Picture the moment that makes this urgent. You are on a call and the prospect mentions they asked Claude to compare options in your category, and your brand never surfaced. The two names Claude offered are competitors you beat on the merits, but Claude had never heard your case made clearly enough to repeat it. That is the problem worth solving, and the good news is that how to get cited by Claude comes down to a few things you can influence directly rather than a mystery you have to accept.
Where Claude actually gets its sources
Claude answers from two places, and telling them apart is the whole strategy. The first is its training, the vast body of text it learned from, which gives it a general sense of who and what exists in the world up to a point in time. The second is retrieval: when Claude can search the web during a conversation, it fetches live pages, reads them, and cites what it used. Those two sources behave differently and reward different work.
When Claude answers from training alone, it can only mention brands that appeared often and clearly enough in its training data to be remembered, and it will not cite a live URL. When Claude retrieves, it can find and quote your page directly, which is where a fresh, well-structured page pays off immediately. Most brands ignore the difference and treat AI citation as one thing. It is two, and you need a plan for each.
Which mode Claude is in changes what a citation even looks like, and that is worth internalizing before you plan anything. In retrieval mode, a citation is a live link to a specific page, delivered in the moment, traceable and clickable. In training mode, there is no link at all, just Claude mentioning your brand as part of what it knows, with no URL to click and no page to point to. Both are valuable, and they are valuable to different parts of your funnel. The live citation can drive a click today. The remembered mention shapes whether you are on the shortlist a buyer forms before they ever open a browser.
Confusing the two leads to bad strategy. A brand that measures only clickable citations will undervalue the reputation work that makes Claude name it from memory, because that work produces no link to count. A brand that chases only broad mention will neglect the page-level clarity that wins the live, traceable citation. The Corpus-and-Context split exists precisely to keep these straight, so you invest in both the durable presence that earns the unlinked mention and the sharp pages that earn the linked one.
The Corpus-and-Context split
Here is the framework I use to keep it straight. Call it the Corpus-and-Context split. Corpus is the durable presence you build across the open web over months and years, the mentions, the coverage, the consistent story that seeps into training data and long-term reputation. Context is the immediate, retrievable material Claude can pull into a single conversation right now: your current pages, the fresh third-party coverage, the clearly-stated answers a live search can reach.
You need both, because they cover different questions. Corpus is what makes Claude name you unprompted when someone asks a broad question with no live search, because your brand is simply part of how the model understands the category. Context is what makes Claude cite your specific page when it does search, because your page was the clearest, most current answer available. Brands that only chase context get cited when Claude searches and forgotten when it does not. Brands that only build corpus get remembered vaguely but never quoted precisely. To get cited by Claude across both modes, you build the split on purpose.
The split also tells you where to spend a limited budget. Context work is faster and cheaper: you control your own pages, so you can make them quotable this week and start winning retrieval-mode citations almost immediately. Corpus work is slower and more expensive, because it depends on coverage, reputation, and time you cannot shortcut. A brand just starting out should usually front-load the context work, capturing the citations available whenever Claude searches, while beginning the patient corpus work that will pay off later. Trying to do only the slow half leaves easy citations on the table for months.
There is a sequencing logic here worth stating plainly. Fix your pages first, because that is the fast win and it also feeds the slow one, since well-structured pages are exactly what earns the coverage and links that build corpus over time. Then invest in the reputation work that makes Claude name you from memory. The two halves are not independent tracks you run in isolation. They reinforce each other, and the brands that get cited by Claude most reliably are the ones that started the context work immediately and let it compound into corpus rather than treating the two as separate projects.
Be the page Claude retrieves and quotes
Start with context, because it moves fastest. When Claude searches and lands on your page, it is looking for a passage that answers the question cleanly enough to quote and attribute. That means leading your sections with direct answers, writing in plain declarative sentences, and making claims specific enough to be useful but grounded enough to be safe. A page that states its answer in the first two sentences of a section is far more quotable than one that builds to the point across six paragraphs.
Accuracy and clarity matter more here than persuasion. Claude is cautious about repeating claims it cannot verify, so a page thick with unsupported superlatives is a page it will read past. Give it verifiable specifics, clear definitions, and answers it can stand behind, and you become the source it reaches for. The same discipline that makes a page genuinely useful to a human reader is what makes it citable, which is a relief, because it means you are not writing two versions of the truth. You are writing one clear one and letting both audiences benefit.
Write for the two-reader problem
Every page you publish now has two readers, and they judge it differently. The first is the human who might land on it, skim it, and decide whether you are worth their time. The second is Claude, which might retrieve it, read it in full, and decide whether to quote it in front of that human before they ever visit. Most content is written for the first reader and accidentally fails the second, because the things that make a page skimmable for a person are not the same things that make it quotable for a model.
The good news is that the two readers agree more than they conflict. Both reward a clear answer stated early. Both reward specifics over vague claims. Both reward an obvious author and a credible source. Where they diverge is subtle: a human tolerates a page that builds slowly to its point, while Claude prefers each section to state its point immediately so a passage can be lifted cleanly. Writing for both means front-loading your answers without stripping the depth underneath, so the human who keeps reading gets the full argument and the model that only needs one sentence gets a quotable one.
The failure mode to avoid is writing two different pages, a shallow one tuned for machines and a rich one for people. That splits your effort and usually produces something worse for everyone. The better path is a single page that respects both: answers first, evidence and nuance second, a clear voice throughout. When you get cited by Claude, it is almost always because you wrote one honest, well-structured page that served both readers at once, not because you gamed a system with content no human would want.
A five-minute test: interview Claude about your category
The fastest way to see where you stand is to ask Claude directly. With web search enabled, put the questions your buyers ask into Claude and watch what happens. Ask it to compare the options in your category, to explain how to solve the problem you solve, to recommend providers or approaches. Then read the answer like a scoreboard. Did your brand appear? Was it cited from a live page or mentioned from memory? Which competitors showed up, and what did their cited pages do that yours did not?
This exercise turns an abstract worry into a concrete list. If Claude never mentions you at all, your corpus presence is thin and the long game of coverage and reputation needs work. If Claude mentions you vaguely but never cites a page, your context is weak and your on-page answers are not quotable enough to lift. If Claude cites a competitor’s page over yours on a question you answer better, compare the two pages side by side and you will usually find the competitor stated the answer more plainly or backed it with corroboration you lack.
Run the test again a month after you make changes, and it becomes a feedback loop rather than a one-time audit. You will watch your citations improve as you sharpen pages and build reputation, and you will catch the questions where you are still losing while there is time to fix them. The brands that get cited by Claude consistently are not the ones that guessed right once. They are the ones that kept asking the model what it thought and kept closing the gaps it revealed, until showing up in the answer became the normal result rather than the lucky one.
Build the reputation Claude already trusts
Corpus is the slower, deeper work, and it is where digital PR earns its keep. The brands Claude names from memory are the ones that showed up consistently and credibly across the web: covered by real publications, referenced in industry conversations, associated clearly with the topics they want to own. That presence is what a model absorbs as it learns, and what makes your name part of the category rather than a stranger to it. You cannot fake it in a week, which is exactly why it is defensible once you have it.
The move is to make your best thinking impossible to miss and easy to corroborate. Publish a genuine point of view under real author names. Earn coverage that repeats your claims in credible places. Keep your story consistent, so every source Claude might encounter tells the same version of who you are and what you are known for. Over time that consistency hardens into the kind of reputation a model treats as settled fact. Do the context work so Claude can quote you today, do the corpus work so Claude remembers you tomorrow, and how to get cited by Claude stops being a question you ask on sales calls and becomes something you have already built.