Citation share is the percentage of AI answers, across a fixed set of queries, that name your domain as a source. ChatGPT, Perplexity, and Google's AI answers each attach a short list of credited domains to whatever they compose, and your slice of those lists is the closest thing to share of voice the answer engines have produced. The number is real. Watching it makes sense.
The part the tracking vendors leave out is how the number behaves. Citation share moves the way a search ranking moves: an output that settles wherever your inputs put it, on the engine's schedule.
The input this number reads is provenance, meaning where the figures on your pages actually come from. We traced 1,006 SaaS blog citations to the end of their reference trail to see where that resolution lands, and only 17.2% reached a primary source. Most pages carry numbers that belong to someone upstream, and a page full of borrowed numbers has already told the engine who deserves the credit.
What Citation Share Means in AI Search
Take a set of queries your buyers ask. Run each one through an answer engine and record every domain the answer cites. Citation share is your cited answers divided by all cited answers, measured against every other domain that got named:
citation share = (your cited answers / total cited answers) × 100
Two properties matter for reading it honestly. It's a percentage of a fixed query set, so changing the queries changes the metric, and comparisons only hold while the set stays put. And it has to be sampled per engine, because ChatGPT, Perplexity, and Google's AI answers each assemble their source lists differently. Within answer engine optimization, citation share is the outcome the whole discipline points at.
Citation Share vs Share of Voice
Share of voice added up your presence across a channel: paid impressions, organic positions, social mentions. Citation share counts something one layer deeper, the times an engine named you as a source it trusted while composing a reply. Presence earned you share of voice. Credit earns you citation share.
The two metrics share a failure mode too. Teams learned to treat share of voice as something budget could lift, and for years it could, because impressions were for sale. Nobody sells the position of being where a number originates. By the time an engine composes an answer, the question of whose page a figure traces back to was settled when the pages were written.
Citation Share Works Like a Rank
You already know how to live with this kind of number. In 20 years of SEO you never once typed yourself into position three. You changed titles, earned links, sped up pages, and the ranking moved later, wherever the algorithm's read of those inputs put you. The dashboard number was a report on work that happened somewhere else.
A citation-share dashboard is the same species of report aimed at a newer surface. It counts how often past answers credited you. The next answer will credit whoever holds the strongest position in the reference chain behind the claim being answered, and no dashboard reaches back into that chain. Vendors lead with the thermometer because a thermometer demos well. The percentage moves when provenance moves.
Provenance being the input raises the question of what you currently read. How do you check whether AI search is citing you?
Poll: How do you check whether AI search is citing you?
Only one of those five checks runs before the engine composes anything. The other four read a result the chain already decided.
A tracker, a referral chart, a manual spot-check all read the scoreboard after the engine has already chosen. The one method that touches what the scoreboard reads is checking whose numbers sit on your page in the first place. As readers weigh in above, the split between watching share and owning what produces it will come into focus.
What Decides Citation Share
Behind every statistic on the web sits a trail: the page that cited the page that cited the measurement. An engine assembling an answer walks that trail and tends to credit the page where it ends. Restate a figure that lives elsewhere and you become one of the hops, a page the engine passes through on its way to the source. The mechanism has its own writeup in why ranking does not equal citation; the working rule is that credit lands where the trail bottoms out.
Entering the trail at all still runs through retrieval. When Ahrefs studied where cited URLs come from, 88% of the pages ChatGPT cites were taken directly from search, so the search visibility you already maintain gates the candidate pool. Reachability gets you considered. Chain position decides who gets named.
That same mechanism settles how much of the leaderboard was ever really up for grabs.
The Citation Leaderboard Is Mostly Settled
Follow one number through its life. You measure something and publish the figure. Three blogs repeat it without linking back. An engine resolving the claim credits whichever of those pages verifies fastest, and the citation stops one hop short of where the number was born. Multiply that story across a corpus and you have the leaderboard as it stands.
The top rows were never in play. When Ahrefs examined the thousand pages ChatGPT cites most, 67% sat in categories ordinary outreach cannot touch: encyclopedias, homepages, educational domains, app stores. No content strategy makes you Wikipedia for a definitional query.
Below that ceiling sits the contestable band, and inside it the seats fill unevenly. The restating hops are crowded, because restating is easy. The origin seat, the page a trail actually bottoms out on, stays open on most chains, since only 17.2% ever reach one. Across the sample, the seat that carries the credit went unclaimed far more often than anyone fought for it.
Owned vs Borrowed Numbers
One input sits under your hand, and you can count it today: the ratio of owned to borrowed numbers across your pages. Open your 20 most important pages. Mark every statistic that carries an argument, and tag each one either measured here or borrowed from a page upstream. That ratio leads the metric; the share a tracker reports months from now is this count after retrieval and time have finished with it.
For a long time I ran the count by hand on our own posts, picking the figure I'd most hate to see credited to someone else and chasing it backward to see where it landed.
The Claims layer and Citation Provenance run that trace as a product. A pass through the Content Health Scanner extracts each statistical claim and assigns a claim origin from five deterministic values: original when the number is backed by your own poll, chart, or confirmed research; first-party when you authored it but haven't yet pinned it to data; sourced when it points to a named outside source; unattributed when it cites nothing; unknown before classification runs. Rules assign the label rather than a model, and the same pass returns an Originality Score, the proportion of your claims resolving to your own data. You can scan a page for owned versus borrowed claims this afternoon.
Owning a number also earns citations on its own weight. The research behind generative engine optimization found that adding original statistics a reader cannot get elsewhere can raise visibility in generative answers by up to 40%. Turning a borrowed figure into one you own the number for is the step after the count, with a playbook of its own.
How to Measure Citation Share Yourself
Reading your share needs no subscription. Take queries your buyers actually ask, run them through the engines that matter to you, and note whether your domain appears among the cited sources. Hold the query set fixed, repeat on a slow cadence, and you have a defensible read on where you stand.
For verifying a single citation, the AI Citation Checker reads the live page behind a claim and quotes back the exact sentence that supports it, one source at a time. For how your position drifts across months, watching position over time covers the longitudinal reading.
One trap while you sample: freshness looks like the deciding factor. Cited pages do skew newer. Ahrefs found AI assistants cite content averaging 1,064 days old against 1,432 for organic results, about 25.7% fresher, and that fresher average still sits near three years old. A recent date improves your odds of being retrieved. The citation itself follows provenance.
Why LiquiChart Has No Citation Share Tracker
LiquiChart tracks no citation share and ranks no competitors, and no leaderboard is coming; I made that call, because a product built on a readout points its users at the wrong work. What living content infrastructure operates on instead is the input. Citation Provenance records where each statistic's reference chain bottoms out. The Claims layer tells you which of your numbers an engine can attribute to you and which point past you. Living Content keeps the figures that are yours current as the data under them moves, so you stay the freshest path to a number you own.
Citation Share Follows Position
By the time any tracker updates, the contest it reports has ended. It ended on your pages, on the day each one either measured its central figure or pasted it in. Every page you publish contains that seat, the place a reference trail can bottom out, and a borrowed number leaves it to whoever you borrowed from. You cannot set your citation share. You can set the one thing it reads.