Answer engines cite the passage that answers the question as asked, in a form that can be lifted out of the page without losing its meaning. A page can be accurate, compliant, and well written, and still miss that standard. Common causes are a buried answer, a hedged answer, a missing detail, or an answer to a neighboring question. The gap closes when you audit at the level of the question and the passage, not the page.
An answer engine is a search tool that writes a response and attaches source links to it. Instead of returning a ranked list of documents, it generates a written answer with citations. Perplexity and Google AI Overviews are examples. Being found and being cited are separate events, and this article is about the space between them.
What the Citation Gap Is
The citation gap is the difference between the questions your firm should be able to answer and the questions where an answer engine actually names you as a source. It has two parts.
- Not found: the engine never retrieves your page for that question.
- Found but not cited: the engine may retrieve your page, then write its answer from another source.
The second part is harder to see. You cannot watch retrieval happen. You infer it from what you can see: your page ranks in ordinary search, or your firm is named in an answer, but no link points to you.
Why Engines Cite Only a Few Sources
Answer engines generally work in stages. They retrieve candidate material, ground a response in it, and attach attribution. The engine may read many pages and credit only a few.
The number of citations differs by engine. One academic comparison found that SearchGPT cited a median of 3 sources per query. The same study found that SearchGPT's most-cited domains overlapped only about a quarter with Google's organic results. So a page that earns traffic from classic search is not guaranteed a place among the sources an answer engine credits.
The selection step is also not visible from outside. Source selection, answer generation, and citation display are platform decisions that the platforms do not fully publish. Anything said about what engines prefer comes from outside testing. It is evidence, not a rulebook.
A Worked Example: A Composite RIA
Marcus is a composite, an invented advisor, not a client of ours. He runs a three-advisor registered investment adviser (RIA) firm. His site has a page called "Our Approach to Fees." A prospective client might ask an answer engine, "How do fee-only advisors charge?" Marcus's firm could plausibly appear in that answer.
Here is how his page is built. It opens with 300 words on the firm's philosophy. The fee schedule sits in a PDF linked near the bottom. Every sentence that mentions cost carries its own qualifier, so no sentence states a plain fact. The page has no date and no named author.
Now take a hypothetical directory page on a third-party site that covers the same question. It opens with one short paragraph that names the common fee models and defines each in a sentence. It carries a "last updated" date.
Picture the engine assembling an answer. It needs a passage that can stand alone, and the directory's paragraph is one. Marcus's page may have been retrieved. It offered nothing short enough to quote.
The fix is mostly structural. Marcus puts a direct answer at the top of the page and moves the fee schedule into the page text. He adds a visible review date, plus the name and credentials of the person who wrote it. Disclosures move into a clearly labeled block near the answer instead of into every sentence. All of this goes through the firm's normal compliance review. The wording remains the compliance officer's decision. The layout is a publishing decision.
Five Reasons Compliant Content Gets Passed Over
These are patterns an audit tends to surface. Each has a mechanism behind it.
- The answer is buried. An engine extracts passages, not whole pages. If the answer needs three paragraphs of setup, no single passage contains it.
- The answer is hedged into nothing. A sentence that qualifies every claim often asserts none. A passage with no claim gives the engine nothing to attribute.
- Specifics are missing. Outside testing suggests that concrete details, such as a stated price or a recent date, make a passage easier to attribute. This evidence is early. A page that says "competitive, transparent fees" and names no structure offers no such detail.
- The page answers a nearby question. A page about "our fee philosophy" matches the question "how do fee-only advisors charge?" less closely than a page that states that question and answers it. Closer topical match gives the engine a clearer reason to pick the passage.
- A third party has the clearer passage. A directory or trade publication that answers in one paragraph hands the engine something easy to quote. Your page competes with it for the same slot in the answer.
Structure alone does not close the gap. Headings and bullets help an engine find a passage. They cannot supply a claim the passage lacks. A page with no clear, specific answer still has none after it is reformatted.
How to Audit the Citation Gap
An audit is a question list, a record of what engines return, and a comparison. It takes a few hours for a small firm. It needs no special tools.
Build the Question List
Write 15 to 25 questions that a prospective client would type before contacting an RIA. Phrase them the way a person would. Include questions about how fees work, how a firm is regulated, and what a first meeting involves. Keep to questions your firm can answer in its own marketing voice.
Record What Each Engine Returns
Run each question in at least two answer engines. Log the result in a simple sheet with one row per question and engine.
- Question: [the exact wording you typed]
- Engine and date: [engine name and the date of the run]
- Sources cited: [the domains linked in the answer]
- Our firm named in text: [yes or no]
- Our firm linked as a source: [yes or no]
- Our page that covers this question: [web address, or "none"]
Results vary from run to run. Repeat the set on a later date before drawing conclusions.
Compare Your Page to the Cited One
For each row where you are named or ranked but not linked, open the page that was cited. Read its first 100 words. Check whether it states the answer, gives a specific detail, shows a date, and names an author. Then read your own page the same way. The difference is your edit list.
Fix, Then Retest
Change one page at a time through your usual review. Rerun the same question on the same engines later. If the result changes, you have a data point. If it does not, you have still learned something about that question.
What We Do Not Know
Being cited is not a pure measure of merit. One study that tested several answer engines found cases where a citation pointed to a page that did not support the sentence beside it. The research cited above was run under test conditions, not on every live engine. Its findings are directional.
A restructured page does not guarantee a citation. What the audit gives you is a clear view of what your page offers an engine to quote, and a short, specific list of what to change.
