When Customers Ask AI Instead of Google

Last week, three different prospects told me they found something on this site by asking ChatGPT a question. Not by searching Google and clicking the third result. By typing a question into an AI tool and acting on whatever it said. Two of those prospects had never opened my site before that moment.

This pattern is now common enough that it has a name, or rather two. AEO covers showing up in tools that return a direct answer to a question, like ChatGPT, Perplexity, or Google's own AI Overviews. GEO covers a broader category of search experiences where the model summarizes multiple sources before deciding what to cite. The terms overlap. The practical implication is the same: optimizing for ten blue links is not enough when most of your customers may not see ten blue links anymore.

The numbers behind the shift

OpenAI reported 800 million weekly active ChatGPT users in October 2025. Perplexity passed 22 million active users in the same period. Google's SERP now includes AI Overviews on roughly half of all informational queries in North America (StatCounter, Q1 2026).

What this means for a small business: a growing share of the queries that used to send traffic to your website now end with the user reading a summary the AI generated. If the summary cites your site, you may still get the click. If it does not, the customer has their answer and the click never happens.

How AI search picks what to cite

Different tools work differently. Most share a common pattern: the model retrieves a set of candidate pages, reads them, and synthesizes an answer that may cite the strongest source or sources. The factors that influence which pages get pulled in look very similar to traditional SEO factors, plus some new ones:

• Authority signals. Backlinks, mentions on trusted sites, and domain-level reputation still matter. Models trust well-cited sources more than fresh anonymous content.
• Clarity of answer. A page that directly answers a question in a short, well-structured passage gets cited more than one that buries the answer in 2,000 words.
• Structured data. JSON-LD schema, FAQ markup, and HowTo structures help models parse the page faster and more reliably.
• Conversational phrasing. Page content written the way humans actually ask questions ranks higher than keyword-stuffed copy aimed at older search engines.
• Recency. AI tools often favour content published or updated within the last 6-12 months for time-sensitive queries.

What AEO actually requires

A clean, fast, well-structured site is most of the work. The principles are not new. The emphasis is.

Write content that answers questions directly. If a customer asks "how much does a website cost in Ontario," the page that answers it should open with a clear range or framework, not a 600-word preamble about the history of web design. Front-load the answer.

Use proper heading structure. An H2 phrased as a question, followed by a paragraph that answers it, is the structure AI models parse most reliably. Headings that read like ad copy ("Unlock Your Potential") teach the model nothing about what is on the page.

Add FAQ schema where it fits. A short FAQ block at the bottom of a service page, with three to five common questions and direct answers, gives AI tools an explicit, structured set of question-answer pairs to draw from. This is one of the highest-leverage changes a small site can make.

Keep your NAP consistent across the web. AI tools use your GBP, your website, your directory listings, and your social profiles together. If they disagree, the model has to pick which one to trust, and that often is not yours.

What GEO adds

Generative Engine Optimization extends AEO with an additional concern: ensuring your site looks credible to the model evaluating a wider set of sources. The model is comparing your page to others on the same topic and deciding which to weight more heavily. Several things help:

• Original data or analysis. Models prefer to cite sources that introduce information rather than paraphrase others. A blog post built around primary research or original framing outperforms one that summarizes the top 10 articles.
• Clear authorship. A visible byline, an "about" page that describes who you are, and consistent voice across the site help models identify your content as authored rather than scraped.
• Linkable assets. If your page is the kind of resource other sites link to, that authority compounds. Pages with concrete data and clear opinions get linked. Pages of generic copy do not.

For small businesses specifically

This matters more for small businesses, not less. The mathematics is unforgiving. If a customer asks an AI "who builds websites in Pembroke" and gets three names, you either are one of them or you are not in the conversation. There is no second page of results. There is no "and 1,400,000 other websites."

The same applies to every local service category. Plumbers, electricians, roofers, accountants, lawyers, restaurants. The AI returns a short list, and the names not on the list are functionally invisible.

The fix is the same as the fix for good SEO: clear content, working structured data, consistent NAP, fast pages, and original material that answers questions directly. The bar is higher than it was three years ago. The reward, on small markets where the candidate list is short, is also higher.

What I am doing about it on this site

This site uses JSON-LD on every page, with per-city LocalBusiness data on local landing pages, FAQPage schema on services and local pages, BlogPosting schema on every article, and consistent NAP across the footer, contact page, and GBP. Headings are written as questions where they make sense. Content opens with direct answers, then expands.

It is not magic. It is the same SEO discipline applied with awareness that the consumer of the page may be a model rather than a person. The model still reads HTML.

If you are reviewing your own site for AI search visibility, get in touch and I will tell you what I see.