Why AI answers are different from Google rankings
A Google ranking is a position on a list of links. An AI answer is a short piece of writing that names a handful of businesses and explains why. Those are different products, built differently, and being good at one doesn’t make you good at the other. Whether the underlying work is even the same discipline as SEO is a real question; here the focus is narrower: what the customer actually sees, and why the answer format punishes different weaknesses than the list ever did.
Ten blue links vs one synthesized answer
A search ranking gives the customer ten options and lets them choose. An AI answer makes the choice for them: it reads several sources, weighs them, and writes back three to five business names with a reason for each. If you’re not one of the names, you were never seen at all.
That’s the whole structural difference. A list distributes attention unevenly but widely. An answer concentrates it on the few businesses the AI decided to trust. There is no page two of an AI answer.
Position three still gets clicks. Omission gets nothing.
Ranking third on Google is a fine outcome. Plenty of customers scroll, compare, and click the third result. Being left out of an AI answer has no equivalent consolation prize, which is a big part of why a business that looks fine on Google can be absent from ChatGPT.
The click data shows the squeeze from both sides. By early 2025, 18% of Google searches already showed an AI answer at the top1. And when that summary appears, people click the traditional results beneath it far less: 8% of visits with an AI summary ended in a click on a regular result, versus 15% without one2. That’s Pew Research Center’s browsing study from March 2025, so treat it as a dated snapshot of a moving number, not a law. The direction is what matters: the answer absorbs attention that used to flow down the list.
To be clear about what this data does and doesn’t say: it shows clicking behavior changed when summaries appeared. It doesn’t prove any one business lost customers, and we won’t pretend it does.
Rankings are per-page. Mentions are per-business.
Google ranks pages. Your service page might rank for “water heater replacement” while your homepage ranks for your brand name. SEO work reflects that: you optimize page by page.
AI assistants evaluate the whole business as an entity. When one decides whether to recommend you, it’s weighing what your reviews say about you, what your Google Business Profile feeds it, what directories list about you, and what your own site says, all at once. A perfect page attached to a business with thin reviews and mismatched listings still loses. The unit of competition moved from the page to the reputation.
Rankings update on a crawl. Answers change per conversation.
Rankings move slowly enough that agencies report them monthly. AI answers can differ between two conversations on the same day, because the model may pull different sources, interpret the question differently, or simply phrase its shortlist another way.
This is uncomfortable and worth saying plainly: nobody holds a stable “position” in AI answers. What you can influence is how often you appear across many askings, which is why a single screenshot of an AI answer proves very little and why watching your mention rate over time beats checking once and declaring victory.
What carries over from ranking work
Don’t torch your SEO. Most of the foundations that make a business visible to AI in the first place are inherited from honest ranking work: real reviews, a complete profile, a crawlable site, pages that answer real questions. If you’ve done that work, you’re carrying an inheritance, not starting over.
Here’s our opinion, and it’s one plenty of the industry won’t like: a monthly report that only shows rankings and traffic is measuring the old game. Rankings tell you where your pages sit on a list. They don’t tell you whether an AI names your business, and nothing in a rank report will warn you when it doesn’t. The fix isn’t to abandon rank tracking. It’s to stop treating it as the whole scoreboard.