Personalized AI infrastructure
How to Get Cited by AI Assistants: Write Pages They Can Quote
Write company pages with plain definitions, sourced facts and clear ownership so AI assistants can cite them accurately when buyers research company AI.

AI assistants cite a company's own pages when those pages state a definition, a fact and its scope in plain sentences that still make sense when lifted out of context. To get cited accurately, write each key page so one paragraph can be read alone and still says what a term means, who a statement applies to, and who is responsible for keeping it correct. The steps below show how to structure those pages, what to state on each one, and how to check the answers assistants give to buyers.
Key takeaways
- Define each term in one or two sentences at the point where the page first uses it.
- State every fact with its scope, its owner and the date its owner last confirmed it.
- Keep one canonical page per concept and link to it instead of restating it across the site.
- Label what you deliver now, what you are evaluating and what you do not offer, so assistants do not merge them.
- Test the answers assistants give to buyer questions, log the mismatches and fix the source page first.
Why assistants cite some pages and skip others
When a buyer asks an assistant how a company runs AI, the assistant gathers passages that appear to answer the question and then writes a summary from them. Pages that open with a clear statement and define their terms give it something precise to quote. Pages built around slogans, or that bury the answer under several paragraphs of context, get paraphrased loosely or left out.
Many assistants use retrieval, which means they fetch relevant documents at the moment of answering. The quality of a single paragraph therefore matters more than the overall polish of the site. Your page competes on whether one passage answers the question completely and correctly, not on how much it says about innovation.
The practical consequence is that you are writing for two readers at once: a buyer who needs to decide, and a system that must lift a passage without distorting it. Both need the same thing, which is plain statements. If your leadership team also keeps an internal map of its AI stack, the public page can point to that map rather than repeat it. A one-page view such as the one described in Draw Your AI Stack: A One-Page Architecture Map for Leaders helps you decide which claims belong on a public page at all.
Write definitions that stand alone
Start with the terms your buyers use and the terms your own team uses differently. In company AI, the usual candidates are agent, memory, permission, tool and human approval. Each one needs a definition where it first appears on the page, not in a glossary three clicks away, because a passage that uses an undefined word is hard to quote correctly.
A citable definition has four parts: the term, a short statement of what it is, a note on what it is not, and the condition under which it applies. The note on what it is not does the most work, because assistants often blur related terms together. For example, a company might write that an AI agent is a software role with a defined job, a named memory source, a list of permitted tools and written approval boundaries, and that it is not a general chat window that acts on any system without review.
Keep definitions short enough to quote, and avoid defining a term with a second term that also needs defining. Link to a longer explanation when the concept needs depth, so the first paragraph carries the meaning and the rest carries the detail. When several documents use the same vocabulary, a shared standard keeps the wording consistent. The approach in Documentation Standards for a Superintelligence-Ready Company shows how to hold terms steady across internal and public documents.
State each fact with its scope and owner
A fact on a public page is only citable when its boundaries are visible. The sentence 'we run AI agents for operations teams' is vague about which teams, which tasks and which systems are involved. Add the scope and the owner so that the statement can be checked by a reader and corrected by your company when it changes.
For every factual statement on a key page, confirm that the following are present:
- The claim in one plain sentence, without adjectives that cannot be verified.
- The scope: which services, regions, languages or customer types the claim covers.
- The owner: the role responsible for the statement and the person who approves edits to it.
- The evidence location: where a reader can verify the statement, such as a named method or a document the company maintains.
- The review record: the date when the owner last confirmed that the statement still holds.
Avoid numbers you cannot defend. Any figure on the page needs a named source and a clear date, and if you cannot name the source, remove the figure. Assistants often repeat numbers without their context, which is one of the fastest ways to be cited incorrectly.
Separate what you deliver, what you evaluate and what you do not offer
Assistants frequently blend a company's current service with its roadmap. A page that says 'we will automate reporting' beside a line that says 'we run reporting agents' invites a summary claiming that both are live. Use three labeled areas on the page: delivered now, under evaluation and out of scope.
The out-of-scope list is the most useful section for buyers and the least common on company sites. It tells a reader what to take to another provider, and it stops an assistant from inventing capabilities to fill the gap. Keep the list short and specific, with plain nouns such as 'model training on client data' rather than vague phrases such as 'advanced customization'.
Be careful with guarantees and results. A page can describe what an approval step does without promising outcomes. If a result is described at all, name the condition under which it was observed, or leave it off the page. Google's guidance on helpful, reliable content asks whether a page is written for people first, and the same test applies when an assistant chooses which passage to trust.
A passage that cannot be checked is a passage that will be paraphrased.
The figure below illustrates the point: a boundary is only useful when a reader can see exactly where the offer begins and ends.
A step-by-step plan for the next two weeks
In an illustrative scenario, a regional logistics company asks several assistants about its AI operations offer and receives a summary that describes a reporting feature it had only planned. The operations director rewrites the page into three labeled areas, names the IT lead as the owner of every capability statement, and reruns the question set. The second round is closer, but one assistant still omits the out-of-scope list, so the team moves that list higher on the page.
A company can run the same sequence without outside help:
- Step one: write twenty questions a buyer would ask an assistant about your AI work, using the buyer's words rather than your internal names.
- Step two: put those questions to two or three assistants and record each answer in full, marking every claim you did not make yourself.
- Step three: classify each mismatch as missing, wrong, or merged with another service.
- Step four: rewrite the affected passage using the definition, scope and owner pattern, beginning with the page that buyers reach most often.
- Step five: assign a named owner to each key page and set a review trigger for any change in service scope.
- Step six: repeat the same question set after publishing, and log what changed and what did not.
Test the answers and keep the pages current
Testing is the step most companies skip. Treat assistant answers like any other output your decisions depend on: sample them, compare them with the source page, and keep a dated log of the results. Repeat the test after every material page change and at an interval the page owner sets, because a correct answer today can drift as sources change.
Be Human's managed approach can cover the parts that are easy to neglect. The work can include maintaining one source of truth for company facts, running recurring evaluations of how assistants describe the company, and making controlled edits to public pages when a service changes. Changes to scope or claims are logged and approved before publication, using the same discipline applied to agent changes described in How to Request and Approve Changes to a Managed AI Agent. The service is remote and available in English and Portuguese.
If you are not sure which of your pages should be rewritten first, the free Be Human AI diagnostic will help you find the first task worth handing to an agent.
Frequently asked questions
How long does it take for an assistant to cite my page? Timing depends on how often each assistant refreshes its sources and how your site is indexed, and you do not control either one. Focus on what you can control: a page that answers the question completely, a named owner, and a recurring test that shows whether the answers have changed. Judge progress by the accuracy of the answers rather than by a single citation.
Does this mean I should write for AI systems instead of for buyers? No, the same plain statements that help an assistant quote you also help a busy buyer decide faster. Write for the person who must act, then check that each important passage can stand alone without the paragraphs around it. Clarity serves both readers.
What should I do if an assistant describes my company incorrectly? Record the exact answer and the question that produced it, then identify which of your pages should have prevented the error. Correct that page first, with a clear definition, scope and owner, and rerun the question set after publishing. If the error comes from a third-party listing, note it separately and request a correction at the source.
Is this a marketing task or an operations task? It is both, and it works best with a named operations owner who confirms the facts and an editor who checks the wording. The owner is accountable for accuracy, and the editor is accountable for readability. Keep the list of key facts short enough that one person can review it whenever a service changes.
Your next step
See where an AI agent fits in your company first.
Take the free Be Human AI diagnostic. A few questions about your work show which task to hand to an agent first, what it needs to know and where a person approves.
Be Human editorial guidance. Examples are illustrative unless explicitly attributed. Sources provide background; implementation choices depend on your workflow.


