AI Citations Are the New Backlinks

A backlink pools authority; an AI citation hands over the answer

The difference between the two prizes is mechanical, not philosophical. A backlink is a vote stored in a graph. Every link Google has crawled feeds one shared scoring system, the descendant of PageRank, and the authority a page banks there travels with it into every query Google ranks. Earn enough good links and you lift across the whole index at once. An AI citation works the other way. Ask ChatGPT, Claude, Gemini, or Perplexity to name the best roofer in Hamilton and the engine consults no stored leaderboard. It fetches a small set of pages it can reach and trust at that moment, reads the business names inside them, and writes its answer from those pages. The prize is not a rank position. It is your name appearing in the sentence the user reads instead of a results page.

That shift rewrites off-site work. Under link equity, the question was how much authority you had banked. Under citation equity, it is whether the specific sources a given engine grounds against happen to name you, right now, for this query. Authority still helps a page get fetched and trusted, but it no longer pools into one number that carries everywhere. Formative Digital measured how far that pooling has broken down, and the figure is starker than the consensus coverage suggests.

Why are AI citations becoming the new off-site currency?

Citation equity matters now because the answer, not the blue link, is increasingly where the decision gets made, and a growing slice of that audience is Canadian. CIRA's 2025 Canadian Internet Trends report, a national survey of 2,000 adults, found 33% of Canadians used generative AI tools in the past year, double the 16% recorded in 2024, and that 47% of those users employ AI as a search engine. When nearly half of AI users treat the tool as search, the surface that decides who gets considered is the cited answer. A business that ranks on Google but is named in none of the four AI answers is invisible to that share of demand, however many links it holds.

Links alone cannot carry you into that answer, for a mechanical reason. The peer-reviewed GEO paper by Pranjal Aggarwal and co-authors (arXiv:2311.09735) showed that source-side optimizations, citing your own sources, adding quotations, adding statistics, can raise a page's visibility inside generative-engine responses by up to 40%. The levers that earn a citation sit at the content and fact level, not in a link count. Kevin Indig's February 2026 Growth Memo analysis adds position: about 44% of all AI citations were drawn from the first 30% of a page's text, so the answer has to be front-loaded to be selected. Where SEO let you buy authority and spend it everywhere, citation equity is earned inside the material an engine reads, and earned again on each engine that reads different material.

What did 1,732 real AI citations reveal about citation equity?

They revealed that citation equity does not pool: 83.7% of every source our four engines cited appeared inside one engine and nowhere else. We ran a DataForSEO scrape against the live answer engines in May 2026, 176 successful queries and 1,732 individual citations, every one a real local prompt of the form "who are the best {vertical} in {city}, Ontario," across nine Ontario cities and five verticals. For the cross-engine view we widened to the full set of 583 distinct domains over 44 city-and-vertical cells, then asked the database one plain question: how many of those domains did two or more engines cite? The answer was 95. Ninety-five out of 583 is 16.3%; the other 83.7% lived in a single engine's answers only. A link earned on a strong domain helps you everywhere Google looks. A citation earned in one engine's source layer helps you, on this evidence, almost nowhere else.

That first-party number is what separates this from the abstract coverage, which argues the link-to-citation case with generic figures and stops at the definition. The data says the shift is not one trend but four separate contests. Backlink authority resolves into one PageRank-style graph, so off-site work compounds into a single asset; citation equity splits across four grounding layers that barely touch, so the same work is done four times against four different source sets. The job moves from "build links" to "earn presence inside each engine's distinct grounding set."

One honest limit: the 16.3% overlap uses a generous bar, cited by two or more engines. Raise it to three or more and the shared core shrinks to a few aggregators, mainly homestars.com and opencare.com, the rare domains where a citation behaves a little like a backlink and carries across engines. Everything else is single-layer. We unpack the divergence itself in our companion study measuring how little the four engines actually agree on, the cornerstone this off-site argument sits on.

Why does each AI engine cite a different web?

Each engine cites a different web because each company wired its retrieval to a different source layer, and those choices show up cleanly in which domains get named. This is where the link-versus-citation difference stops being abstract: you cannot earn citation equity in general, only on a specific layer, and the four layers want different things. Here is the grounding fingerprint our scrape recorded for each engine, and the off-site move it implies.

Read those four together and the off-site implication is unmistakable: no single backlink-style deposit lifts you across all four. A Google Business Profile fix moves ChatGPT and barely touches Claude. A directory placement moves Claude and does nothing for Gemini's Vertex pipe. Schema work feeds Gemini and is invisible to Perplexity's review breadth. The gap that separates this from old-style SEO, where one ranking did not equal one citation, is laid out in our piece on where GEO and SEO part ways. Citation equity is portfolio work across four channels, not one balance you grow and spend everywhere.

How do you earn citation equity instead of link equity?

You earn it by winning a place in each engine's preferred source layer at once, rather than banking one stock of authority and expecting it to travel. The four layers are known and stable, so this is concrete work. The GEO paper found targeted content optimization lifts generative-answer visibility by up to 40%, and that the effective tactics are domain-specific. Two of Formative Digital's 12 Vectors carry most of it. Vector 5, Cite, earns placement in the third-party sources each engine trusts. Vector 10, Localize, makes your local entity unambiguous so the same business resolves cleanly whether it is read through Google's Knowledge Graph, a directory, or a review aggregator.

Measurement is where the old link habit does the most damage. If 83.7% of cited sources sit inside one engine, checking "do I appear in ChatGPT" says nothing about the other three, and a single check on a single day says little even about that one, because the engines wobble run to run. Track all four as four separate scores, not one blended AI rank, and sample each several times to average out the noise. Watch the source layers too, since the layer leads and the citation lags: win a spot on threebestrated.ca and expect Claude to move first. The honest first step is a baseline read of which engines already surface you, our Vector 1. We run the four layers in parallel through the Formative Forces, our orchestrated multi-agent system, rather than chasing one engine at a time. One caveat, since this is YMYL work owners spend real money on: outcomes depend on your industry, your competition, and your existing digital presence.

Do backlinks still matter when the prize is being cited?

Backlinks still matter, but in a narrower role. Links remain part of how a page becomes trustworthy enough to be fetched and grounded, so a well-linked page is likelier to enter an engine's candidate set. What links no longer do is finish the job. Being in the candidate set is not being named; the naming is decided by the content-level and directory-level signals above. Authority still buys the ticket to the door. Citation equity is what gets you through it, and it is bought separately at four doors.

A brand mention is not quite the same prize as a citation, and the difference is practical. A citation is a named, often linked source the engine grounds a claim against; a brand mention is your name in the generated text whether or not a source is attached. Search Engine Land's reporting argues the off-site target is now both, since mentions across the web, not only backlinks, drive combined LLM and SEO visibility. In our data the two travel together, because the directories and profiles that get cited also carry the mention, but a mention without a grounded source is the more fragile half.

"For twenty years off-site SEO meant one thing: bank links, and the authority follows you everywhere. That model is over for AI answers. We scraped 1,732 real citations across nine Ontario cities and 83.7 percent of what the engines cited appeared in only one of them. Citation equity does not pool the way link equity did. You earn it four times, on four different source layers, and you measure each one on its own terms. That is not link building with a new name. It is engineering presence into four separate retrieval systems, and it is the whole reason a well-built local business can be named in four answers while a competitor with more backlinks is named in none."

Matt Griffin, Founder, Formative Digital, Brantford, Ontario