Generative Engine Optimization vs SEO

In Google’s own guide, generative engine optimization and SEO are the same craft. The page still has to be worth reading by a person. The idea that you must pick a side is a sales pitch, and it is not how the guide is written.

Cited in this note

40%

Google says answer-engine optimization and generative-engine optimization are still SEO.¹ The page has to be in Google’s list of pages, and it has to be allowed to show as a snippet. Google does not use llms.txt, and it does not ask you to chop the page up in any special way. We keep those lines together because owners are being told to pay for a second program beside the first one. The guide does not describe a second program. It only describes the page you already owed your reader, which is now being quoted inside an answer.

The visibility guide is how you check whether that page can be found and quoted. The answer-engine note lists the methods in one place, if you want them. The llms.txt note covers the file Google says it ignores. That is good to know before someone sells it to you as the gap between the two names.

What changes, and what does not

The 2024 paper says that pages written for a generated answer gained up to 40% in how often they showed up.² Adding quotes, sources, and facts with numbers went with gains of over 40%. On Perplexity the gains went up to 37%. The study also saw a 41% rise in word count, adjusted for position. It saw a 28% rise in how the reader felt about the answer. All of this came from a set of 10,000 questions. So what changes, in that study, is how easy it is to spot a source inside the answer when the writing carries proof. What does not change is the need for a real page that a robot can read and a person can use.

Keep the two layers apart when a pitch blurs them. SEO, the way Google uses the word, gets the page into its list and ready to be quoted as a snippet. The generative layer is what happens when an answer borrows from that page. You can fail the first layer and never reach the second one. You can also get quoted and still fail the business, if the quote told the reader enough and the visit never came. Both failures belong to the same page, so they are not two departments.

Pew found, in July 2025, that people clicked a link 8% of the time when a search showed an AI summary. When it showed none, they clicked 15% of the time.³ SEO still gets the page into the list, and the page still has to be worth the click.

That is the plain split we want owners to remember. Getting into the list is the price of a seat. The click is what pays for the chair time, the staff, and the front desk. A page written only to be summed up will do the first job and skip the second. A page that is only a thin list of keywords will do neither one well. A summary has nothing solid to lift, and a person has nothing solid to trust.

A worked example

Take a small dental clinic as an example, and not as a case with a score. The front desk lead already answers the same questions that people type into the search box. The dentist will not let the website make up promises about care. The questions the owner cares about are the ones new patients ask before they call. They want to know what happens at a first visit. They ask whether it is okay to bring up a cracked tooth. They also ask whether the clinic will see someone who is nervous.

The page they have today is an SEO page in the old, thin way. It has a city name, the words “family dentistry,” a list of services, and a map. It might be in Google’s list, but a snippet from it would be mush. Say someone asks an answer engine about a cracked tooth and a nervous patient. It will skip that page for one that says what the first hour is like. That other page tells you what to bring and what the clinic will not promise. The owner does not need a second site for these readers. She only needs this one page to do both jobs.

What does not change is the floor. The page has to be in Google’s list and ready to show as a snippet. The front desk lead should be able to read the first lines and say, “Yes, that is what I tell people.” The dentist should be able to read the clinical lines without wincing. If those two cannot sign off on the page, no engine should be quoting it. Google’s no to special chopping and to llms.txt fits here too. Breaking the page into bits does not help, and neither does adding a side file. Nothing takes the place of the owner standing behind each sentence.

What changes is the shape of the sentences. The paper is why we say so, though we will not turn it into a promise. In that 2024 work, quotes, sources, and facts with numbers went with gains of over 40%, on a set of 10,000 questions. For this clinic, the honest version is small. Quote the after-care steps you already hand out on paper. Name the rules you really follow when you talk about a cracked tooth, and skip the blog that only sounds like a doctor. If you give a number, it should be one you can explain. A patient may read it in an answer and walk in holding it. Word count and the reader score moved in the study, by 41% and 28%. We do not make those a goal for the owner. We turn them into a simple rule, which is to write enough real detail that a reader can feel the visit before they book.

The click problem sits on top of that. Pew is why we will not let the page stop at what a cracked tooth is. A summary can explain the tooth, but the page has to explain this clinic. First visits start with a talk and not a surprise procedure. Nervous patients can say so when they book, and the front desk lead will set aside a longer time in the chair. The clinic does not do every kind of care, and it will say which kinds it sends elsewhere. A summary that only wanted a definition has no use for those lines, and they are the reason a person still taps through. In the July 2025 Pew look, people clicked 8% of the time when a summary showed, against 15% when none did. So write the lines the summary cannot cover.

Now the jobs again, because this is where projects get sloppy. The owner approves what the clinic claims. The front desk lead checks the practical lines against the phone script, including the list of places the clinic sends people. The dentist owns anything that could sound like a diagnosis or a promise of no pain. A freelancer can tidy the headings, but a freelancer cannot decide where the clinic sends people. Say the generative work and the SEO work are billed as two projects. You will get two drafts that disagree, and the patient will find that out at the front desk.

How we’d run it

We would run one page, one time, for both names.

First, pick the one question the front desk lead answers every week, and write about only that. It might be a first visit, or a cracked tooth, or how the clinic treats a nervous patient, with one page for each question. A page that tries to be the whole clinic will be quoted for nothing in particular.

Second, make the SEO basics easy to see. The page should be in Google’s list, open to anyone without a login, and ready to show as a snippet. It should also start with sentences that answer the question. If a snippet of those lines would embarrass the dentist, they are the wrong lines. Fix them before you worry about any generative layer.

Third, add proof the way the 2024 paper describes it, but only proof you really have. That means a source you would show a patient. It means a quote that is truly yours, or the steps you already print. And it means a number you can defend. The study saw gains of up to 40% when pages were written for a generated answer. It saw gains over 40% tied to quotes, sources, and numbers. On Perplexity the gains went up to 37%. Use the method on your own material, and do not paste the study’s percentages onto the clinic.

Fourth, write for the click. Say what the summary will not know. Tell people how booking works with the front desk lead and what to bring. Say what you refuse to do, and which fear you will take seriously. This is the reply to Pew, because being the source is not the same as being visited.

Fifth, say no to the split invoice. Do not hire one firm for SEO and another for generative work on the same page unless they edit one document. Two plans make two tones, and the answer will quote whichever is cleaner, which may be the one the dentist never saw.

Sixth, read it the way the desk will hear it. The front desk lead reads the practical half out loud, and the dentist reads the clinical half. Then the owner asks, “If this is the only paragraph a summary keeps, is it still true?” If the answer is no, cut until it is. Then publish the one page. The names GEO and SEO can both point to it, but they do not each get a website.

Common mistakes

The first mistake is thinking the two names are rival crafts. Google says they are still SEO, so a clinic that stops its normal work to “move to GEO” has misread the guide. A page that Google has not listed will never be the quoted source.

The second mistake is the side file. Google does not use llms.txt for this. If you buy a file, or a service that starts with the file, the first-visit page stays vague. You have spent the week on the part the guide shrugs off.

The third mistake is adding special structure for its own sake. Google asks for no special chopping of the page. Headings that match real questions help people, and they give an answer a clean edge to lift. Headings that are only bait, with nothing under them, help no one. We would rather see four plain sections a patient can scan than a pile of bits.

The fourth mistake is using the study as a forecast for your town. The paper reports gains of up to 40%, over 40% with quotes, sources, and numbers, and up to 37% on Perplexity. It also reports a 41% rise in word count, a 28% rise in reader feeling, and a set of 10,000 questions. That is what the paper measured. It is not the owner’s sales line, and putting it on the homepage would be the wrong kind of number.

The fifth mistake is ignoring the click. Pew’s 8% with a summary, against 15% without, is the business problem hiding inside the win. You can win a mention that takes the place of the visit. Put the detail that a summary cannot hold on the page, like the person, the things you refuse, and how booking really feels.

The sixth mistake is having two owners for one page. One person edits for keywords and another edits for answers, and neither sits with the front desk lead, so the patient meets the gap. We want one page, one checker at the desk, and one dentist who can veto a line.

What we’d do

We would keep a single page and a single editor, and we would drop the idea that GEO replaces SEO. Our view is that the guide is right and the fight over names is not. Write the page so a person can use it. That is what makes it ready for Google’s list, ready for a snippet, and worth quoting. The 2024 paper says that proof on the page goes with more reach inside generated answers. Pew says the click gets harder once a summary shows up. Both facts point the same way. Be clear enough to be quoted, and rich enough in detail that the quote cannot say it all.

We would spend the owner’s money on the front desk lead’s questions and the dentist’s veto, and not on a second strategy deck. If a firm cannot show the same page doing both jobs, they are selling a fight that Google did not ask for. The page still has to exist for the list of links and for the quoted answer. So make it true first, then make it useful. Leave it where a robot and a nervous patient can both find it.