How AI Search Decides Who Gets Mentioned

What happens when someone asks about your industry

When a customer asks ChatGPT “who's a good bookkeeper in Irvine for a small construction company,” the AI does not simply recall an answer from memory. It runs a live search, collects chunks of text from the pages it finds, and then writes an answer based on those chunks. The technical name for this is “retrieval-augmented generation,” but the plain-English version is what matters: it retrieves first, then it writes. Everything you do in this course is about making sure your text is what gets retrieved.

Step 1: your one question becomes a dozen questions

Google confirms in its own documentation that AI Overviews and AI Mode may use a “query fan-out” technique, issuing multiple related searches across subtopics at once. So that one bookkeeper question quietly becomes eight or twelve hidden searches: construction bookkeeping, Irvine accountants, QuickBooks for contractors, job costing, and so on. This is genuinely good news for a small business. You don't have to win the one big competitive phrase. You have a dozen smaller doors to get in through, and a page that answers a specific sub-question well can be pulled in even when you'd never rank for the headline term.

Step 2: it pulls passages, not pages

This is the single most important idea in the course. The engine splits pages into segments and matches those segments against the question. It is not weighing your website as a whole, and it is not weighing your home page as a whole. It is looking for a specific paragraph that answers a specific question. That paragraph then gets read completely out of context, with nothing around it to explain what “we” or “our service” refers to. A paragraph that only makes sense after reading the three paragraphs above it is almost useless here, no matter how well written it is.

Step 3: it writes the answer and decides who to credit

The retrieved passages get ranked once more, and the model writes an answer grounded in the best of them, attaching citations to some. Different engines make this call differently, which is why you'll test more than one. Google's own AI Mode and AI Overviews share only about 14% of their cited URLs, so even two surfaces from the same company disagree about who deserves the credit.

Where each engine looks

• •ChatGPT and Microsoft Copilot lean heavily on Microsoft's Bing index. If Bing can't find you, ChatGPT usually can't cite you.
• •Google AI Overviews, Google AI Mode, and Gemini use Google's index.
• •Perplexity runs its own crawler and leans unusually hard on community sources like Reddit.
• •The practical upshot: being crawlable and indexed is table stakes, not an advantage. It's the price of entry, and you'll check yours two lessons from now.

Ranking #1 is not the same as being cited

You might assume the AI simply quotes whoever ranks first. It doesn't. A December 2025 analysis of 173,902 URLs found that 68% of the pages cited in AI Overviews were not in the organic top 10 at all. Ranking well certainly helps you get found, but it does not buy you a citation, and not ranking well does not lock you out. What gets you quoted is having the clearest, most self-contained answer to the specific sub-question being asked.

The research on what actually works

The strongest evidence here comes from an academic paper, not a marketing blog. A team from Princeton, Georgia Tech, and IIT Delhi tested roughly 10,000 queries and published the results at the KDD 2024 conference. Their headline finding, in their own words, is that including citations, quotations from relevant sources, and statistics can significantly boost source visibility, with an increase of over 40% across various queries. Three tactics, all of which come down to the same thing: back up what you say.

• •Cite your sources. Naming where a claim comes from made sources measurably more likely to be quoted.
• •Add real statistics. Specific numbers give an AI something concrete to lift.
• •Add quotations. A quoted line from a credible source lifted visibility roughly 22%.
• •Do not stuff keywords. Repeating your keyword scored about 8 to 10% below the untouched baseline. It actively hurt.

Why this favors small businesses

Here's the finding that should decide whether you spend the next two hours on this. In that same research, the sites that gained the most were the ones starting further down: a source sitting in fifth place saw its visibility rise by up to 115% just from adding proper citations. Researchers call it the “equalizer effect.” In classic SEO, out-ranking a national competitor with a twenty-year-old domain is close to hopeless for a small local business. In AI answers, a well-structured, well-sourced paragraph from a two-page local site can beat them, because the engine is looking for the best answer to a narrow question, not the biggest brand.

But if nobody clicks, why bother?

It's a fair question, and the honest answer has two halves. Yes, clicks are down: Pew found 8% click-through when an AI summary appeared versus 15% without, and Seer Interactive measured organic click-through on those queries falling from 1.76% to 0.61%. But being named in the answer still pays, in two ways. Brands cited in AI Overviews earn meaningfully more clicks than uncited competitors, and the visitors who do come through convert far better, around 4.4 times organic in Semrush's data. On top of that, plenty of value here never shows up as a click at all: when an AI tells a customer your business is the right fit, that recommendation does its work whether or not they visit your site first.

The golden rule, same as always

Every prompt in this course drafts something for you: a score, a rewrite, an FAQ, a correction. AI drafts, you verify. That matters even more here than in other courses, because the whole point is to publish text that AI engines will repeat to your customers. Publish a number you never checked, and you've taught the machines to spread your mistake. Never let a statistic, price, or claim go live until you know it's true.