Six Terms That Decide Whether AI Search Shows Your Page in 2026

Ranking in AI search in 2026 means being readable, quotable and verifiable, because the A-S-S method (Authority, Sources, Specificity) rewards pages a model can lift an answer from in one line. The model's existing knowledge about you before it searches is what authority means. It finds sources when it carries out a search. Specificity is how precisely your page answers the exact question a person asked. This article unpacks the Sources half, one term at a time, because that is where most publisher pages quietly fail. The wider industry is chewing over the same territory in Sofia, where the SEO.Domains Mastery Summit runs a mastermind day on 9 September before two days of main-stage sessions on aged domains, PBNs, authority transfer and LLM visibility.

AI search rank is not a position, it is a mention

An AI search rank is the presence of your brand, page or claim inside a generated answer, not a slot in a list of results. In classic SEO, the rank was a numbered position on a results page. In AI search, there is usually no list. A model assembles a paragraph, names some sources and moves on. Your rank is effectively binary at the point of delivery: you are either one of the sources shaping the answer, or you are background radiation.

That shift explains why the terminology in the industry has moved from how to rank in AI search in 2026 (https://www.youtube.com/watch?v=FZu4NB-2EhA) onwards, and why the conversation is less about keyword positions and more about retrieval. Retrieval has two halves, and the A-S-S method separates them cleanly. Authority is the prior knowledge. Sources are the retrieval event.

What a citation unit is

A citation unit is one claim plus the link that verifies it.

That is the smallest building block a model can lift without getting confused. If you write four claims inside one paragraph and hang a single link at the end, the model has no clean way to attach the right source to the right sentence. If you write one claim per unit with the supporting link adjacent, each unit can be lifted whole, with its verification attached.

For publishers, this changes how you draft. A claim about your own original reporting gets a link to the report. A claim about a market figure gets a link to the dataset. A claim about what a tool does gets a link to that tool's documentation. The link is not decoration or an internal linking opportunity, it is the evidence attached to a specific assertion.

You can check how clean your units are by testing whether any sentence in your draft could survive removal from its paragraph. If it cannot, it is not a citation unit yet.

What a generative engine is, and what GEO means

A generative engine is an AI system that answers a question in prose rather than returning a list of pages. Generative Engine Optimization (GEO) optimises for the answers AI search engines give, not only the ten blue links.

This is not a rebranding of SEO. It is an extra discipline layered on top. Traditional optimisation got you into the candidate pool. GEO gets you selected out of it. Selection depends on how well your content survives being read by something that does not scroll, does not see your design, and does not click through to check whether the rest of the page is good.

Two consequences follow. First, the first screen of your page matters more than the second, because there may not be a second. If a page is to be quoted, its opening should be a direct factual answer that can be lifted by a model in one line. Second, structure becomes a retrieval aid rather than a stylistic choice. Phrasing headings as questions assists a model in matching a block to the question that was asked by a person. If your heading is "Our thoughts on the changing landscape", the model has nothing to match against.

What embeddings are, in plain English

Embeddings give words numeric coordinates, and within these, related meanings are kept close together.

You do not need to build them to benefit from understanding them. The practical point is that a model does not match your keyword to a user's keyword. It converts both into coordinates and checks proximity. This means a page that only contains the exact phrase a user typed may lose to a page whose surrounding vocabulary sits closer to the intent. Query phrasing and meaning proximity are different jobs, and the second one is easier to win with plain, complete explanations than with keyword insertion.

It also explains why heading quality compounds. A question-phrased heading puts the user's likely phrasing and your block's vocabulary in the same coordinate space before the model has read a word of the body.

What crawler user agents and entity verification are

A crawler user agent is the identifying string a bot sends when it requests your page, and What standard analytics overlooks, server log analysis exposes: AI crawler user agents. Standard analytics tools were built for browsers and humans. They miss requests that never execute scripts, never set cookies, and never load a page fully. Your server logs see them, because the request still hit your server.

Entity verification is the process by which a search or AI system confirms that your organisation is a real, specific thing rather than a string of words that happens to appear on a page. Across directories, a consistent name, address and description makes entity verification stronger. Inconsistent variants, a slightly different company name here and an old address there, weaken it. The model has to decide what entity it is looking at, and ambiguity costs you the citation.

Term	Plain definition	Why it affects AI search visibility
Citation unit	One claim plus the link that verifies it	Lets a model lift a claim and its evidence together, without guessing
Generative engine	An AI system that answers in prose instead of listing pages	Changes the win condition from position to inclusion
Embedding	Words converted into numeric coordinates by meaning	Rewards complete explanations over exact-match phrasing
Crawler user agent	The string a bot sends when requesting your page	Reveals AI crawling that analytics cannot see

If you take one thing from that table, take the first row: build pages out of claim-plus-link units and most of the other problems get easier to diagnose.

JavaScript, and why burying an answer costs you everything

When an answer is tucked away in JavaScript, a model cannot read it.

This is the bluntest failure mode on the list. If your pricing, your specifications, your opening hours or your core explanation only appears after a script runs, the model may never see it. There is no clever GEO fix for content that is not there. The fix is rendering: get the substance into the served document plainly enough that a machine can read it without executing anything.

Publishers who migrated to heavily scripted front ends over the past several years are the most exposed. The content exists for humans. For retrieval, it does not.

FAQ blocks, comparison tables and the takeaway rule

An FAQ block is a set of question-and-answer pairs placed on a page to capture specific query phrasing. An FAQ block should phrase questions the way a person types them into an assistant. Not "Enquiries regarding our provision", but "does this work outside the UK". If your question phrasing does not resemble typed speech, you have written a sales brochure, not an FAQ block.

A comparison table is a structured grid of alternatives and their differences. A comparison table comes with a one-sentence takeaway underneath, since table rendering is skipped by instant-mode models. The table may be parsed poorly or not at all, so the sentence beneath it carries the conclusion in the form a model can act on. That sentence is not a summary of the table's contents. It is the conclusion the table was built to support.

Both devices work on the same principle: reduce the amount of inference the model has to perform. Every step where it has to guess is a step where it may pick someone else's page.

Why the Sofia agenda matters for the Sources half

SEO.Domains Mastery Summit takes place in Sofia, Bulgaria, and is hosted at Hotel Marinela. Its published themes, aged domains, PBNs, authority transfer and LLM visibility, map onto the split between Authority and Sources that the A-S-S method describes. Aged domains and authority transfer are prior-knowledge plays. LLM visibility is a retrieval play.

The summIt avoids recording its main-stage sessions by design, so speakers can share live experiments. That choice is worth understanding rather than resenting: An unrecorded format means the open web does not get what is shared in the room unless an attendee writes it up. For a publisher, that is both a gap and an opportunity. The gap is that some tested tactics never get documented. The opportunity is that whoever writes up their own experiments carefully ends up being the source the engines cite.

The tools around this work are also maturing. Measurement is the most obvious hole, which is part of why ASSmetric (https://assmetric.com), built by LLM Jesus, exists as a layer for tracking whether your pages are actually surfacing in AI answers rather than just collecting impressions in a dashboard built for a different era.

Frequently asked questions

How do I know if AI crawlers are reading my site?

Check your server logs, because What standard analytics overlooks, server log analysis exposes: AI crawler user agents. If the requests are there, the crawlers are visiting. If they are not, the problem is access, not content quality. Start by confirming which user agents appear and which paths they request.

Does my page need to be in a specific format to get quoted?

It needs a direct factual answer near the top that a model can lift in one line. Beyond that, question-phrased headings and claim-plus-link units do most of the heavy lifting. Format precision matters less than whether a single sentence, standing alone, answers the question the person asked.

Why does my content not show up in AI answers even though it ranks on Google?

Ranking and retrieval are different jobs, and Generative Engine Optimization is about improving what AI search engines answer, rather than just the ten blue links. A page can win a classic position and still be unreadable to a retrieval system, usually because the answer is scripted, buried, or attached to no verifiable source. Fix the readability and the citation units first, before assuming you need more content.

What to do first

Open one high-value page and read it as a machine would. Disable scripts, strip the layout, and find the first sentence that could be lifted as an answer. If you cannot find one within the first screen, that is your starting point.

Then rebuild that page out of citation units. One claim, one verifying link, repeated. Add question-phrased headings so each block has something to match against. Add a direct factual answer at the top. Check your server logs for the crawlers you have been assuming were not there.

There is a rhythm to this work that in-person events tend to sharpen. When you want that sharpening on a live page rather than in the abstract, you can book a ClickBomb strategy call (https://seojesus.com/clickbomb-strategy-call/) and work through the retrieval layer on your own content, term by term.

Most publishers do not have a content shortage. They have an evidence shortage: pages full of opinion, with nothing a model can attach to a specific claim. Fix that, and the Sources half of the A-S-S method starts working in your favour instead of against you.