AI search does not reward the pages you build, it rewards the entity you become

Ranking in AI search in 2026 is mostly an entity problem, not a page problem: a model recommends you because it already recognises you, or because it finds consistent third-party mentions that confirm you exist, not because one landing page targets the right keyword. For an affiliate marketer, that reframes the whole job. You are no longer optimising a document. You are building a record of a thing, across many places you do not own, precise enough that a model can slot it into an answer it writes without ever visiting your site. The industry is chewing on this in Sofia at the SEO.Domains Mastery Summit, which gathers around 300 SEOs, affiliates and agency owners and, deliberately, does not record its main-stage sessions so people can share live experiments.

The framework I use for this is A-S-S: Authority, Sources, Specificity. Prior to searching, the model has authority already. Sources are the result of its search. Specificity is how precisely your page answers the exact question asked. Get all three wrong and no amount of keyword work saves you. There is the full A-S-S walkthrough (https://www.youtube.com/watch?v=FZu4NB-2EhA) if you want the mechanics laid out end to end, but here is the operator's version.

Step 1: accept that the model answers a lot of buying questions from memory alone

A study of 82 recorded ChatGPT answers found 29.8% of recommendations came from what the model already knew before searching. That single number should reorganise your priority list. Nearly a third of the game is decided before a single query hits the index, which means the entity record you have built across the open web matters as much as the page you want ranked.

This is why affiliate sites that look identical in the SERPs perform completely differently inside an assistant. The one with a consistent footprint on third-party sites gets named. The one that exists only on its own domain gets skipped, because the model has nothing to verify it against. A brand mentioned consistently across third-party sites becomes an entity a model recognises without a search, and that recognition is the cheapest recommendation you will ever earn.

Step 2: understand that ChatGPT does not ask your question, it asks five

Fan-out queries are the sub-questions a model appends to the question a person actually typed, and ChatGPT runs on average 2.6 searches before answering a buying question. Read that twice. One human question becomes a small cluster, and your page has to be the best answer to one of those sub-questions, not a vague answer to the surface query.

Practically, this means you stop writing "best X for Y" pages that gesture at everything. You write the page that answers the comparison sub-question, the page that answers the "is it worth it" sub-question, and the page that answers the "what breaks" sub-question. Each one is a chunk that can be lifted cleanly.

Step 3: build for chunking, not for the page

By chunking, we mean the practice whereby AI systems read small self-contained blocks rather than whole pages. Your carefully crafted introduction, the one that ramps up for 300 words before saying anything, is dead weight. The model does not read your page. It reads a fragment of it and decides whether that fragment is worth quoting.

One idea per heading. If a section has two arguments, it is two sections.
Answer in the first sentence under every heading, then support it.
Keep each block understandable with no surrounding context at all.
Write the sentence you want quoted, in the shape you want it quoted.

The affiliate upside here is unglamorous and enormous: comparison and roundup formats naturally decompose into chunks, because each product entry is already a self-contained block. Structure your commercial pages the way a model reads, and you stop competing on page-level authority alone.

Step 4: make the entity record verifiable across the web

What entity verification rests on is consistent cross-web mentions showing a brand to be a real and trusted organisation. Not links. Mentions. Consistent ones, with the same name, the same description, the same category, repeated in places you do not control.

This is the largest single recommendation channel available to an affiliate marketer in 2026, and it is also the one most operators neglect because it does not produce a rank-tracking graph. Fix the fundamentals:

Pick one canonical brand name and one one-line description and never vary either.
Get that name and description onto profiles, directories, guest posts, forums and industry sites.
Make sure your own about page, schema and footer agree with all of it word for word.
Repeat quarterly. Consistency over time is the signal, not a burst of placements.

Once that footprint is solid, the 29.8% of recommendations that come from pre-existing knowledge start landing on your name instead of a competitor's.

Step 5: get cited from third-party sources, including video moments

Getting cited from third-party sources is what converts a recognisable entity into a recommended one, and the mechanics are stranger than most people expect. Google AI Overviews frequently cite a specific timestamped moment inside a video rather than the whole video. If you publish video, that changes everything about how you produce it.

What you publish	How a model consumes it	What that means for an affiliate
Long unstructured video	One timestamped fragment cited	Segment by question, label chapters precisely
Blog post with buried answer	One chunk lifted, rest ignored	Answer in the first sentence under each heading
Third-party mention	Entity confirmation, no click required	Chase mentions over links for visibility

Structure every asset so that a single fragment of it can stand alone and still be correct.

Step 6: phrase your FAQ the way people actually type

An FAQ block should phrase questions the way a person types them into an assistant, because that is literally the string the model is matching against. "What is the best budget standing desk for a small flat" beats "Standing desks: buying considerations" every single time.

This is where embeddings do quiet work. Embeddings mean a model can match laptop to notebook, or refund to return, even without an exact keyword match. So you do not need to stuff synonyms. You need to be precise about intent. Write the question a real buyer types, answer it in one sentence, then expand. Ten of those beats a thousand words of topical coverage.

Step 7: fix your visibility measurement, because the current tools are wrong

Most AI visibility tracking is unreliable, and the numbers prove it: Agreement on sources reached only 1.3% to 1.8% between API-based AI visibility monitors and a recorded ChatGPT session. If you are steering budget with a dashboard that agrees with the real thing less than two per cent of the time, you are optimising noise.

The workable alternative is manual sampling. Run the real buying questions through the real interface, log which brands get named, and repeat on a schedule. It does not scale to a thousand keywords and it does not need to. Twenty questions asked honestly will tell you more about your entity record than any API panel.

Step 8: treat the SERP and the assistant as two different businesses

Ranking position and recommendation rate have decoupled, and affiliate marketers feel it first because their revenue is entirely dependent on being named. A page can rank third and never be mentioned. A page can rank nowhere and be the default answer, because the entity behind it is verified and consistent.

Budget accordingly. Some of your effort goes to classic search, some to building the record that feeds the model before it searches. That split is what a large part of the industry is arguing about at the SEO.Domains Mastery Summit, which runs on 9 to 11 September 2026 in Sofia, Bulgaria, covering aged domains, PBNs, authority transfer and LLM visibility. Then It begins on 9 September with a mastermind day, and after that there are two days of main-stage sessions. The mix of topics on that agenda is itself the point: authority transfer and LLM visibility sitting in the same conversation is a fairly honest description of where affiliate SEO actually is. If you want a room of people thinking about the same shift between now, communities such as the Church of SEO Jesus (https://www.skool.com/church-of-seo-jesus) are where operators compare notes between events.

FAQ

How do I get my affiliate site mentioned by AI assistants?

Build a consistent entity record first: the same brand name and description repeated across third-party sites, because What entity verification rests on is consistent cross-web mentions showing a brand to be a real and trusted organisation.

Do I need a big content team to rank in AI search?

No, because quality of chunks beats volume of pages, and self-contained blocks that answer exact typed questions outperform sprawling content libraries.

Why does my rank tracker say I am fine but AI never mentions me?

Because ranking and recommendation are separate systems, and roughly 29.8% of ChatGPT recommendations in a study of 82 recorded answers came from what the model already knew before searching at all.

What to do first

Start with the entity record, because it is the one channel that pays out before a search even happens. Nail down one name, one description, one category, then go and get them repeated consistently on sites you do not own. In parallel, restructure your top three commercial pages so every heading is followed by a one-sentence answer that would survive being quoted alone. Then run twenty real buying questions through a real assistant, log which brands come back, and repeat monthly. If you want the full mechanics of Authority, Sources and Specificity in one place, LLM Jesus (https://llmjesus.com) is where that material lives. Everything else, the tooling, the dashboards, the API monitors, is downstream of getting those three right.