Search used to end with a list. You typed "best roofer in Austin", got ten blue links, and picked one. Ranking third still won you traffic, because the customer saw all ten.
That is not how a growing share of customers search anymore. They ask ChatGPT or Google's AI for the best roofer near them and get back one confident paragraph naming two or three businesses. There is no page two. If you are not in the paragraph, you did not lose by a little. You were not in the conversation.
Answer engine optimization (AEO) is the practice of making your business the one the AI names.
What actually changed
Three things, and only three:
- The result is a recommendation, not a list. Ten slots became roughly three. Position six no longer exists as a consolation prize.
- The AI reads about you rather than reading you. It synthesizes reviews, directory listings, news mentions, and third-party pages. Your own website is one input among many, and rarely the loudest.
- The answer is different every time. Two people asking the same question in the same city can get different businesses named. There is no single "rank" to check, which is exactly why you have to sample answers repeatedly instead of looking one up.
What did not change
The underlying signals are boringly familiar. Answer engines lean on the same things search engines did: how many credible sites link to you, how much people say about you and in what tone, whether your details are consistent everywhere, and whether a page on your site clearly answers the question being asked.
If you have done real SEO, you are not starting over. You are optimizing for a different output format using largely the same inputs.
The four levers that move an AI answer
For a local business, in rough order of impact:
- Review volume and specificity. AI engines quote what reviewers actually say. Fifty reviews mentioning "same-day emergency repair" teach the model what you are for. A five-star average with no text teaches it nothing.
- Third-party corroboration. Being listed and described consistently across directories, local press, and industry associations is what turns you from a claim into a fact the model is willing to repeat.
- Question-shaped pages. A page titled "How much does a roof replacement cost in Austin?" that answers it in the first paragraph is citable. A services page headlined "Quality You Can Trust" is not.
- Structured, attributable facts. Hours, service area, pricing ranges, insurances accepted, certifications. State them plainly, in text, on your own site. Models are far more willing to assert something they can point to.
Where to start
Pick the ten questions a customer would actually type before hiring you. Not keywords: questions. Ask each one in ChatGPT, Google, and Perplexity, and write down who gets named.
That list is your real competitive set, and the gap between it and your business is your roadmap. Kaymak automates exactly this loop: it asks those questions on a schedule, records who the AI named and why, and turns the difference into a ranked to-do list.
The question is no longer where do I rank? It is am I the answer?