For most businesses, strong SEO already does most of the work for AI search. GEO is not a separate rulebook. It is a shift in emphasis toward clear answers, real expertise, and clean HTML that machines can read. If your SEO is weak, no amount of GEO will rescue you.
We get this question almost every week. A founder reads that AI search is killing SEO, then reads that GEO is the fix, then gets a proposal from an agency with a shiny new line item on it. The honest answer is less exciting than the pitch, and it saves a lot of money.
We build in Webflow and we watch how our client sites show up in Google, ChatGPT, Perplexity, and Claude. What we see lines up with what Google publishes. The work that gets you cited by an AI answer is mostly the same work that got you ranked five years ago, done more carefully.
GEO stands for generative engine optimization. It means shaping content so large language models pick it up and cite it when they answer a question. The term comes from a 2024 research paper called GEO: Generative Engine Optimization by Pranjal Aggarwal and co-authors, presented at KDD 2024.
That paper is worth knowing about, because it is the source almost every GEO agency quotes without naming. The authors tested content changes against generative search engines and reported visibility gains of up to 40 percent for the strongest methods. The full paper sits on arXiv under the identifier 2311.09735.
Notice what the paper is actually about. It studies how content is written and structured, not some secret file you upload. The winning changes were things like adding citations, adding quotations from credible sources, and adding statistics. That is editing work, not a new technical channel.
Google says there is nothing extra to do. Its Search Central documentation on optimizing for generative AI features states plainly that you do not need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search. Standard SEO practice remains the foundation.
Google’s guidance also sets out the eligibility rule. To show up as a supporting link in AI Overviews or AI Mode, a page has to be indexed and eligible to appear in Google Search with a snippet. There are no extra technical requirements beyond the normal Search technical requirements.
This matters because it is the platform owner speaking about its own product. When a vendor tells you Google secretly ranks AI answers by rules only they understand, you can check that claim against Google’s own documentation in about two minutes. We do this before we accept any new tactic into our process.
Because new names sell. SEO is a crowded market with known price points. A fresh acronym lets an agency reset the conversation and charge a premium for work that overlaps heavily with what they already did. That is a sales strategy, not a technical discovery.
We are not saying AI search changed nothing. It changed a lot. What we are saying is that the change shows up in how you write and structure a page. It does not show up as a separate workstream with its own tools and its own invoice.
Ask any vendor pitching GEO one question. What would you do differently on this page if AI search did not exist? If the answer is a long list of exotic tactics, be curious. If the answer is short and mostly about answering faster and citing better, they are being honest with you.
Three things do most of the work. Answer the question in the first two or three sentences of a section. Name real sources with the organization and the year. Use the exact vocabulary the model expects, so Core Web Vitals rather than a vague phrase like page speed metrics.
The KDD 2024 GEO paper supports the second point directly. Adding quotations and statistics from credible sources was among the highest performing changes the authors tested. Models are trained on text that cites things, so text that cites things reads as more reliable to them.
The first point is about structure. A model pulling an answer needs a clean, self-contained chunk of text it can lift. If your answer is buried in paragraph six after a long story about your founder, it will not get pulled. We write every section so the opening block stands alone.
The third point is about entities. Models connect ideas through named things: tools, companies, standards, and people. A page that names Webflow, Cloudflare, Semrush, and Schema.org gives a model more to hook onto than a page full of generic phrasing. We go deeper on this in our piece on query fan-out and how it changes SEO.
Ranking well now sends fewer clicks. Research published by Pew Research Center in July 2025 found that Google users clicked a search result 8 percent of the time when an AI summary appeared, compared with 15 percent when no summary appeared. That is roughly half the clicks.
The Pew study used browsing data from 900 US adults who agreed to share their activity, looking at searches from March 2025. About 58 percent of those people ran at least one Google search that month which produced an AI summary. So this is not a fringe case anymore.
The honest takeaway is uncomfortable. You can win the ranking and lose the visit. That does not make the ranking worthless, because being the cited source in an answer still builds trust and still sends the highest intent visitors. It does mean judging your SEO purely on raw session counts is now misleading.
Not for Google. Google’s own documentation says llms.txt files and similar resources neither harm nor help your visibility or rankings in Google Search. It is a proposed standard with real supporters, but Google has stated it does not use it. We do not add it for clients as a ranking play.
Schema markup is a different story and worth your time. It is a documented standard that Google uses for rich results, and it helps machines understand what a page is about. That is not a GEO trick. It is basic technical SEO that has been valuable for more than a decade.
If you want the longer version of this argument, we wrote a full breakdown of what llms.txt is and whether you should add it. Our position has not changed. Add it if you like the idea, but do not expect it to move rankings, and do not let anyone bill you for it as a strategy.
Four things. We put a direct answer at the top of every section. We name sources inline with the organization and the year. We keep the article body in the static HTML so crawlers that do not run JavaScript can read it. And we use one consistent term per concept across a page.
That third one gets overlooked, and it is the one that quietly breaks sites. Many AI crawlers do not execute JavaScript. If your content only appears after a client side fetch, those crawlers see an empty shell. We prerender content into static HTML for exactly this reason, and we check it by loading pages with JavaScript turned off.
The fourth one sounds small. It is not. When a page calls the same thing three different names, a model has a harder time deciding what the page is about. Pick the term the industry already uses and repeat it. Consistency reads as clarity.
Start with Google Search Console. Google introduced Search Generative AI performance reports in Search Console in June 2026, and its own optimization guide now points site owners there to measure how content performs in generative AI features. That is the closest thing to a first party number you will get.
Beyond that, the honest answer is that measurement is still rough. Referral traffic from ChatGPT and Perplexity shows up in analytics, but a mention inside an answer with no click leaves no trace in your reporting. Anyone promising a precise GEO score across every model is selling confidence, not data.
We tell clients to track two simpler things instead. First, are you ranking in Google’s top results for the questions that matter, since that is where AI features draw from. Second, when you ask the models your own key questions, do they name you. Both are easy to check by hand. Our guide to optimizing for Google AI Overviews walks through this in detail.
Probably not as a separate hire. If your SEO foundations are solid and your content answers real questions clearly, you already have most of what AI search rewards. Spend the budget on better content and a faster, cleaner site before you spend it on a new acronym.
The businesses we see winning in AI answers are not doing anything exotic. They publish genuinely useful pages, they cite real sources, they load fast, and they have been at it for a while. Across 150+ projects, the sites that get picked up by AI answers are the same ones that were already well built for search.
If you are trying to work out whether your site is ready for this, or you have a GEO proposal in front of you and want a second opinion on it, we’re happy to walk through it with you. Reach out at phoenix.studio and we’ll tell you honestly what we would change and what we would skip.
Tell us where you want to go. We'll tell you how we'd get you there.