Every agency now sells AI SEO. A lot of it is the same retainer with a new label. Here is what the work involves when it is real, what the naming confusion between SEO, GEO, AEO and LLMO actually means, what it costs per month, and how to test whether an agency can do it before you sign.
Every agency that sold you SEO last year sells AI SEO this year. Most of the time the deliverables did not change, only the deck did. That is worth naming up front, because the underlying shift is real even though a lot of the marketing around it is not.
Here is the shift, measured rather than asserted. Pew Research tracked the actual browsing of 900 US adults across 68,879 Google searches. When an AI summary appeared, users clicked a traditional search result in 8% of visits. When no AI summary appeared, they clicked in 15% of visits, nearly twice as often. And the links inside the AI summary itself? Clicked in 1% of visits.
Read that last number carefully, because it reframes the whole exercise. Getting cited inside an AI answer is not primarily a traffic strategy. Almost nobody clicks. What you are buying is being the source the answer is built from, which means your business is the one described, recommended and named when a buyer asks. This post covers what that work actually is, what the competing acronyms mean, what it costs, and how to test an agency before you pay them.
SEO, GEO, AEO, LLMO: what the names mean
There is no settled term for this yet, which is itself a signal about how young the discipline is. The four you will encounter:
- GEO, generative engine optimization. The most widely used term. Optimizing to be cited by generative systems: Google AI Overviews and AI Mode, ChatGPT, Perplexity, Claude.
- AEO, answer engine optimization. Slightly older, and originally about featured snippets and voice assistants. Now used more or less interchangeably with GEO.
- LLMO, large language model optimization. Least common. Emphasizes the model rather than the search product wrapped around it.
- AI SEO. What most buyers actually type, and an umbrella for all of the above. It is also sometimes used to mean the unrelated practice of using AI tools to produce SEO content faster, which is a different thing entirely. Worth clarifying which one an agency means.
Do not spend energy on the vocabulary. Any agency that leads with which acronym is correct is selling you a taxonomy rather than an outcome.
What the work actually involves
The honest version is that classic SEO and AI SEO share most of their foundation. A page an assistant cannot crawl, parse or understand is not going to be cited. Where they diverge is the target: classic SEO optimizes for a ranked position, AI SEO optimizes for extractability. You are writing to be quoted, not clicked.
In practice that means four things.
Answer pockets. Self-contained passages of roughly 40 to 80 words that answer one specific question completely, including the concrete details (a dollar figure, a timeframe, a named entity). An assistant assembling an answer needs a chunk it can lift without surrounding context. A 2,000 word article with the answer distributed across it gives the model nothing clean to take.
Entity clarity. The model needs to resolve who you are, where you are and what you do without inference. That means consistent naming, explicit geography, schema markup that mirrors the visible page content, and a real author with a real profile behind the writing. Vague, geography-free copy is invisible here in a way it is not in classic search.
Comparison and list structure. Assistants disproportionately cite pages that name and compare specific options, because the question they are usually answering is comparative. Tables, ranked lists with position markers, and explicit "best for X" labels are formats models can restructure into an answer.
Third-party corroboration. This is the part agencies skip because it is not billable content work. For local and service queries, assistants lean heavily on directories and on review data from Google Business Profile and Yelp. If a model is assembling a "best X in Y" answer from review counts, no amount of on-site optimization substitutes for having reviews.
What this looks like when it works
A worked example from our own site, because we would rather show a result than describe a methodology.
We queried ChatGPT with web search for “custom software development company near Chino Hills CA” in August 2026. It returned six firms, listed LinkTech Solutions first, cited our homepage and our about page as sources, and closed with: “If being genuinely local to Chino Hills is important, I'd start with LinkTech Solutions and EDISON.”
The pages that earned that were not clever. They were an about page whose title states the city, service pages that name the region explicitly, and a regional landing page dense with the entity information a model needs to place us. Nothing exotic, just unambiguous.
The same exercise also shows the limits. Ask the same assistant for “best web design company in Orange County” and we do not appear at all. That answer is assembled from directories and from review-count-sorted local listings, and we are not in them. On-site work does not fix that. Knowing which of your target queries is winnable through content and which requires something else is most of the strategy.
What AI SEO services cost
Three broad tiers in the market right now.
- $500 to $1,500 a month, content-led. A defined number of published pieces per month, structured for citation, with the technical foundation handled once up front. Realistic for a small business with a working site.
- $1,500 to $6,000 a month, full program. Content plus technical work, schema, entity cleanup, directory and review infrastructure, and ongoing measurement against the assistants. This is where most mid-market engagements land.
- $6,000 and up. Multi-location, multi-service, or competitive national categories. At this level you should expect named strategists and custom measurement, not a template.
One pricing red flag worth stating plainly: nobody can guarantee an AI citation. There is no placement to buy, no submission process, and no relationship with an assistant that produces one. An agency promising guaranteed citations for a flat fee is either misunderstanding the mechanism or counting on you not to check.
How to tell a real offering from a rebrand
Four questions. They take about ten minutes on a sales call and they are hard to bluff.
- “Show me a before and after assistant response.” Not a ranking report. The actual answer a model gave before the work and after. If they cannot produce one for any client, they are not measuring this.
- “Which of my target queries are winnable with content, and which are not?” A real answer distinguishes between queries assembled from web content and queries assembled from directories and review data. An agency claiming everything is winnable with blog posts has not looked.
- “What are you doing about my Google Business Profile and directory presence?” If the answer is nothing, the program cannot touch local “best of” queries at all.
- “How does this change what you actually publish?” Listen for answer pockets, schema mirroring visible content, comparison structure and named authorship. Listen for whether “keyword density” comes up, which tells you the deck changed and nothing else did.
How to measure it
Measurement is genuinely harder here than in classic SEO, and most reporting glosses over that.
Search Console does not separate AI surfaces from ordinary results. Google folds AI Overviews and AI Mode impressions into the standard Web search type with no filter to isolate them, so anyone showing you an “AI impressions” chart sourced from Search Console is showing you something the export does not contain.
What works is direct: build a list of the prompts a real buyer would type, run them against ChatGPT, Google AI Mode, Perplexity and Claude on a schedule, and record whether you are named, whether you are cited as a source, and where you sit relative to competitors. It is unglamorous and it is the only honest version. Track the competitor set too, because the useful question is rarely whether you appear, it is who appears instead of you and why.
Where to start
If you are doing nothing right now, the order that produces results fastest is: fix entity clarity on the pages you already have, then build the answer-pocket content for the questions your buyers actually ask, then work on the third-party layer of directories and reviews. Most people do that in reverse and wonder why the content is not landing.
The longer technical treatment is in our guide to LLM SEO, the measurement side is covered in AI visibility, and if you are comparing this against a classic retainer, what SEO costs has the traditional numbers. We do this work ourselves through our SEO service and through monthly blog writing starting at $500 a month.
Frequently asked questions
What is AI SEO called?
There is no settled term. Generative engine optimization (GEO) is the most common, answer engine optimization (AEO) and large language model optimization (LLMO) also circulate, and many agencies just say AI SEO. They all describe the same goal: being the source an AI assistant cites when it answers a question, rather than a blue link someone has to click.
How much do AI SEO services cost?
Monthly retainers for small and mid-market businesses generally run $1,500 to $6,000. Content-led programs aimed specifically at earning citations start lower, around $500 to $1,500 a month for a defined number of published pieces. Anything promising guaranteed AI citations for a flat fee is selling something nobody can deliver, because no assistant sells placement.
Is AI SEO different from regular SEO?
It overlaps heavily. Crawlability, structured data and topical depth matter for both. The difference is what you optimize toward. Classic SEO optimizes for a ranked position. AI SEO optimizes for extractability: short factual passages, unambiguous entity information, comparison tables and schema an assistant can lift straight into an answer.
How do you measure AI SEO results?
By querying the assistants directly and recording whether your domain is cited, for which prompts, and in what position. Classic rank tracking does not capture it, and Search Console blends AI surfaces into ordinary web impressions with no way to separate them. If an agency cannot show you before and after assistant responses, they are not measuring this.