What are the economics behind web spam?

In this question, I'm referring to "web spam" on search engines (not email spam), such as placing misleading keywords on one's web page so that a pornographic site, for example, would get a lot of traffic that isn't necessarily related to what the site has to offer.

I am interested in understanding the dynamics, or rather, the economics behind this kind of activity, preferably with some numbers as examples or as data.

Links to studies are also appreciated.

3 Answers

I used to work for a major online advertising network, and it's pretty much an open secret in the industry that a substantial portion of ad impressions are either bot-generated, scams to exploit misleading search engine results, or both. There are typically several layers of "middlemen" involved between the company whose products are being advertised and the websites running the ads, and everybody has a financial incentive to keep their mouths shut about the rampant fraud involved.

The basic economics is this: an unscrupulous website joins an advertising network (which often have insufficient resources dedicated to their review/approval process for new websites joining their network) and then either purchases botnet traffic (at a lower price than they're receiving in ad revenue) or tries to pull in search engine traffic through misleading search terms, for as long as they can get away with it. They're typically caught fairly quickly, but by that time they may have made several hundred (or several thousands) of dollars in profit. Then they register a new scam website and repeat the process. Remember: websites are typically paid for the ads that appear on their website, even if nobody ever clicks on them.

Since nobody in the entire industry is really interested in exposing this aspect of the business, it may be difficult to find actual studies. I'll take a look though, and would suggest that "fraud in online advertising" is the appropriate search phrase.

Update: It looks like the Wikipedia article on Ad Fraud has a number of references that may serve as a good starting point. An online search turns up a good number of results as well, although many of them appear to be marketing material for software that claims to help prevent/stop ad fraud.

• Thanks Bill. I wonder why websites typically get paid for impressions rather than clicks. Surely this would incentivize the unscrupulous website to drive irrelevant or fake traffic to the ad-bearing page? It seems like a shift to PPC would make more sense, right? Although, I am aware that there is click-fraud. It just might make it harder for the unscrupulous party, that's all. May 31 '19 at 4:26
• Some of the companies that advertise largely just want their brand "out there" and are happy to see it plastered everywhere. They pressure their ad agencies to "reach" as a large an audience as possible, which is why those agencies then turn a blind eye to the fraud (because it helps them make their targets, as well) and so it goes through the chain of middlemen. May 31 '19 at 13:16

I am not a data person. So I can only provide insights into the theoretical incentives behind the phenomenon. Moreover I think that there are multiple forms of such web spam and each is rooted in distinct incentives. Hence, there is not a single true answer. Let me first express my view on classical (email) spam and relate it to what you have in mind. I will do this in this answer by using a baby model any undergrad could come up with. Then, I will point to search diversion incentives of an intermediary in a second answer building on the model by Hagiu and Jullien (RAND 2011). You (and the community) can then choose what fits best.

Classical spam

I guess the classical incentives behind (email) spam arise because the marginal cost $$c$$ is vanishingly small while benefits $$b$$ are comparatively large, $$c \approx 0, b>c$$. Although only extremely few people click the spam link, it can pay off to send a ridiculous amount of emails, because some do click and thereby end up on the homepage. The spammer maximizes $$(b-c)q$$ which is strictly increasing in $$q$$, the quantity of emails sent. Hence, she chooses the maximal possible amount $$q$$, i.e., send the spam to any address available.

The benefit $$b$$ here is that each visitor to the homepage may buy a good or generate ad revenue. In the first case, $$b=p_1 p_2 V$$, with $$p_1>0$$ being the probability of the average email reader actually clicking the link and $$p_2$$ the probabiliry of buying a (possibly fraudulent) good from the spammer (or transferring money to the Nigerian prince) and $$V>>0$$ being a large profit from this transaction. Since $$c \approx 0$$, spam pays even for extremely low probabilities $$p=p_1p_2$$.

Alternatively, the spammer is just an intermediary that leads possible customers to a seller's homepage. This intermediary then gets paid either per visit or per transaction. Both lead to the same model as above. The seller would be willing to pay the spammer some fixed amount $$x$$ per visitor with $$x \in [0,p_2V]$$ such that spammer and seller can agree on some spam benefit $$b \in (c,pV]$$, and the above model applies. If the seller pays some share $$s$$ of $$pV$$ per transaction, we can also arrive at the same model with $$b=spV>c$$.

Alternatively, the link just leads to a homepage full of ads. Then, the spammer gets paid per click on the ad or per second the ad is watched. Both is isomorphic to the model above with additional middlemen. Another alternative is that the link downloads malware, resulting in the same model with $$V$$ being the value from successful infiltration of the malware and $$p$$ being the probability of a successful undetected download.

Instead of $$q$$ being a quantity of emails sent it could be an additional keyword for search engines. Since adding one more keyword is quasi costless and attracts an additional number of expected visitors $$r>0$$, we arrive at a similar model with $$(rb-c)q$$.

This baby model abstracted away from a lot, but I think this is crucial incentives are rooted in very low marginal cost and the law of large numbers. Of course spam bothers people and engaging in it comes with a loss of reputation. For that reason, you do not see established firms doing it.

• HJ2011 look at the incentive of an intermediary to "diverting" the customer an undesired store first and then to the desired store to profit from "per-click" (or "per-visit") payments by the store. I type something about that another time. May 31 '19 at 12:45

This is an alternative explanation. Here, the "spam" due to "misleading keywords" on search engines does not work through the mechanism in my previous post. Instead, an intermediary platform diverts a customer to garner referral fees. Through your browsing history, the platform has good information about your taste. Through a recommendation mechanism, a search platform can direct attention to a particular product. Is this recommended product the one that suits your taste best? Hagiu and Jullien (RAND 2011) raise the point that it might not be.

Search diversion

Suppose there are only two products $$i \in \{A,B\}$$, and two types of consumers, one tends to prefer good $$A$$ and good $$B$$. The consumer incurs a cost $$c$$ for visiting each good-$$i$$ homepage/shop and can visit the sellers only sequentially. The value for the good is learned upon inspection. There is an intermediary who knows the consumers type and leads the consumer to the first shop. The intermediary gets an exogenous per-visit fee for each consumer she leads to a seller. Because of this incentive to generate traffic, the intermediary may like to divert the search process in the sense that it leads you to the least preferred product first. The authors also show that the diversion process is a mechanism to influence the seller prices on the platform.

Note that the consumer in this model actually likes both goods, and the valuations are independent draws from different distributions (the preferred good's value is drawn from a better uniform distribution). That is, this model does not apply to a setting in which you look for a hotel room and end up on a scam homepage that nobody values. Since the valuations are independent it also does not apply to a setting in which you look for adidas trainers and are diverted to the nike shop. More like, good $$A$$ are sunglasses and good $$B$$ is a watch.

Also note that the intermediary commits to the diversion process publicly before the consumer decides to use its service. If it is known that the intermediary likely diverts, then consumers with high search costs prefer not to use the intermediary.

• Speaking of search... when products and their prices are very easy to compare over retailers, demand is price-sensitive. Hence, retailers may want to "obfuscate" this process so that consumers do not learn so much about prices. I belive Elisson 2009 is the relevant citation here. May 31 '19 at 16:05