Affiliate SEO stopped being a craft of “build a site — get it to the top” a long time ago. The economics here are closer to buying traffic: there is a launch cost, a hit rate, and an asset lifespan. Below is how this pipeline works as of mid-2026: how much it costs to enter, where domains and links come from, what agentic models have changed, and which risks eat into the margin.
One Website Is Not a Strategy but a Lottery Ticket
The main mistake beginners make is investing in one project and waiting. Practitioners cite a figure: out of a hundred launched websites, seven or eight gain traction. “Gaining traction” here does not mean reaching the top-1, but entering the top-30, after which the asset can be pushed further with links and traffic. The remaining ninety hover at the bottom of the top hundred or never enter the index at all.
This is where the pipeline logic comes from. One market participant refers to a calculation based on a Google patent concerning algorithmic noise thresholds: for the statistics to become meaningful at all, at least fifty to sixty properties need to be launched. The calculation is someone else’s and cannot be verified, but the practical takeaway matches the observed hit rate: with fewer than fifty websites, you cannot distinguish a working scheme from random chance.
This also leads to a conclusion about further development. Resources are invested not in a favorite project, but in the one that has already shown results. Scaling up every launch equally is a sure way to spread the budget across ninety dead domains.
Launch Cost: What Agentic Models Have Really Changed
A couple of years ago, a pipeline required a layout designer, a CMS developer, and a webmaster. Today, the setup looks different: a subscription to an agentic development tool (Claude Code and similar tools, around 100 dollars per month), one specialized SEO professional, and one mid-level developer — not for layout work, but for architecture. Once the network reaches two hundred properties, you need your own search-results scraper and a management dashboard that will not collapse under the load.
The figures for a single website are roughly as follows:
| Item | Cost |
|---|---|
| Website generation (tokens) | around 1 dollars |
| Content, 1 pages | 3–3,5 dollars when generated by agents |
| Initial links | around 400 dollars |
| Pushing a site that has gained traction further | up to 10 000 dollars |
Developing a hundred properties costs a hundred dollars — an item that used to consume most of the budget has practically disappeared today. The money has shifted to links and domains. Total entry into the niche is estimated at around 50 thousand dollars, but the model becomes three times cheaper if links are built only for what gains organic traction. A testing reserve of less than twenty thousand is not worthwhile: the first batch may fail to get off the ground entirely.
This also settles the old debate about machine-written code. Rankings depend not on who typed the HTML, but on measurable parameters — rendering speed, layout stability, and time to first byte. A model will write both standards-compliant semantic markup and a tangle of inline styles: as requested. Poor, heavy code is equally harmful regardless of who authored it.
Domains: New Registrations vs. Drops
There are two approaches. A new registration is a fresh domain containing a keyword; these are usually bought in batches of a hundred with different formulations. A drop is someone else’s expired domain with accumulated link equity, sometimes still alive in the index.
At the start, new registrations are more sensible: they cost almost nothing while showing you what the search results look like and what the competition is actually about. Drops are a tool for those who already know what they are doing: prices range from 2 000 to 50 000 euros per domain, and the result remains probabilistic.
Drops are selected by parameters through specialized auctions and services such as ExpiredDomains and Spamzilla — the latter also shows web-archive snapshots, so you can see what the domain was previously used for. The working filters look like this:
- Live links — at least twenty for general tasks, at least a hundred for competitive niches.
- Time since expiration — no more than two years from when traffic actually fell.
- Language match — the content should have been in the language of the geo you are targeting.
- History — checks for trademarks and past content.
The history check is not a formality. Rebuilding a domain around someone else’s registered brand ends either in seizure or in the property being shut down. A separate category of problems involves domains whose former owner simply forgot to renew them: if they return and see someone else’s content at their address, the dispute will be with you, not the search engine.
The main risk of a drop lies elsewhere. A sharp change in subject matter is interpreted by the algorithm as abuse, and the accumulated trust is reset. There is a known case of a volunteer organization’s domain failing to get off the ground at all: the organization moved to a new address, the search engine linked the two entities, and treated the purchased domain as a satellite site. Approaches to changing the subject matter vary — some gradually add new sections, others upload everything at once to understand the asset’s fate faster, while the most aggressive teams override the subject matter with external links: ten thousand links with new anchors in a day, indexing, and a month later the domain works in a new niche.
Links: Where They Are Placed and How Many Are Needed
The range of approaches is enormous. Some place fifty thousand links at once on the principle of “if it holds, it will reach the top; if it doesn’t, no great loss.” Others believe that hundreds of live links for a geo are enough, and anything beyond that is just burning the budget.
The difference is explained by the asset’s lifespan. An expensive link from a strong property makes sense where the site is expected to live for years: review sites, major projects, and parasite placements. For a one-off property built around someone else’s brand, it is pointless — the site will die first, and the link cannot be moved.
Reddit has dropped out of the platforms used for mass link placement: moderation removes links there, leaving only mentions. Twitter, meanwhile, continues to work as a spam playground — five-dollar accounts live for years. One practical detail is worth remembering: it is better to write in English. Russia and India are the two largest sources of search spam, and content in Russian or Hindi by itself increases the likelihood of an additional review.
Parasite SEO and Its Downside
The scheme is simple: instead of building trust in your own domain for years, you place content on someone else’s strong domain. Formally, this is media partnership — you buy a section on a trusted publication with the right to publish an agreed number of pieces per month. The price for renting a section is said to range from several hundred to tens of thousands of dollars a month; major properties charge millions.
The advantages are obvious: placement takes a day instead of six months, the page starts on a powerful host, and antispam filters are much more lenient toward it. Links from such a page lead directly to the product rather than to an intermediary site — essentially, it is a landing page on someone else’s domain, where you can also send traffic from paid channels.
The problem is that this scheme is the most visible one in the industry, and penalties are imposed conspicuously. The segment’s largest player, the affiliate-project network Finixio (now Clickout Media), built its business precisely on parasite placements in well-known publications. After its methods were publicly analyzed, the properties it owned — including Techopedia, ReadWrite, and Business2Community — suffered a sharp drop in organic traffic and visibility. A model that delivers a multiple-fold acceleration increases the cost of failure by the same multiple.
HubSpot learned the same lesson in the white-hat segment. For years, the company bloated its blog with material about anything and everything for the sake of traffic — from résumés to real-estate-agent licenses. After the March core update and the 2024 antispam update, followed by the December one, the blog lost around 80% of its organic traffic: from 24,4 million visits in March 2023 to 6,1 million by January 2025. The analyses identify one reason — dilution of topical authority.
Behavioral Signals: Why Schemes Differ Across the Two Search Engines
According to practitioners’ estimates, Russian search results remain vulnerable to behavioral-factor manipulation: with a functioning bot farm, a query can be kept in the top results even with weak text on the page. Cases are cited of reaching the top-1 within minutes for branded queries and within a couple of days for commercial ones.
This does not work the same way with Google. The check happens at the moment of the click, on the redirect layer from the search results, where device fingerprints are collected. Among other things, practitioners list battery, camera, and SIM-tray identifiers — precisely what farms assembled from bare motherboards do not have. This can be bypassed only with real mobile farms: we are talking about tens of thousands of devices and local addresses, which puts the scheme out of reach for most people.
Instead of manipulation, viral traffic is used: thousands of real users from the target geo are sent to the page, the search engine records a spike in interest, and raises the document in the results within a few hours. The source is redirects from owned properties whose audiences have already passed antifraud checks. The method does not work universally: according to those who sell it as a service, it succeeds in six or seven cases out of ten, and works best on strong hosts. The cost of maintaining a top ranking is said to be around 1 200 dollars per month for developed markets and around 900 for less affluent countries.
What Happens When You Work With Someone Else’s Brand
Sites that collect traffic for someone else’s name live only until the brand owner notices them. Then complaints are sent to the registrar, and infrastructure hygiene becomes important: the network is not kept in a single account, otherwise everything will be de-delegated at once. Advanced teams have automated the migration — a complaint comes in, and the project appears on another CDN account within seconds.
There is a risk that is most often underestimated in this scheme. If traffic for someone else’s brand is directed to a licensed product, that product itself comes under attack: regulators have revoked local licenses, and market participants recall such cases in the United Kingdom and the Netherlands. The price of the issue is measured in millions, while the consequences are borne by someone other than the person who built the site.
It is also worth keeping in mind that practitioners themselves exclude two methods as unquestionably criminal — hacking other people’s websites to place links and denial-of-service attacks. Everything else in this area is governed by platform rules and trademark legislation, which differs in every jurisdiction.
What Changes With the Arrival of AI-Powered Search Results
AI answers are built on the same index, so the basic condition has not changed: to be cited, a document must be in the index. The specifics begin after that.
AI-powered search results really do take over informational queries — users read the answer and go nowhere. But models avoid sensitive verticals: no one will recommend an unlicensed product to a user because responsibility for the consequences would fall on the platform. For such niches, the familiar link-based search results will remain for a long time.
Affiliates face a separate problem: chatbots strip referral tags. Such traffic is useful to a direct advertiser, but an intermediary cannot monetize it — the conversion will not be credited.
As for appearing in answers, user discussions remain the key source. According to a June Semrush study of 150 thousand citations, Reddit accounted for 40,1% of links in language-model answers — more than Wikipedia and YouTube combined. The share fluctuates: for ChatGPT, it fell from nearly 60% of mentions to around 10% over a month and a half. Hence the guerrilla marketing on forums, where only a few survive out of four dozen accounts created.
Where to Start
- Calculate the unit economics before launch: affiliate-program payout, domain and link costs, and the expected hit rate.
- Build a pipeline, not a website. The benchmark is at least fifty properties; otherwise, the result is indistinguishable from chance.
- Start with new registrations. Drops cost several thousand euros and guarantee nothing.
- Check the domain’s history for trademarks and past content — it is cheaper than a dispute.
- Set aside a reserve for a second round of tests: the first batch may fail to get off the ground entirely.
- Don't spread the budget evenly—put more behind what has already delivered results.
- Distribute infrastructure across accounts and registrars in advance, not after the first complaint.
- Assess the risk to the end product: losing a license costs more than any earnings from traffic.
The niche has ultimately turned into a form of arbitrage where success depends not on the elegance of any individual solution, but on disciplined calculations and a willingness to write off ninety attempts out of a hundred.
