TL;DR
Answer engines that show sources cite specific pages that answer the question asked. They do not cite a brand in general. The citation gap is the distance between the questions your buyers ask and the pages you have that answer them directly.
Closing it means building pages around real buyer questions, putting the answer near the top in plain words, and backing each claim with a source a machine can check.
Citations follow pages, not reputations
A business can be well known locally and still be absent from AI answers about its own field. That looks unfair until you look at what gets cited: a directory page listing five providers, a comparison article, a competitor's guide that answers the question in its first paragraph.
Each of those is a page built around the question. Your homepage, which describes your business in general, is not. The system isn't choosing a competitor over you. It's choosing a page that answers over a page that doesn't.
Three patterns behind the gap
No matching page. The buyer asks about a specific service in a specific town and the business has no page for that combination. Our BIC field study is a worked example: the agency was already recognised in two markets but had no page tying its business insurance work to Lakeland, the market where it was missing.
The answer is buried. The page exists but opens with a slogan and gets to the point four paragraphs down. A system extracting a short answer takes the clearer page.
Nothing to check. The page makes claims with no source, date or named author. Pages that show where a fact came from are easier to trust and quote. That is an inference, not a proven rule, and we label it that way.
Closing the gap
Start with the questions, not the keywords. Write down the ten questions buyers actually ask before they hire someone like you, in their words.
Run those questions through the assistants your buyers use and note which pages get cited. That list is your real competition, and it often includes directories and publishers, not just rivals.
Build or fix one page per question: the direct answer in the first two sentences, then the detail, then the source for each claim, with the author and date visible.
Re-run the same questions on a schedule and record what changes. Without a recorded before-and-after, you can't tell a real improvement from a lucky week.
What we're testing
An early five-model comparison is now logged on our Experiments page with exact answers. For the Lakeland insurance and deathcare records prompts, none of the five models returned a BackTier link or any source annotation. They had no browsing or retrieval access, so this is not a valid citation-rate test and cannot show whether field studies outperform blog posts. The scheduled multi-engine run and comparison remain open.