AI SEO Is Still SEO. Here's What Changed.

TL;DR: AI SEO is still SEO. What changed is structure. AI engines need answers they can lift straight out of your page, which means a direct one-sentence answer under the heading, a real TL;DR, honest FAQs, and schema code that tells the engine what your business is. Schema is the piece most businesses skip because it looks like programming, and it is also the change that moved our own client work the furthest. Off your site, watch where the engines buy their data, because ChatGPT now reads Yelp reviews.

Business owners ask us some version of the same question every week. Do I need AI SEO, or do I need regular SEO?

The answer is yes.

That sounds like a dodge, but it is the honest answer. Nearly everything that made your site rank five years ago still counts. What sits on top of it is a set of organizational requirements that did not exist before, and those requirements are where we keep seeing movement.

What Google actually said about AI search

You have probably heard that Google is going fully AI by the end of the year. Google has never said that.

Here is what Google did say. At I/O in May, Sundar Pichai reported that AI Overviews passed 2.5 billion monthly active users and AI Mode passed 1 billion in roughly a year, with AI Mode queries more than doubling every quarter. Asked on the New York Times Hard Fork podcast in June whether Google would simply switch everything over to AI Mode, Pichai called the transition a continuum and said "sources and links will always be there as part of it."

There is no deadline and no announced switchover date. What Google describes is a gradual absorption of classic search into an AI answer, already well underway, with citations still pointing somewhere.

That last part is the whole opportunity. If sources and links stay in the answer, then the question for your business is whether the link is yours.

AI SEO is still SEO, with new requirements

We treat AI SEO as one discipline with several names. Generative engine optimization, answer engine optimization, AI overview optimization, LLM citation optimization: same job, different vocabulary, and the vocabulary mostly tells you who is selling it.

The foundation has not moved. You still need a site that loads fast, pages that cover a topic properly, a clean Google Business Profile, real reviews, and links from places that matter. We wrote about whether AI replaces SEO outright and the short version holds up: it has replaced tasks, not the work.

What is new is that a language model has to be able to pull a usable answer out of your page without reading the whole thing. That is a formatting problem before it is a writing problem, and most sites fail it.

The structure AI engines can actually use

Put a direct one-sentence answer immediately below the heading. If the heading asks how much a service costs, the next sentence gives a number or a range, in plain language, before any context. The engine takes that sentence. A reader who is skimming takes it too.

Open longer pages with a TL;DR paragraph that answers the question outright. Give the actual answer in three or four sentences at the top, rather than a preview of what the page is about to cover.

Then write real FAQs. Questions your customers ask, answered in one to three sentences each, factual, no sales pitch. An FAQ that says "Can our team help you?" is filler. An FAQ that says what a permit costs in Potter County is a citation waiting to happen.

None of this is difficult. It is just different from writing a page that was designed to hold a reader's attention for two minutes.

Schema is the part most people skip

Schema markup is code that tells a search engine what your page is about in a format built for machines: this is a business, this is its address, these are its services, this is a question and this is its answer.

It intimidates people. It looks like development work, so it gets pushed down the list, and the effort goes back into the thing that feels familiar. That is usually keyword stuffing, which stopped working years ago and works even less well on a system that reads for meaning rather than repetition.

We are starting to see more businesses adopt FAQs. Far fewer pair them with proper schema markup, and the pairing is what performs. FAQs give the engine an answer. Schema tells it that the answer is an answer.

What moved the needle for AW Broadband

AW Broadband is an internet provider in Amarillo, and their AI citation numbers are the clearest before-and-after we have.

In August 2024 they had zero citations in AI answers. Two things happened next. We implemented comprehensive schema coding across the site, and we resubmitted a batch of URLs through Google Search Console. After that it took on a life of its own.

By May 2026 they peaked at 418 citations. As of July they sit at 94, with 49 brand mentions and an average position of 8.6 in AI answers, all measured in SE Ranking.

We are publishing the drop along with the peak because it is the honest picture and because it tells you something. AI citation counts move. They are not a ranking you hold. The engines change models, re-weight sources, and reshuffle who gets named, and any agency showing you a number that only climbs is showing you a slice.

We still do the off-site work every month, and it matters. But on this account it functions as maintenance. The schema and the resubmission were what broke it open, and that was genuinely new to us.

Watch where the engines get their data

In July, Axios reported details of a deal between Yelp and OpenAI that lets ChatGPT surface Yelp reviews, star ratings, photos and business information inside its answers to local questions, with Yelp branding and links alongside. Yelp had disclosed the agreement in its annual earnings release back in February. The companies plan to add Yelp's Request a Quote feature so people can contact a service business without leaving ChatGPT.

Read that as a business owner. Your Yelp reviews are now feeding the model that people are asking for plumber recommendations. If you have been treating Yelp as a legacy directory you check twice a year, that assumption just expired. We signed on as a Yelp partner agency because of this.

The wider point is that AI engines are buying and licensing data, and every one of those deals redraws who gets recommended. Most agencies in markets like Amarillo and Lubbock are not watching the announcements. They are running tactics that worked in 2016 and waiting for someone to tell them the rules changed.

Searches are turning into conversations

The query itself is changing shape, and this is the shift we find most useful to explain to clients.

People used to type "plumbers in Amarillo," scan a list, and pick. Now we see something closer to a back-and-forth: "who are some reputable plumbers with a money-back guarantee in Amarillo, TX," followed by more questions about pricing, availability, and warranty terms.

That is a better question. The trouble is that the businesses named in the answer are not always the reputable ones. They are the ones whose information the engine could find, parse and trust. Being reputable and being easy for a machine to retrieve are two different problems, and at the moment the second one decides who gets named.

This is where the work pays off, and it is also why AI answers now shortlist local businesses before anyone sees a results page at all. If your money-back guarantee lives in an image on your homepage, it does not exist as far as the model is concerned. If it lives in a sentence under a heading, backed by schema, it does. We have written more about where local search is heading if you want the longer view.

Where to start

Take it in this order.

Get schema on the site first, sitewide and accurate rather than a plugin default. Then restructure your highest-value pages: one-sentence answer under each heading, TL;DR at the top, FAQs that answer real questions. Resubmit those URLs in Search Console rather than waiting to be recrawled. Keep building reviews, and treat Yelp as a live data source rather than an old listing.

Then change how you judge it. Traffic is a poor measure now, because a customer can read your answer inside ChatGPT and call you without ever loading your site. Watch calls, forms and booked jobs. Those still count the way they always did.

In our market, the businesses pulling ahead on this are mostly the ones who started restructuring early, while their competitors were still arguing about whether any of it was real. Budget has had less to do with it than timing.