What is AI SEO (and LLM SEO)?

What does AI SEO look like on a real brand?

AI SEO is easiest to see on a real brand, so start with ours. Attensira runs its own crew on attensira.com. We track 61 questions a buyer of an AI SEO product might ask and read the answers on ChatGPT and Perplexity on a daily schedule; in the 30 days from 4 September to 3 October 2026, answers were stored on 24 of those 30 days. These are the numbers from that window, read over the Attensira MCP on 4 October, with the sample size on every row. The table at the top of this page shows three more questions from the same window.

• what is Attensira

  Answers read

  60

  Answers naming Attensira

  60

  Named most often instead

  nobody else

• Attensira vs Profound

  Answers read

  60

  Answers naming Attensira

  60

  Named most often instead

  nobody else

• best GEO tools

  Answers read

  114

  Answers naming Attensira

  0

  Named most often instead

  Profound, in 92

• best answer engine optimization tools

  Answers read

  115

  Answers naming Attensira

  0

  Named most often instead

  Profound, in 90

• AI search visibility software

  Answers read

  109

  Answers naming Attensira

  0

  Named most often instead

  Semrush, in 63

• how to get cited by ChatGPT

  Answers read

  117

  Answers naming Attensira

  0

  Named most often instead

  LLMPulse, in 4

Across all 61 questions, the answers named Attensira in 4.1% of 4,374 answers. Every one of those mentions came from the three questions that already contain our name. On every unbranded question, the number is zero.

That table is the whole case for AI SEO in six rows. A brand can be known to the assistants, named every time someone asks about it directly, and still be absent from every answer where a buyer is choosing. The branded number feels like visibility. The unbranded number is where the customers are. AI SEO is the work of moving the second number, and we publish ours while it is still zero because a page about measurement that hid its own baseline would not be worth reading.

Two more things the same data shows. First, the leaders differ by question: Profound owns the "best GEO tools" answers, Semrush owns "AI search visibility software", and the how-to questions are mostly answered without naming any vendor at all. There is no single competitor to beat, only answers to win one at a time. Second, the sources matter more than the site. In the same 30 days, the answers to our 61 questions carried 97,149 citations across 3,604 domains. This is where they pointed:

Our own site was one of the smallest sources in our own answers. Most of AI SEO happens on pages someone else owns.

AI SEO is search engine optimization for the answer, not only for the list of links. When a buyer types "what is the best expense tool for a 20-person agency?" into ChatGPT, the assistant writes a paragraph that names two or three products. AI SEO is everything you do so that one of those names is yours, and so that the link under the paragraph points at your page.

The phrase has several spellings. LLM SEO, LLMO, AI search optimization and AI search engine optimization all describe the same work, and Wikipedia's article on the field records that usage varies across practitioners, vendors and publications with no consensus definition. This page treats AI SEO and LLM SEO as one term and says plainly where its neighbours differ.

What does AI SEO change about search?

Classic SEO competes for a position in a ranked list that a person then reads. An AI answer is prose. The assistant reads some pages, writes a summary, and attaches a few links, or none. There is no position seven. You are either in the paragraph or you are not.

That changes three things about the job.

The unit of success is a sentence, not a position. The question stops being "where do we rank for this keyword?" and becomes "when a buyer asks this question, does the answer name us, and does it link to us?" Those are two separate events. An assistant can name a brand it never links, and it can link a page on your domain in an answer that never names you.

Most of the deciding happens on other people's pages. Assistants lean on review sites, comparison articles, documentation hubs, forums and Wikipedia. Your own site is one source among many, and usually a small one. The work therefore runs off-site as much as on-site.

The click matters less, and is rarer. Pew Research Center tracked 68,879 Google searches by 900 US adults in March 2025. On pages with an AI summary, people clicked a traditional result in 8% of visits, against 15% on pages without one, and clicked a link inside the summary in 1% of visits. If the answer is where the decision is made, being in the answer is the goal in itself.

Is AI SEO different from GEO, AEO and LLM SEO?

Mostly in emphasis. All of these terms describe overlapping work, and the honest answer is that the vocabulary is not settled.

• AI SEO, LLM SEO, LLMO

  Where it comes from

  Practitioner and vendor usage

  What it emphasises

  The whole job of being found in AI answers, as an extension of SEO

  Our page

  This page

• Generative engine optimization (GEO)

  Where it comes from

  A 2023 paper by Aggarwal and co-authors, published at KDD 2024

  What it emphasises

  Measured content edits that change visibility inside generated answers

  Our page

  Generative engine optimization

• Answer engine optimization (AEO)

  Where it comes from

  Practitioner usage, older than LLMs

  What it emphasises

  Being the passage an answer engine quotes

  Our page

  Answer engine optimization

• AI search visibility

  Where it comes from

  Vendor usage

  What it emphasises

  The measurement: how often answers name and cite you

  Our page

  AI search visibility

Generative engine optimization is the only one of these with a citable origin, a paper that defined the problem and measured which edits worked. If you need a definition you can point at, GEO is the one. If you are hiring someone, AI SEO is the word most buyers search for.

Forrester's Nikhil Lai wrote in November 2025 that AEO is significantly, but not fundamentally, different from SEO. We agree with the first half. The tactics overlap heavily. What is genuinely new is the measurement problem, and the fact that the deciding pages are mostly not yours.

The glossary owns the short definitions. A large language model is the kind of model behind these assistants. AI search is the broader shift to answers that are written rather than ranked. A source citation is the link an assistant attaches. Search everywhere optimization is the wider idea that buyers search across assistants, video and communities, not only Google.

What work does AI SEO involve?

Four kinds of work, in roughly the order they pay off.

1. Let the assistants read you

An assistant can only cite a page its search system has fetched. Each operator publishes separate crawlers for separate jobs, and blocking the wrong one costs citations. OpenAI documents that a site which disables OAI-SearchBot will not be shown in ChatGPT search answers, though it can still appear as a navigational link. GPTBot, by contrast, collects content for training, and blocking it does not remove you from ChatGPT search.

For Google, the rules are the old ones. Google's Search Central documentation says a page must be indexed and eligible to be shown with a snippet to appear as a supporting link in AI Overviews or AI Mode. It also says there is no special schema.org markup, no AI text file and no new machine-readable file you need to add. Anyone selling you a mandatory "AI file" for Google is selling something Google says you do not need.

The AI crawler guide covers every operator's agents, what each one is for, and what allowing or blocking it costs.

2. Publish the page the answer needs

Assistants answer questions, so the pages that get used are the ones that answer a question directly, early, in a passage that still makes sense when it is lifted out of the page. That means a plain answer near the top, numbers with their sources and dates, comparisons that name the alternatives fairly, and headings phrased the way people ask.

Google says AI Mode and AI Overviews may use a query fan-out technique, issuing several related searches across subtopics before writing the answer. In practice, one question becomes many smaller ones, and a page that answers the sub-questions too gets more chances to be read. The AI search visibility guide walks through how assistants find pages and what makes a passage survive extraction.

3. Be present on the pages that decide the answer

This is the part classic SEO plans rarely budget for. When an assistant names a product, it is usually drawing on someone else's page: a roundup, a comparison, a README, a forum thread, a directory listing. The work is to find which pages decide each answer, and then to get onto them honestly, by pitching the author, listing the product where listings are open, contributing to the repositories and threads people actually read, or publishing the original page that does not exist yet.

4. Measure answers, not rankings

Rank tracking does not translate. An answer has no stable position, and the same question asked twice can come back with different names. So the measurement unit is the answer itself: ask a fixed set of buyer questions on each engine every day, and count how often you are named, how often your domain is linked, and who is named instead. Every number needs its sample size, and a change only counts when it is larger than the noise. The GEO measurement guide is the long version.

How do you measure AI SEO without fooling yourself?

Three rules cover most of the mistakes.

Track the unbranded questions. A prompt that contains your brand name measures whether the assistant knows who you are. It does not measure whether a buyer finds you. Our own table above is the example: named in every answer to the branded questions, and in none of the rest. Build the set from the questions a buyer asks before they know your name.

Read the sample size before the percentage. A rate built from three answers moves a third of the way every time one answer changes its mind. Attensira reports every rate with its n and tests every change with a two-proportion z-test at 95% confidence. A change that does not clear the test is reported as "nothing we can prove yet", never as a movement.

Keep "not measured" apart from zero. An engine you never asked is not an engine that ignored you. Reporting an unasked engine as a zero invents a failure. The AI search visibility page goes through which numbers survive scrutiny and which are constructed.

Who does AI SEO: you, an agency or software?

Three honest options.

You. The method above needs a list of buyer questions, a spreadsheet, the free crawler and passage tools on this site, and a few hours a week. It works. What breaks it is cadence: reading several engines every day, finding the deciding pages for each question, and then actually writing the pages and the pitches. Most teams do the first reading, see the gap, and never get to the work.

An SEO agency. An agency does the work through people, on a monthly retainer. That is the right shape for some companies. It is also slow, and the reporting is often a monthly deck rather than the answers themselves.

Attensira. Attensira is an AI-native SEO agency run by software: the service of an SEO agency at the price of a tool. The workspace reads the answers to your buyers' questions on a daily schedule, a crew of AI agents finds why you are missing from each one, and the crew does the work: a researcher finds the pages that decide the answer, a writer drafts the page that is missing, a web engineer fixes crawler access and structured data, outreach drafts the pitch to the site that is cited, and an analyst reads what moved with its sample size. Workspaces start in Draft, so nothing is published, sent or written outside the workspace until you approve it in one inbox. You pay in credits for the work done, up to a budget you set. How it works shows the loop from day one.

What objections should you raise before paying for AI SEO?

"Can anyone guarantee I will appear in ChatGPT?" No. Nobody controls what an assistant writes, and a vendor who promises a placement is promising something they do not own. Attensira promises the work, not the outcome, and reports when a change made no difference. Ask any provider what they guarantee, and in which words.

"Isn't this just SEO with a new name?" Partly, and that is fine. The crawling, indexing and page quality work is the same. What is new is that most deciding pages are other people's, and that the result has to be measured by reading answers. If a provider's plan is your old SEO plan with "AI" added to the title, that is the sign to ask harder questions.

"Our brand already shows up when I ask ChatGPT about us." That is the branded number, and it is the easy one. Ask the question a buyer asks before they know your name, ask it twenty times on each engine, and count. Our own workspace shows how far apart those two numbers can be.

"Will AI answers send us any traffic?" Less than a blue link does, if Pew's figures are a guide: a 1% click rate on links inside Google's AI summaries. Being named is still worth something when the answer is the shortlist. Measure mentions and citations as the outcome, and treat referral visits as a bonus you can see in analytics, not as the scoreboard.

"Can't I just add an llms.txt file?" You can, and it does no harm. Google says it needs no AI text file for its AI features. llms.txt is a proposal, not a standard any major engine has said it relies on for ranking or citation. It is a minor technical step, not a strategy.

Which AI engines does AI SEO have to cover?

Each assistant builds its answers from a different mix of pages, so a plan for one does not transfer to the next. Our own readings show how far apart they are. For the question "what is generative engine optimization", ChatGPT's answers leaned on google.com and a specialist GEO site, while Perplexity's leaned on Wikipedia, Coursera and Built In. The two engines barely shared a top source. The same brand can therefore be well placed on one engine and invisible on the other, for the same question, on the same day.

That has three practical consequences.

Measure each engine separately. An average across engines hides exactly the difference you need to act on. Attensira never blends engines into one number, and reports an engine it did not read as not measured rather than as zero.

Plan the work per engine. The page that wins a Perplexity answer may be a Wikipedia section or a course page. The page that wins the ChatGPT answer may be a specialist article that Perplexity never reads. One blended content plan fits neither.

Remember that Google is still an engine. AI Overviews and AI Mode draw from Google's own index, so the Google work is still classic SEO with snippet-eligible pages. The engines and channels pages describe how each assistant finds and cites pages.

What does a month of AI SEO work produce?

The output of AI SEO is ordinary work, done for a specific answer. In a typical month for one brand, the list looks like this:

• New pages that answer a buyer question nobody on your site answers yet, with the answer in the first paragraph and the facts sourced and dated.
• Rewritten passages on pages that already exist, where an answer misstates your price, your product or your category, so the passage an assistant lifts is the right one.
• Technical fixes: robots rules that block a search crawler by accident, structured data that contradicts the visible page, redirects that send crawlers in circles, sitemaps that leave out the pages that matter.
• Off-site work: a pitch to the author of the comparison article an answer cites, a corrected directory listing, a contribution to the repository or thread people actually read.
• A reading of what moved: per question and per engine, with the sample size, and an honest "no change" where nothing moved.

What it should not produce is a monthly deck of screenshots. If the work cannot be named page by page and answer by answer, it is hard to know whether any was done.

How is AI SEO paid for?

Three models exist, and each prices a different thing.

Retainers price people's time. An agency quotes a monthly fee for a scope of work, and the scope is usually fixed for a quarter. You pay the same whether the month produced ten pages or two.

Tool subscriptions price access. A monitoring tool charges for prompts tracked and engines read, and the work of acting on what it shows is yours.

Work-based pricing prices the output. Attensira charges a plan fee and meters agent work in credits, so a month with more work costs more and a quiet month costs less, up to a monthly budget you set. Work it does not deliver is not charged. The pricing page has the current plans.

Whichever model you choose, ask the same question: what exactly will exist at the end of the month that does not exist today, and how will we both know whether it changed an answer?

Where should you start with AI SEO this week?

1. Write down twenty questions a buyer asks before they know your name, in their words, not your category language.
2. Ask each one on ChatGPT, Perplexity and Google, five times each, and note who is named and which pages are linked.
3. Check your robots.txt allows the search crawlers of the engines you care about. The crawler URL inspector shows what each one is allowed to fetch.
4. Pick the three questions where a competitor is named and you are not, and list the pages the answers link.
5. Decide who will do the work each week, and how you will know in thirty days whether it moved anything.

If step five is the hard part, that is the part Attensira does. The free audit runs step two for your domain across several engines and shows who they name instead of you.