The short answer
Because AI Overviews do not rank pages, they assemble answers. Google's system picks sentences it can lift cleanly and trust, so a clearly worded, well structured page often beats a higher ranking one. Your competitor is easier to quote, not more popular. Fix the wording and structure and you can be cited within weeks.
Ranking and being quoted are two systems
Your competitor is easier to quote, not more deserving. Google AI Overviews do not simply promote the top ranked page. They read many sources, extract sentences that answer the question directly, and stitch them into a summary. A page can sit below yours in the blue links and still supply the exact line the model wanted. Position ten can win the citation if position one buries its answer.
Classic ranking rewards backlinks, domain history and search intent signals built over years. Generative answers reward clarity and structure in the moment. The model needs a passage it can reuse without editing, so it favours pages that state the answer in plain sentences near the top. If your page opens with a long story and hides the answer in paragraph six, the model may skip you and take a cleaner line elsewhere.
There is also a timing gap. AI Overviews and normal search results refresh on different cycles, so your ranking today does not guarantee your presence in an answer today. Google draws on a shortlist of trusted sources for each topic, then chooses passages from within it. Being in that shortlist matters more than your exact position. Winning the answer is about being the clearest voice, not the highest one.
How AI Overviews decide who to cite
Google builds the summary first, then attaches sources to the claims inside it. So the question is not who ranks best, but who wrote the sentence the model chose to reuse. It looks for passages that match the query wording, read as fact rather than opinion, and stand alone without surrounding context. A definition, a direct answer or a short list is far easier to lift than a paragraph of persuasion.
Trust signals still count, just differently. The model prefers sources it can corroborate, so if two or three other sites confirm the same fact your competitor stated, that fact becomes safe to quote. Consistency across the web acts as a vote. If your figures, service descriptions and claims differ from page to page, the model has reason to doubt you and reaches for the source it can verify more easily.
Structure does heavy lifting. Clear headings, a question phrased the way people ask it, and an answer in the first two lines all help the model find and trust your text. Schema markup, such as FAQ and Article data, tells the crawler what each block means. None of this changes your ranking much. It changes whether your words are ready to be picked up and repeated in an answer.
Why a lower ranked rival still gets named
The most common reason is answer placement. Your competitor probably states the answer in the first sentence under a heading that matches the question. You may rank higher because of stronger links, yet lead with background before you get to the point. The model reads top down and grabs the first clean, confident answer it finds. Burying the payoff is the fastest way to lose the citation to a weaker page.
The second reason is coverage of the exact question. AI Overviews trigger on specific phrasings, and your rival may have a page built around that phrasing while you cover it in passing. A dedicated page that names the question in its heading and answers it plainly will beat a broad page that mentions the topic once. Match the question people actually type, then answer that question and nothing else on the page.
The third reason is corroboration and mentions off your own site. If your competitor is described the same way across directories, review sites and articles, the model sees a consistent story and trusts it. If you appear rarely, or your details clash, you look risky to quote. This is why presence beyond your website matters. Being named by other sources, in the same words, makes an assistant confident enough to name you too.
What to check on your own pages
Start by reading your page the way a machine does. Open the target page and ask whether the answer to the question appears in the first two lines. If it does not, that alone can explain why a rival is quoted. Then check that your heading matches the question a person would type, not a clever marketing line. These two fixes often move a page from ignored to cited faster than any link building.
Confirm the crawler can read you
Googlebot and Google-Extended must be allowed in robots.txt. If either is blocked, you are invisible to the answer engine and every citation goes elsewhere.
Lead with the answer
Put a direct, factual reply in the first two lines, under a heading that matches the exact question a person asks.
Add structure
Use clear subheadings, short lists and FAQ schema so the model can lift a clean, self contained passage.
Finally, look for contradictions between your pages and your profiles elsewhere. Make your service description, your location coverage and your key claims identical wherever they appear. Assistants reward a single, consistent story. When your own site, your directory listings and any articles about you all say the same thing, the model has no reason to hesitate. Consistency is quiet work, but it is often the difference between being read and being repeated.
How to earn the citation within weeks
You can usually influence AI Overviews faster than classic rankings. Rewrite the page so the answer leads, the heading matches the question, and the supporting facts sit in short, liftable sentences. Then request indexing in Google Search Console so the crawler revisits sooner. Changes framed this way are often reflected in AI answers within a few weeks, well before a ranking shift would show. Speed here comes from clarity, not from spending.
- Write the answer first in two lines, then explain.
- Name the exact question as your heading, in plain words.
- Build corroboration so other sites describe you the same way.
- Keep claims consistent across your site and every profile.
- Track the answer by asking ChatGPT, Perplexity and Copilot the question weekly.
Do the same across the assistants that matter to your buyers. ChatGPT, Perplexity and Microsoft Copilot each read the web differently, so test your question in all of them and note who they name. Where a rival appears and you do not, study the exact sentence quoted and write a clearer version of that fact on your own page. Repeat the check every week. Progress is visible, measurable and usually quicker than people expect.