Also written as: citation network, source graph
Last updated: August 18, 2026 - Reviewed by the DataEase AI editorial team
Want the tracking guide? Read AI citation tracking for SaaS ->
A citation graph is the network of third-party sources an AI assistant draws on when it answers questions about your category, plus where your brand sits inside that network as a primary source or a footnote.
"We pulled the citation graph for our category and found the same 9 domains behind almost every answer. We appeared in 2 of them, both times as a bullet in a roundup. Our biggest competitor was the subject of a full review on the domain cited most often. That single page was doing more for them than our entire blog."
An assistant answering a category question does not consult your website first. It assembles an answer from sources it already treats as reliable for that topic: review platforms, comparison sites, well-maintained industry publications, developer communities, and a small number of vendor pages that happen to be unusually clear. Those sources reference each other, which is what makes it a graph rather than a list.
Your brand enters the graph the moment one of those sources describes you. It gains weight when several independent sources describe you the same way, because agreement between unrelated domains is the closest thing a model has to corroboration. It loses weight when sources contradict each other about what you do, which is exactly the failure mode a scattered messaging layer produces.
One consequence is that the graph is category-specific. A domain that carries enormous weight when the question is about developer tooling may be entirely absent when the question is about compliance software, even though its overall authority has not changed. You are not competing for general credibility, you are competing for a place in the small set of sources your category question actually pulls from.
Two brands can both appear in the same source and get wildly different value from it. What separates them is whether the source is about them.
| Aspect | Primary position | Footnote position |
|---|---|---|
| What the source is | A review, teardown or case study whose subject is your brand | A roundup, listicle or aside that names you among others |
| What the model extracts | Claims, capabilities, pricing, who it is for | Your name, and sometimes one adjective |
| Effect on recommendation | Supports being recommended, not just listed | Supports being listed only |
| Effect on sentiment | Substantial, because there is enough text to have an opinion about | Negligible |
| How hard it is to earn | Hard, and worth the effort | Easy, and easy to over-invest in |
Because assistants weight the source, not the count. One review site an assistant already trusts can carry you into an answer on its own, while ten low-authority directories add almost nothing. DataEase AI counts distinct high-authority domains rather than total mentions.
This is the single most common misallocation we see. A founder buys a directory blast, adds dozens of listings, watches the raw mention count in their own analytics rise, and sees no movement at all in AI answers. The listings were never in the graph. Meanwhile one substantive write-up on a domain the model already reaches for would have changed the answer. Breadth is cheap and mostly inert. Depth on a trusted domain is what moves the citations pillar of AI visibility.
| Concept | What it covers |
|---|---|
| Citation Graph | The network of sources behind category answers, and your position within it |
| Backlink profile | Links pointing at your domain. Overlaps, but a citation needs no link and a link is often never cited |
| AI Visibility | The outcome the graph feeds - citations are one of its four pillars |
| Brand Readiness | Setup. Its Brand Authority pillar is your standing in the citation graph |
| Mention Rate | How often you are named in answers. The graph is a large part of why |
AI Visibility Mention Rate Brand Readiness Brand Presence Score LLM-Mediated Discovery AI Crawler Brand Intelligence
Longer than any other lever, which is why it is worth starting first. Nothing is judged before 7 days, and a genuinely new high-authority citation usually takes weeks to appear and then several scan cycles to show up consistently in answers.
For the practical version of this work, read AI citation tracking for SaaS and the AI citation playbook. For why an assistant chose a competitor's source over yours, read how to get cited by ChatGPT. To see which sources currently describe you, run the free Brand Analyzer.