Chapter 5: GEO, AEO, LLMO: What Has Really Changed?

Last updated: November 1, 2026.

“GEO services” in an agency proposal, “AEO strategy” on a conference agenda, an “LLMO score” in a tool’s sales pitch. Over the past two years, acronyms like GEO, AEO, LLMO, AI SEO and AI visibility have multiplied, and each one gets presented as a separate discipline.

There’s a fair question behind all this. As the first four chapters showed, search is changing, AI writes answers in its own words, and the platforms work differently from one another. Something has clearly changed. Whether that change needs a new name, or whether the same work is being sold under a new label, is a separate question. The checklist at the end of this chapter is there to help you weigh any “GEO advice” that lands in front of you.

1. One job, many names

GEO stands for Generative Engine Optimization: work aimed at appearing more often, and in a better light, in AI-generated answers. “Generative engine” here means systems that write the answer themselves, such as ChatGPT, Perplexity and AI Mode. AEO, Answer Engine Optimization, focuses on systems that give a direct answer instead of a list of links. The term was in use before AI search, back when people talked about featured snippets and voice assistants.

LLMO, Large Language Model Optimization, looks at the model itself rather than the search engine, and also covers how a brand is represented in the model’s memory, meaning its training data. AI SEO is also the title of this book: a general term for treating visibility in AI search as part of SEO.

Google’s guide to AI search gives these terms their own section. It defines AEO as “answer engine optimization” and GEO as “generative engine optimization,” and says both are used to describe work focused specifically on improving visibility in AI answers (Google Search Central).

The differences between the definitions are mostly about emphasis. AEO looks at the answer, GEO at how it’s generated, LLMO at the model. All three deal with the same question: when AI answers a question, does my brand show up in that answer correctly? So many terms are floating around because the field is new and moving fast; there aren’t separate disciplines hiding behind them.

2. The paper GEO came from

The term GEO came out of an academic paper. Researchers from Princeton University and IIT Delhi first published “GEO: Generative Engine Optimization” on November 16, 2023, and it was accepted to the KDD conference in 2024 (arXiv).

The paper claims that content creators can raise their visibility in AI answers by up to 40% by editing the text on their pages in certain ways. To measure this, the researchers also built a test set called GEO-bench, with questions from different fields.

The results for the methods they tried are interesting (arXiv). Citing sources (Cite Sources), adding quotations (Quotation Addition) and adding statistics (Statistics Addition) worked best, giving relative improvements of around 30% to 40% on the paper’s own visibility metrics. Improving fluency added 15% to 30%. Keyword stuffing didn’t work and actually did worse than the baseline. The researchers write that methods like this from classic SEO brought little or no improvement in AI answers.

These results get quoted a lot, but it’s worth looking at how the experiment was set up. In the main experiment the researchers didn’t use a real search engine but a simulation: a two-step system that retrieved the top 5 results from Google and wrote the answer with gpt-3.5-turbo. They also tested Perplexity as a real platform on a subset of 200 questions; there the best method was adding quotations, with a 22% improvement.

That setup has a consequence that’s easy to miss. The paper measures how much an already-retrieved page gets used in the answer, not how a page gets retrieved in the first place. In the experiment, the pages were in the top 5 results from the start. As we saw in Chapter 3, in real life the hardest step is often exactly that one: getting the page found in the fan-out searches so it ends up among the pages that get read. That step is still SEO’s job.

Look at the methods the paper found useful and you get a familiar list: back up claims with sources, give concrete numbers, include expert views, write fluently. A good editor would have recommended all of that before AI. There’s one difference, though. In the paper, the statistics and quotations were added to the text for the experiment; on your page, they have to be real. A made-up statistic or a quote with an unclear source just increases the risk of misrepresentation described in Chapter 3, by your own hand.

3. AEO and LLMO

AEO is older than GEO. SEOs used the term for featured snippets and voice assistants too: Google showing a short answer taken from one page at the top of the results, or a voice assistant responding with a single answer. The advice from that period still sounds familiar: state the question clearly in the heading, give the answer in the first sentence, write clear definitions.

With AI answers, AEO’s scope has widened. Instead of a single excerpt from a single page, there’s now an answer synthesized from several sources. The principle of “give the answer clearly” hasn’t changed, and the advice on passage-level reading in Chapter 3 follows from it.

LLMO looks somewhere else: the model’s memory. In Chapter 3 we saw that AI knows things from two sources, live search and training data. GEO and AEO mostly deal with the live search side, while LLMO also accounts for how your brand appears in training data. LLMO’s question is what the model knows about your brand without searching at all. That knowledge comes from what has been written about you on the web over the years, and it’s covered in Chapters 13 and 14, on brand and third-party sources.

Put the three terms next to their classic SEO counterparts and you get this:

Every term has a counterpart in classic SEO. What’s new is how much weight those counterparts carry and how they connect to each other.

4. According to Google, it’s still SEO

Google is unusually direct on this. In its AI search guide, right after defining AEO and GEO, it says: “From Google Search’s perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO.” (Google Search Central)

The guide also gives the reasoning. Because Google’s AI features rely on its core ranking and quality systems, SEO best practices still apply.

What Google employees say out loud points the same way. At the Search Central Deep Dive event in Bangkok in July 2025, attendee Kenichi Suzuki reported Gary Illyes as saying: “To get your content to appear in AI Overview, simply use normal SEO practices. You don’t need GEO, LLMO or anything else.” (Search Engine Land) In Chapter 3 we also saw Danny Sullivan answer the advice to chop content into bite-sized chunks with “We don’t want you to do that.”

The guide also lists, one by one, things you don’t need to do:

• You don’t need to create new machine-readable files, AI text files, markup or Markdown versions to appear in Google Search.
• Creating an llms.txt file is fine, but Google Search doesn’t use these files; it neither hurts nor helps your site’s visibility.
• You don’t need to break content into small pieces so AI understands it better. There’s no ideal page length either.
• You don’t need to write in a special way for AI search. There’s no need to worry that you don’t have enough long-tail keywords or haven’t covered every variation someone might search for.
• Structured data isn’t required, and there’s no special schema.org markup you need to add.
• Chasing inauthentic “mentions” on the web isn’t as useful as it might seem.

Every item on that list matches something that has been sold somewhere as a “GEO tactic” in the past two years.

5. Why Google says this

Google is talking about its own search. The guide says “Google Search’s perspective”; it says nothing about how ChatGPT, Perplexity or Claude work. As we saw in Chapter 4, those platforms use different indexes, and their sources overlap very little with Google’s top 10. “It’s still SEO” is a strong statement for Google Search, but on its own it isn’t a sufficient answer for the other platforms.

Google also has an interest here. It built its AI features on top of its ranking systems and doesn’t want a separate industry growing up around manipulating them. Saying “you don’t need to do anything special” reassures site owners and also reduces the risk of spam.

Having an interest doesn’t make what it says wrong. In Chapter 3 we saw that Google retrieves pages for its AI features with its core ranking systems. And according to the Ahrefs data in Chapter 1, 37.9% of pages cited in AI Overviews rank in the top 10 for the same query. That share is falling, but the explanation for the drop is also a search concept: query fan-out. A large share of the cited pages rank well not for the main query but for the sub-queries.

My conclusion is that Google is technically right to say it’s still SEO. But “still SEO” isn’t the same as “nothing has changed”; the boundaries of the work have widened.

6. What stays the same

Being crawlable and indexable is still the first requirement. All of Chapter 2 was about this, and a page that isn’t in the index can’t be cited in AI Overviews. On the news site I described there, the sitemap was cached for about 10 minutes, so new stories were crawled late. Today the same problem also means showing up late in AI answers.

A site that’s up and fast is just as basic. One of our clients’ sites was down for about an hour; for that hour it couldn’t respond to search engines or AI bots. That’s why at İmza we monitor uptime for every client’s site.

Understanding search intent hasn’t changed either. A page that misreads what the user wants struggles both in classic search and in the sub-searches of query fan-out. Quality and trust, meaning E-E-A-T, also apply in AI answers. If AI relies on a wrong source, it repeats the mistake in its own voice, so the search for trustworthy sources grows rather than shrinks.

Original content keeps its value too. Google’s guide gives two examples. Something like “7 tips for first-time home buyers” is made of general information anyone could write; Google calls this commodity content. A post explaining why the authors skipped the home inspection and how they saved money by having the sewer line scoped, on the other hand, offers first-hand experience beyond the ordinary. Google says that over the long run, this kind of original content will shape your presence in AI search the most.

The spam rules are the same as well. The scaled content abuse policy from Chapter 3 applies to AI answers too. The guide says Google’s core ranking systems focus on quality content, other systems block spam, and its AI features rely on both.

SEO’s foundation is still standing; what has gone up is the cost of gaps in it. As I said in Chapter 2, a mistake now gets paid for twice.

7. What’s new

I think there are seven things that are new in AI search. None of them needs a new acronym, but all of them change the day-to-day work of SEO.

1. The unit of success has changed. In classic SEO we measured success by rank; in AI there’s no rank. In the SparkToro and Gumshoe study from Chapter 3, the chance of getting the same brand list twice for the same question was under 1%. What we measure now isn’t a position but how often, and how, we appear across many answers. Chapter 7 handles this with five metrics.
2. There’s no single engine. Classic SEO was, in practice, Google SEO. Today ChatGPT uses its own crawler, Perplexity has its own index and Copilot relies on Bing. As Chapter 4 showed, their sources differ clearly from one another and from Google’s results.
3. The shape of questions has changed. Instead of short keywords, users ask long questions with constraints. In Semrush’s data, the average length of ChatGPT prompts that triggered a search went from 4.7 words in early 2025 to 8.7 words in early 2026 (Semrush). Prompt Research is being added alongside keyword research; that’s the subject of Chapter 6.
4. Off-site sources carry more weight. In classic SEO, off-site work mostly meant backlinks. AI learns about your brand from Wikipedia, Reddit, YouTube and comparison sites, and cites them directly inside the answer. In Chapter 4 we saw in Profound’s data that each platform leans on different third-party sources; Chapter 14 goes into detail.
5. How your brand is described has become a risk. In classic search, Google showed your title and description. AI describes you in its own words, and sometimes gets it wrong. Checking this is a new part of SEO that needs to be measured; Chapter 13 covers it separately.
6. Measurement tools fall short. In classic search, Search Console gave you rankings, impressions and clicks. Google’s AI report currently gives impressions only, and ChatGPT, Gemini, Perplexity and Claude don’t show site owners their search data. We have to measure visibility ourselves.
7. Agents are coming. Google’s guide talks about AI agents that carry out tasks for users and may visit sites and collect data, and it recommends keeping an eye on protocols such as the Universal Commerce Protocol (UCP) (Google Search Central). When an agent rather than a person reads your site, presenting information in a machine-readable way matters more. We’ll come back to this in Chapter 15.

None of these changes SEO’s foundation, but they widen its scope. An SEO now has to look beyond Google rankings to several platforms, off-site reputation, how the brand is described and a new way of measuring.

8. What the other platforms ask for

None of the platforms’ documentation for site owners asks for a special writing technique, a special file or special markup. What they want is for their bots to be able to reach your site.

OpenAI says any public website can appear in ChatGPT search, and that the requirement is for OAI-SearchBot to be able to crawl it. If you don’t want a page to appear, the method it suggests is a familiar one: noindex (OpenAI). Perplexity says PerplexityBot works to show and link sites in its search results and recommends allowing it in robots.txt (Perplexity). And as we saw in Chapter 2, Anthropic notes that blocking Claude-SearchBot may reduce visibility in search results.

The other platforms’ message is an SEO message too: keep the door open to my crawler, make your page indexable, mark pages you don’t want shown with noindex. The difference is that these rules cover several different bots rather than one. To carry Google’s statement beyond Google, a small addition is enough: It’s still SEO, just for more than one search engine.

9. A filling machine manufacturer’s plan

Take a filling machine manufacturer that builds machines for foaming products, sells in Turkey and abroad, and prices its machines at around €25,000.

A few years ago, this manufacturer’s SEO plan would have looked roughly like this: pick keywords like “filling machine,” “liquid filling machine” and “foaming product filling machine,” optimize product and category pages for them, fix technical issues, write a few blog posts, get links from industry sites, and track rankings and traffic.

Nothing in that plan gets thrown out today. What gets added:

• A list of questions alongside the keyword list. Constraint-heavy questions a purchasing manager would ask an AI, such as “Which machine for 3,000 bottles a day of a foaming product?”, “Which filling machine manufacturers have service in Turkey?” and “Diving nozzle or piston filling?”
• Decision information on the pages. Capacity, suitable product types, service area, lead time and price range should be written out clearly on the page. Information that the old plan treated as nice to have now decides whether you make the shortlist.
• A bot access check. Make sure OAI-SearchBot, PerplexityBot, Claude-SearchBot and Bingbot can reach the site, not just Googlebot.
• An off-site check. The company should be described correctly and consistently in industry directories, on marketplaces, in product videos on YouTube and, if there is one, on Wikipedia.
• A new measurement sheet. Run the question list regularly on each platform and track mention, recommendation, link and accuracy rates.

None of these additions is “special text written for AI.” They’re all a response to the fact that buyers now do part of their decision-making with AI.

10. How to read a list of “GEO tactics”

Some of the GEO advice going around on social media and in tool pitches is useful, some is harmless but unnecessary, and some is harmful. The table below sorts the most common ones according to Google’s guide and known research.

There’s a pattern in the table. Almost all of the advice that works was already known as good SEO and good editing before AI. Most of what doesn’t work, or is risky, promises a shortcut.

11. Which name to use

I called this book “AI SEO” because I see this work as part of SEO and want the name to keep that connection. Saying GEO or AEO isn’t wrong. But a separate name can give the impression that it needs a separate discipline, a separate team and a separate budget.

What gets done under a name matters more than which name is used. When someone offers you a “GEO service” or an “AI visibility package,” ask for clear answers to these questions:

• Which questions are you measuring? Were they chosen with a specific buyer and their constraints in mind, or derived from generic keywords?
• On which platforms, and how many times? Is it a single screenshot, or is each question run on each platform more than once?
• What are you measuring? Only mentions, or also recommendations, links and accuracy?
• How do you diagnose? For a question where we don’t appear, how do you tell whether the cause is technical, content or off-site sources?
• Is the foundation checked? Are crawling, indexing, bot access and JavaScript issues part of the work?

If the answers are clear, there’s measurable work on the table, whatever it’s called. If they stay vague, like “we have a proprietary method,” an old promise may be getting sold under a new label.

I think AI visibility should be treated not as a separate product but as SEO work with broader measurement and diagnosis: decide which questions you want to appear for, measure, diagnose and fix the problem, then measure again. In Chapter 17 we’ll turn this loop from Chapter 1 into a system that runs continuously.

12. After the naming debate

GEO, AEO and LLMO are names for the same work from different angles. Google says that for its own search this is still SEO, and it’s technically right. Still, Google is describing its own search; for ChatGPT, Perplexity and the other platforms that sentence isn’t enough on its own. The scope of the work has also widened: appearance rates instead of rank, several engines, long questions with constraints, off-site sources, being described correctly, and agents.

In practice the first step is finding out which questions you want to appear for. The research and measurement part of the book starts there, in Chapter 6, with the move from keywords to prompts.

Checklist

Use this list when you come across a piece of “GEO” or “AI SEO” advice:

Sources

• Google Search Central: Optimizing your website for generative AI features on Google Search (last updated July 10, 2026)
• arXiv: GEO: Generative Engine Optimization (Aggarwal et al.; first version November 16, 2023, latest version June 28, 2024, KDD 2024)
• ar5iv: GEO: Generative Engine Optimization, full text (for the method and results tables)
• Search Engine Land: Google says normal SEO works for ranking in AI Overviews and llms.txt won’t be used (July 2025)
• OpenAI: Publishers and developers FAQ (help page, checked November 1, 2026)
• Perplexity: Perplexity crawlers (checked November 1, 2026)
• Semrush: ChatGPT search insights (April 7, 2026)

Previous: Chapter 4, Not All Platforms Are the Same · Table of contents · Next: Chapter 6, From Keywords to Prompts