Generative citations are not dofollow, and not nofollow either, because the question imports a concept from an economy they simply do not participate in. Dofollow versus nofollow describes whether a hyperlink on a crawled webpage passes ranking signal through the link graph; a citation in a ChatGPT answer or an AI Overview is not a node in that graph at all, it is a source attribution rendered in an interface, generated per conversation, usually uncrawled and unindexed as a linking document, and carrying no PageRank-style equity in either direction. Asking whether it is dofollow is like asking whether a librarian’s verbal recommendation is dofollow: the frame does not apply. What the question is really reaching for is worth answering properly, though, because citations do carry value, several kinds, some of them worth more than the link equity they lack, and understanding what actually flows through a citation changes how you invest in earning them.
The short version of the real answer: citations pass traffic, trust, brand memory, and second-order links, each on its own mechanics, and they are the KPI of a different economy, one that runs beside the link graph rather than inside it, interacting with it at exactly two points worth understanding.
Why the dofollow frame fails mechanically
Link equity exists because search engines crawl pages, find hyperlinks, and treat them as endorsements flowing through a persistent, recomputable graph. Every single part of that machinery breaks for generative citations. The “page” containing the citation is an ephemeral answer, generated for one user, not crawled, not indexed, not part of any graph an algorithm recomputes. The link is interface furniture, a source chip, a footnote panel, a reference list, whose markup semantics are irrelevant because no crawler is parsing the answer to extract endorsements. And the direction of meaning differs: a hyperlink says “I, this page, vouch for that page,” while a citation says “this answer was assembled from that source,” provenance rather than endorsement, evidence rather than a vote, a bibliography entry rather than a ballot.
The one partial exception proves the rule neatly: AI Overviews live inside Google’s own results page, and their links are real links on a rendered surface, but they function as SERP features, like the links in any search result block, not as equity-passing editorial hyperlinks from a ranking document. Nobody asks whether a blue-link search result is dofollow, because everyone understands results are outputs of ranking, not inputs to it; overview citations sit in the same category, outputs of the answer layer, not votes within the link layer.
What actually flows through a citation
| What flows | Mechanism | How to capture it |
|---|---|---|
| Referral traffic | Users click sources to verify and go deeper | Land them on the exact cited claim, fast |
| Trust transfer | Being the named source reads as authority to the human | Consistent presence across the category’s questions |
| Brand learning | Names seen in answers return as brand searches later | Watch brand-query and direct-traffic trends |
| Second-order links | Writers discover you via answers, then link classically | The only path back into the link graph, and a real one |
| Compounding presence | Cited sources get re-retrieved and re-cited | Stay current; presence begets presence |
Two of those rows deserve expansion because they get systematically undervalued. Referral traffic from citations is small in volume and unusual in quality: the clicker is verifying or deepening, mid-task, pre-framed by the answer, and conversion behavior on these visits runs well above ordinary informational traffic, which several teams discover only after they segment the referrers properly, including the attribution quirk that Perplexity clicks often arrive as direct traffic. Click behavior on answer surfaces overall is genuinely reduced, Pew’s data shows users click less when summaries appear, which makes the clicks that do happen more selected, not less valuable.
Second-order links are the quiet bridge back to the old economy: journalists, bloggers, and researchers increasingly do their discovery through assistants, and the source an answer cites is the source their eventual article hyperlinks, dofollow, on a crawled page, in the classic graph. A citation is not a link, but it sits reliably upstream of links, which means citation presence feeds link equity on a delay, through human intermediaries, and brands tracking both often watch referring-domain growth follow answer-presence growth by a quarter or two.
Citations as their own economy, with their own ledger
The practical upgrade is to stop translating citations into link-era units altogether and account for them natively, in their own books. The citation economy’s ledger: citation-rate per tracked question (how often your domain appears among an answer’s sources), named-rate (how often your brand appears in the answer text, a related but distinct achievement), the spread of questions on which you are cited, and the stability of citations across engines and re-samples. Those are the units that actually respond to your work, and they are measurable with the standing methodology, a fixed question set, sampled monthly, per engine, with sources logged, that runs through whether ChatGPT citations are worth tracking and the whole measurement discipline around it.
What earns entries in that ledger is also its own craft, related to but distinct from link building: extractable, verifiable, current content in the formats language models actually lift, on pages the engines can read, with claims specific enough to be worth attributing. Link building earned votes by relationships and worthiness-to-endorse; citation earning is worthiness-to-quote, and the difference shows in what works: original numbers, plain-stated facts, and answer-shaped structure out-earn the guest-post circuit completely, because the machine is sampling for evidence, not counting endorsements.
The two economies do interact at the selection stage, which keeps classic SEO relevant to citation work: engines’ retrieval layers lean on search-style ranking to pick candidate sources, and analyses like Ahrefs’ study of most-cited domains in ChatGPT show citation flowing disproportionately to established, authoritative, well-ranked properties. Authority built in the link economy buys candidacy in the citation economy; the citation economy then judges the content on quotability. Neither substitutes for the other, and the strongest competitive positions hold both at once.
A quarter of ledger-keeping shows the dual economy in motion. A data-tools company starts tracking fourteen category questions in January: citation-rate near zero, named-rate low, rankings mid-page-one for most. February’s format surgery, answer-first openings and original benchmark tables on six pages, moves nothing in rankings and starts appearing in the citation logs by March: four questions now sometimes cite the benchmark pages, exactly the pages with numbers nobody else published. April brings the second-order effect on schedule: two industry newsletters, both of whose authors research with assistants, link the benchmark study, real dofollow links on crawled pages, and the rank tracker finally twitches. By May the brand-query trend line has a visible bend, and the leadership deck can tell the story in the correct causal order: quotable content earned citations, citations earned human discovery, discovery earned links and brand searches, and the link-era metrics moved last, as effects rather than causes. Read backwards through link-era glasses, the same quarter would look like unexplained ranking luck; read through the citation ledger, every step is attributed.
What this means for how you spend
The budget implications are concrete and immediate. Chasing citations for equity is chasing a payout that does not exist, so any vendor pitch implying AI mentions boost your rankings through some link-like mechanism deserves the skepticism owed to category errors, and there are such pitches circulating. Chasing citations for their real yields, selected traffic, trust, brand learning, second-order links, prices the work correctly: it is content and evidence work, not outreach work, and its ROI shows up first in the citation ledger and then, on a one-to-two-quarter lag, in brand queries and referring domains, in that order, every time.
The measurement stack should reflect the dual economy: classic rank and link tracking for the selection layer, and the tracked question set for the citation layer, read together every month, because divergence between them is where the diagnosis lives, well-ranked but never cited points at quotability problems on pages that already have the authority, while cited but poorly ranked points at niche authority the link economy has not caught up with yet, both of which are specific, fixable states rather than mysteries. SQSEO covers the citation half free, pairing the longtail question research that finds what your buyers ask with tracking of what the engines answer and cite, which is exactly the ledger this economy needs, and the divergence reads fall out of putting its data next to your rank tracker’s.
And one defensive note completes the picture: because citations carry trust to humans reading the answer, being cited for wrong or stale claims is negative value flowing through the very same channel at the very same strength, which is why the freshness and accuracy disciplines are citation-economy hygiene rather than mere content maintenance. The source an answer quotes into an error loses standing with both the audience and, over re-sampling cycles, the engine; the ledger runs in both directions, and it never stops posting entries.
Frequently asked questions
Are generative citations dofollow?
No, and the frame does not apply: dofollow describes ranking signal passing through hyperlinks on crawled pages in the link graph, while generative citations are source attributions rendered in ephemeral answers, uncrawled, unindexed, and outside any equity graph. Even AI Overview links, which are real links on Google’s results page, function as SERP-feature outputs rather than equity-passing endorsements. Citations carry other value, selected traffic, trust, brand learning, and second-order links, on their own mechanics.
Do AI citations help SEO at all?
Indirectly and genuinely: citation presence drives discovery by the humans who create actual links, writers, researchers, roundup authors, so referring-domain growth often follows answer-presence growth on a delay; cited-source traffic converts unusually well; and names learned in answers return as brand searches, which are classic positive signals. What citations do not do is pass link equity directly. The two economies interact at retrieval, where search-style authority buys citation candidacy, so classic SEO and citation work reinforce rather than replace each other.
What is the difference between being cited and being named in an AI answer?
Cited means your domain appears among the answer’s sources: the engine used your page as evidence. Named means your brand appears in the answer’s text: the engine considers you part of the answer itself, a recommendation-level achievement. They respond to different work, citations to quotable, verifiable, well-formatted content; naming to entity strength and category consensus, and tracking them separately per question per engine is what makes the difference actionable rather than blurred into one score.
How do I measure the value of AI citations for my brand?
Run the citation ledger: a fixed set of your category’s real questions, sampled monthly per engine, logging citation-rate, named-rate, and the cited sources, trended over quarters. Pair it with the capture metrics, assistant-referred traffic segmented in analytics (minding attribution quirks), brand-query trends, and referring-domain growth on a lag. SQSEO runs the ledger half free, pairing the question research with the answer tracking, so the citation economy gets the same instrumentation the link economy has had for decades.
Can I buy or request citations in AI answers?
No mechanism exists for either: citations are selected by retrieval and synthesis from what the engines can read and verify, and no engine sells placement in its sources. The earnable path is quotability, extractable, current, verifiable content with specific claims worth attributing, on crawlable pages, plus the authority that wins retrieval candidacy. Anything marketed as guaranteed AI citations is either ordinary content work in costume or nothing, and on-page tricks aimed at manipulating answers are treated as adversarial by every platform that catches them.
