Gemini tries to avoid the worst of the web, and for most people ads and tracking are high on that list.
But, there is a lot about how ads work that is not common knowledge.
You almost certainly know what online ads look like; you may even have clicked on them from time to time—possibly by mistake. You might have considered what information is flying around in the cloud to support those ads. You’re less likely, I think, to have considered how much money changed hands, and between whom.
If you know the enemy and know yourself, you need not fear the result of a hundred battles — Sun Tzu
Let’s postulate that ads are the enemy—so, we should learn something about how they work.
I have a pretty good knowledge of how online ads work, so, I’ll share.
I’ll start from the point of view of a small business: a bike shop. I make and sell great bikes, and what to let people know, so they’ll come to my shop and buy something.
Let’s say my profit margin on a bike is $50. Revenue is important, not just profit—because revenue pays wages and bills. So I’m willing to pay up to $50 for an additional sale.
Enter the web. There are obviously people on the web who ride bikes. How do I reach them?
Here are the pieces that will be involved in helping me to sell more bikes.
Here is the first big piece you might not have given much thought to.
As the bike shop owner, I need some way to arrange for ads to run.
I need to be able to put together text, images, and so on, to describe what my ads should look like. I need to some way of describing who the ads should show to. And, I need to say when the ads should run, and how much money I want to spend.
Here is the second big piece.
As a website owner, I need some way to be able to say “please run ads on my site and give me some money”.
I’ll probably want some kind of control over the type of ads that are allowed to run on my site; at the very least I’ll need reasonably guarantees that they are not malicious or outrageous.
The third big piece is reasonably obvious: there needs to be some server that serves ad content at the point where you view an ad.
And, site owners need to somehow integrate with it.
And the fourth big piece is positively notorious: the user tracking.
But why does user tracking exist at all? What’s it for?
Let’s go back to the bike example.
It so happens that there is a local web enthusiast who creates `bikes.example.com`, a forum for bike enthusiasts in the same area as the shop. It’s run by Ross, who pays the cost of hosting out of pocket.
One day he wonders about getting some of that money back, and signs up at some Tool For Site Owners. He sets everything up to serve ads and waits for the money to roll in.
Let’s name our bike shop owner, too; she can be Rachel. So Rachel signs up at some Tool For Advertisers, creates an ad, and waits for the additional sales to roll in. (And the bikes, I guess, to roll out).
If these are the only two entities in the ad space, great! All of Rachel’s ads get served on `bikes.example.com`. The only ads on the biking site are about bikes, and because the forum is for users in the same area, all the ads are relevant and reasonably interesting. Ad clicks are reasonably likely to lead to shop visits and sales, so the clicks are high value. Rachel is happy to pay for them, Ross is happy to get the money, the site users aren’t too annoyed by the ads—and sometimes they’re useful. The cost of the forum is paid. Everyone wins.
But expand this up to the whole web, with sites about everywhere and everything, and users from everywhere and with all kinds of interests, and it obviously doesn’t work.
Rachel’s bike ads will show to the whole Internet, and be just spam; they’ll drive no sales, and be worthless. Ross will get ads from the whole world; they’ll drive no sales, and be worthless. Users will see worthless ads. Everybody loses.
This is how we get to the subject of targeting.
Targeting means showing ads to the “right” people; in the case of direct sales ads, that means to people who are likely to buy the product; and in the case of brand advertising, it means to the people who you want to be aware of your brand or products.
The most important types of targeting are so basic you may never have even thought about them; let’s start with those, then work to the most specific, ending with the most “creepy”.
This is pretty fundamental: at a minimum you need to show ads to people in the right country, and in many cases it needs to be the right region.
An ad that is for a product you have to take a plane to buy is a bad ad. Nobody wins.
In Rachel’s case, she wants to target her ads to people in the city her shop is in. This immediately and dramatically improves both the chance of a sale and the usefulness of the ad to users.
Notice how targeting is dual purpose: it increases value for both the advertiser and the user. This is always the case, if the ad is genuine—for something that people might buy and be happy with. If the ad is for a scam, of course, there will never be a happy user.
Geo targeting could be based on the user IP address for broad targeting, or information added to a user account if they are logged in.
Just as fundamental as geo targeting is language targeting.
An ad that you can’t understand because it’s in the wrong language is a bad ad. Nobody wins.
Rachel’s shop is in NYC so she targets English.
Language targeting can be approximated from IP address, with location mapped to language. Another option is to match the language of the site being viewed.
Particular websites often have a theme, and so do ads.
Ross can tag his site with the “bike” theme; and Rachel can target her ads to the “bike” theme. Ross’s site will get more ads about bikes, and Rachel’s ads will show on more sites that are about bikes.
With these three types of targeting, we start to get to a point where we are close to the situation first described—where the only advertiser is a bike shop, and all the ads are about bikes.
But, of course, targeting and tracking go further.
Rachel knows from day to day business the approximate age range of her clientele; they are usually 20 to 50. So, she optimizes by showing ads to just this age range.
Also, she has some bikes she wants to advertise just to men and some she wants to advertise just to women. So she tags those ads with additional targeting.
There are various possibilities for targeting demographics. Sites might themselves be bucketed by their typical demographics, then matched to ads. Individual user demographics might be guessed from behaviour—but more likely, they are simply asked when they create their account.
With topic targeting, we can show ads about bikes on sites about bikes.
Interest targeting is an evolution of topic targeting. Users are tagged with the topics of sites they regularly visit—their “interests”.
So if Rachel turns on interest targeting for her bike ads, those ads will show to people who visit bike sites—but on other sites, too.
This is the big one.
Remarketing is showing ads to people who already interacted with your business in some way.
If you nearly buy a product in an online store, but bail out, and then you see an ad for it immediately afterwards on some other site—this is remarketing.
Remarketing is incredibly high value. Because you almost bought a product, you are obviously the exact target audience—you are right in the decision process about whether to buy, and showing you an ad about it might very well tip the balance. Unfortunately it can also feel intrusive.
Could Rachel use remarketing if you visited her store? Only if she learned your digital identity in some way—for example, by having you sign up for a loyalty card with your email address. Then, the Tool For Advertisers that she uses might allow her to upload that email address and target ads specifically at you. I personally consider this Not Okay—but then, most loyalty cards are already clear “no go” areas for me in terms of privacy. Last I checked, not all online advertisers offer this type of remarketing, and I’m happy about that.
Through all these types of targeting it’s important to notice that the information about you flows no further than the serving stack. It’s used to decide what ad to show, but it never flows on to the advertiser. The Tool For Advertisers does not expose user information.
The Tool For Advertisers probably does show stats on how many times particular ads served, and care needs to be taken to make sure there is no information leak there. For example, an advertiser could come up with targeting criteria so specific that it can match only one person; if they get a report saying that someone saw that ad on Saturday then it’s a report about that individual. So serving stats needs to hidden when the numbers are too small.
All of these targeting possibilities contribute to ad value.
Ad value links to a lot of things:
Again using the bike shop example: the value of one ad targeted to the entire world, is zero.
Targeted to the right region and language it might be a tenth of a cent. With $50 Rachel can buy 50,000 ad views, getting one additional sale.
Targeted to the right topic as well, it could be half a cent; and with demographic, one cent.
Targeting by interest might not particularly increase the value—rather, it gives more opportunity to serve the same ad at the same value, by spreading across more sites.
Remarketing, as I’ve said, is the big one. It might increase that value to fifty cents, a dollar, five dollars. There is a huge difference between someone who is interested in bikes in general, and someone who was already in your store earlier this month; the latter is much more likely to buy a bike from you.
At least two of the big online advertisers let you open a free account and play with these targeting options to your heart’s content; you can tweak the targeting options and see how many users you can expect to reach as a result and how much each click, view or other interaction will cost.
You could exactly see how much it would cost to show an ad about bikes to someone interested in bikes, of a particular age, and living in your city.
If online privacy is interesting to you, I recommend taking a look. It will make concrete for you a lot of what is going on behind the scenes—and help you to understand why there are dollar values attached to personal information, and what the important and meaningful factors are. You can check for yourself how much user information flows back to advertisers; it should be zero.
Learning about ads didn’t make me a fan of ads.
But at least, most of what I learned about ads made the picture better, not worse.
Advertising is always a four way deal, between the user, the content provider (Ross), the company that made the ad (Rachel) and the company with the ad infra (Big Tech). For the system to succeed it usually needs to deliver value, on average, to all four.
The user, unfortunately, gets the least pretty end of the deal—they are going to be annoyed to some extent by ads. But if they never clicked on the ads, there would be no ads. So there is, at least, some amount of balance.
I personally find some types of ads bother me and some don’t; still, it’d be nice to be able to opt out. Since I don’t interact with ads, all ads shown to me count as “bad ads”, and it would actually be an improvement for everyone to not show them to me. There should be an opt out button somewhere for that reason. As far as I can guess, there isn’t one is simply because there aren’t enough users who don’t click on ads for it to be worth having.
By the way, if you think ad companies dislike ad blockers, think again. If you hate ads, you are helping ad companies by installing an ad blocker. You weren’t going to click on the ads, so you increase the value of the ads that do serve by filtering yourself out. Bad ads benefit nobody; ads are not about annoying you.
At least, it’s usually the case that ads are not about annoying you.
When there is the possibility of “upgrading” to an ads free subscription or product, a wrong incentive is set up; now the ads do not have to have positive value, they can also make money by driving users to pay, so there is the possibility that they win by having negative value. Mobile games often use this pattern—that’s a rant for another day.
I glossed over some parts of the picture and there are plenty of big topics about online advertising that I didn’t go into at all; for example views vs clicks, and the importance and value of user intent.
Let me know if something in particular interests you and I’ll probably find time to write about it.
So, to the big question: could we end up with ads, and the associated tracking, on Gemini?
Fortunately I think the answer is a hard “no”, but it is worth looking at in detail. This post is already more than long enough, though, and I’m out of time to write today; so I’ll do that another day.
Until then, enjoy ad-free Gemini in the knowledge that it’s likely to stay that way. I’ll be doing the same.
This post was prompted by a thread on Station about my Identity post:
I responded directly there—I have a Station account now!—and thought I’d respond here in a different way, by starting to address the underlying and very valid concern. Hopefully my next post on the topic can be substantially reassuring.
So far today, 2024-11-25, feedback has been received 2 times. Of these, 0 were likely from bots, and 2 might have been from real people. Thank you, maybe-real people!
———
/ \ i a
| C a \ D n |
irc \ /
———