbike], [Nintendo], [Supreme]. Stationery client – [giraffe], [pictures], [navy blue]. Furniture client – [emerald ring], [church pew]. Google’s suggestion was to add negative keywords. Thanks! Any suggestions about what to add as negatives then Google? Should I add every keyword in the dictionary?

“Any suggestions about what to add as negatives then Google? Should I add every keyword in the dictionary?”

I think we just had to keep a close eye on the search terms and add negatives as they came up. Not much fun. Of course, CPCs increased through the mechanism of Quality Score. If the Shopping ad appears and it’s totally irrelevant to the search term and someone clicks on it (because of course, people do, don’t they?) then QS will be lower and CPCs will be higher – which is bad for the client, paying out for traffic they don’t even want and never bid for, but good for Google which entirely coincidentally is now able to show in its quarterly report that “advertising prices” are not falling and there’s no need for investor unease.

Broader matching means that more advertisers are gathered together to end up bidding in each auction, and CPCs are pushed up. The second effect which compounds this CPC intensification is that the very mechanism which is supposed to ensure fairness and keeps cost low, ie. Quality Score, is used to generate extra $$$ for Google at the expense of the advertiser.

Anyway: pretty crude move. Someone at Google HQ loosens the bolt on the Black Box marked “relevance of query to search term” and Christmas is (nearly) ruined!

No “respectful whistle” from me for pure cynicism.

3. What if… Google Ads isn’t actually based on an auction at all?

Which brings me onto my third, most recent and most famous example of things that Google has done that really enraged advertisers and agencies. What if… there isn’t really an auction at all? What if Google just decides, at random, what ads to put in at the top of the page?

Yes, I am talking about the Randomised Generalised Second-Price ad auctions (or RGSP) which we have already covered on the Kinase blog. But I am happy to talk about it again with anyone who is happy to listen, for as long as possible and in as much detail (perhaps more) than anyone can stand.

What is the RGSP? Well, this is an excellent summary but it’s basically a mechanism through which Google works out who has won an ad auction and then decides to award the top spot, at random, to another advertiser. Advertisers at large – naively assuming that the black box of the auction was a rational or at least a non-random system - will therefore be encouraged to bid more in order to achieve greater prominence, these days through automated bidding tools rather than through setting bids manually (obvi).

This all came to light during the US Department of Justice’s antitrust trial against the company in 2023 – which shed light on some Google internal communication.

As we said in our article last year: “Anything to do with the auction is a hot topic as it’s both a black box, and the determining factor for advertising costs on Google. As Jerry Dischler, VP for Google’s Advertising Products, noted during the trial: “We tend not to tell advertisers about pricing changes.” In other words, there are various ways in which the cost of advertising on Google has risen, and Google is not in the habit of sharing them.

“The auction is intensified and advertisers with smart bidding objectives for top of the page, absolute top, or conversion-oriented goals (if the signals are that being top for this query at this time is best for conversion KPIs) will find CPCs rising.”

So Google had a bid rule option for being ‘absolute top’ while also having a mechanism to make it impossible to really be absolute top. An upward spiral of costs.

“So Google had a bid rule option for being ‘absolute top’ while also having a mechanism to make it impossible to really be absolute top.”

On one level, this was kind of appalling – if not surprising. But on another level I confess I felt relief. I’ve explained the Google Ads auction maybe 1,000 times – to clients, to trainees, even to my mum, as we all have – and I’d described it as a “black box” but I felt a bit shifty. Not so much in the definition of QS or Ad Rank, but the stuff about “only paying 1 cent more than the advertiser below you”.

Really? How can this ever be independently verified? Now I can stop worrying. Even if it were verified, it wouldn’t matter!

As we all know, the DOJ has more recently determined that Google does have a monopoly on search. It’s proposing a series of measures to address this – many make no sense at all (isn’t Google right? If people don’t like Google, they can easily go somewhere else), others go too far, and none of them, as far as I can tell, will address any of the issues I’ve outlined in my three examples about how the Google Ads algorithm actually works.

It remains to be seen whether the DOJ will gets its way – assuming the DOJ still exists after 20th January 2025.

Christmas conclusions of sorts

So those were the three best stories from my Google storybook. In fact the example from last year is so well known and caused such uproar (as far as these things go) that I am surprised that it didn’t make ChatGPT’s list. In fact, using AI to recall the earlier histories of Google’s AI (Auction algorithms, Keyword matching, Quality Score) suddenly struck me as ironic. Unintended consequences and money pouring into black boxes, in the past and the present.

One thing ChatGPT could have pointed out, maybe in the friendly / chatty summary sign-offs that it puts at the end of a lot of its responses, is there’s really no point getting angry with Google. It’s not like there are any other search engines with comparable reach that advertisers can give their money to. At least, not right now. Check back with me again in a year. I’ve never felt so optimistic.