Getting cited by AI: SEO is now a distribution game

For twenty years, brands asked one question: how do I rank first on Google? Today they ask a different one: does AI cite me? ChatGPT, Gemini and Google’s AI Overviews answer users directly. Often without giving them a single reason to click.

SEO is not dying. It is changing shape and breaking out of its silo. It now plays out in influencer marketing, PR and online communities too.

To adapt, you first need to understand the mechanics. How an engine builds its index. Why an AI cites one brand rather than another. Why the consistency of your information matters as much as its quality. The playbook comes after. Applied without understanding, it wears out fast. The operational version, the eleven principles, is the other piece on this site: Getting cited by AI without giving up SEO.

From search to answers

The shock does not come from an algorithm update. It hits the act of searching.

Retrace the path. In the early 2000s, people wondered which tool to use: a web directory, Yahoo, Lycos, AltaVista. By 2010, the question was settled. They went to Google. They typed two or three words, sometimes five or six. Short queries bring volume but stay vague. Queries of five words or more, the long tail, are rarer and far more precise.

Google returned ten blue links. People opened three or four, and their brain did the work: read, compare, summarize, decide. The engine found the documents. The human built the answer.

The industry did go through upheavals, all driven by the same mechanism. Google rewards a signal, SEOs inflate that signal, Google eventually devalues it. Content volume counted: everyone published thousands of 150-word pages, and Panda penalized them in 2011. Links counted as votes: everyone bought them, and Penguin penalized artificial link profiles in 2012.

The paradox of the system: you get penalized precisely because you exploited the loophole well. SEOs switched tactics. Users, meanwhile, kept the same habits.

Now those habits are changing. You ask an AI for the answer, and it does the reading, comparing and summarizing for you. You no longer look up the height of Mount Everest on Wikipedia. The assistant just tells you. You click a source only when you want to dig deeper.

The query itself is changing too. You no longer type keywords. You write a prompt. “I’m going freelance in Austin, which accountant would you recommend?” Ten, fifteen words, context, sometimes a whole conversation. And when the request stays vague, the AI asks you the questions.

The direct result: informational traffic is collapsing. That is not necessarily bad news.

Losing traffic is not losing revenue

Take an online store selling hiking gear. It ranks for “how tall is Mount Everest.” Who types that query? A kid doing geography homework, friends settling a pub quiz, now and then a seasoned climber. Rarely a buyer.

So why build these pages? Because they fed the system. Internal linking, first: a well-ranked informational page earns links, then passes that authority, the PageRank, to the product pages it points to. Retargeting, next: every visitor passes an ad pixel and joins an audience you can target again. Lead capture, finally: an exit pop-up, a lead magnet, an invite to a WhatsApp group. Their real KPI was never the visit. It was the prospect.

AI breaks this model for a simple reason. A factual answer needs no source. A mountain’s height, a date, a definition: the model already knows them. The user gets the answer without ever passing through your site.

Purchase queries hold up better, at least for now. A likely explanation: choosing a pair of shoes requires up-to-date prices, stock levels and reviews. The AI needs fresh documents, and it links to them.

Those are the pages to watch. Your “women’s trail running shoes” page drops from first place to the bottom of the SERP? Now your revenue takes a hit. That is your conversion page.

Watch another signal too: branded traffic. Follow the journey. The AI mentions a store. The user does not click the source. They type the name into Google or go straight to the site. Your analytics file that visit under branded search or direct, never under the AI that triggered it. Informational traffic falls. Direct and branded visits rise. Google has even started separating branded queries in Search Console.

So hold on to the one metric that does not lie: revenue. Is it holding, or falling off a cliff?

GEO is SEO for new engines

GEO, AIO, LLM SEO: the labels keep multiplying. The substance does not change. Optimizing for an engine means understanding its ranking criteria and applying them.

The industry loves to call it “organic.” The acronym says otherwise: Search Engine Optimization. And an engine is not just Google. It is every search bar your customer uses. Amazon, Airbnb, Etsy, the App Store.

Each of these engines ranks results by its own criteria. On a marketplace, reviews, sales history and listing completeness carry a lot of weight. On Google, links, content and visitor behavior. Everywhere, the 80/20 rule applies. A couple of dozen criteria drive most of the outcome. Master them, and you capture most of the result.

Which leads to the first question to ask: where is your customer? An Airbnb host who still makes a decent margin after fees does not necessarily need a website. They need to rank on Airbnb. A knifemaker selling three models may do better dominating Etsy than launching a Shopify store. A meditation app wants the top spot on the App Store, not on Google.

Google itself has moved the goalposts. The 2024 leak of its internal documentation showed how much visitor behavior counts: clicks, time on page, returns to the results page. To measure it, the search engine has a source nobody else has: Chrome and Android data. Many SEOs read it the same way. Google compares how visitors actually behave with how they are expected to behave on that type of page. AI-written content is not a problem in itself. Google does not care who wrote it, only whether people read it.

GEO adds criteria, not a new discipline. The label still has commercial value. Plenty of marketers think they have mastered SEO, and a new word gets them to look at the topic with fresh eyes. Call it whatever you like. What matters is being visible where your customers search, and selling.

Finally, the buyer journey has splintered. Yesterday: Google, a website, a purchase. Today: a question on Instagram, a TikTok video, a chat with ChatGPT, the brand name typed into Google, then the purchase. Pinning down what tipped the buyer is next to impossible.

How AI decides who to cite

Ask an assistant: “Recommend an accountant for freelancers in Austin.” To cite one firm over another, it relies on sources. Which ones? The answer runs through a concept GEO has put back at the center: the index.

What an index is

An index is a search engine’s catalog. It is built in three stages. First, crawlers travel the web from link to link and download every URL, every HTML document. Next, the engine sorts what it collected: quality content or spam, useful page or duplicate. Finally, it settles the hardest question: quality, sure, but on what topic? An excellent page on starting a podcast is worthless for a query about freelance taxes. So the index matches each document to the queries it answers, then ranks them.

That work, accumulated over more than twenty years, is Google’s head start. It owns the servers, the crawlers, the browser, and above all a battle-tested ranking of billions of documents.

Links: a vote borrowed from academia

To rank pages, Google imported an idea from research. Its founders drew on citations between academic papers: the more a paper is cited, the more it matters. Carried over to the web, every link becomes a vote. The link text, the anchor, tells you what topic the vote is about. That is PageRank. It is also why SEOs spent years buying links.

From words to entities

Google then stopped matching plain strings of characters. Since the Knowledge Graph in 2012, it reasons in named entities: a person, a brand, a place, a product. Since RankBrain in 2015, it has relied on vectorization. Each text becomes a series of numbers, a vector, that represents its meaning. Two texts with similar meanings get similar vectors. Language models rest on the same idea. Nothing new, then. SEOs have been talking about it for a decade. How to become one of those nodes is the other essay: Entity building for AI systems.

Why ChatGPT looks elsewhere

A language model is not an index. It learns from a corpus frozen at training time. It knows nothing about yesterday’s news and cannot rank URLs. To answer on a recent or specific topic, it has to fetch documents from existing indexes, then write its answer from them.

Where does it find them? First, in search engine results, Google’s included. Its index gets scraped by plenty of players, to the point that Google sued a service that collected and resold its results in late 2025. Second, on platforms, each one an index in its own right: LinkedIn, Reddit, YouTube. A video gets transcribed, and the transcript becomes a source. Open an AI Overview. The sources listed often mix LinkedIn posts, YouTube videos and websites.

A Google top 10 ranking guarantees nothing. Pages outside the top 100 show up in AI Overviews. To be fair, studies still disagree on how widespread this is.

A probabilistic model

Now for the heart of the mechanism. A language model does not “know” anything in the human sense. It calculates the most likely sequence of words. If your brand shows up often, across varied sources, tied to the same topic, the link between the two becomes the most likely sequence. Your LinkedIn posts, podcasts, videos and articles all say you are Austin’s go-to accountant for freelancers? The AI will end up repeating it.

The reverse holds too. Faced with contradictory information, the probability scatters, and the model prefers to cite someone else, or no one.

There is a flip side. It is far easier to game than Google, which has spent twenty years training spam filters. Panda and Penguin were their ancestors. AI does not have that memory yet. Repeat a false claim on enough sites, and AI will eventually state it as fact. After negative SEO, get ready for negative GEO: contradictory information spread about a brand to make it vanish from answers.

Tomorrow, one answer per person

Personalization will reshuffle the deck. An assistant tied to your account will know your tastes, your favourite restaurants, your latest trips. It will build a kind of second brain in your image, and stop recommending the same thing to everyone. We are only at the beginning.

The method: create, distribute, lock down

Google and ChatGPT do not rank the same way. One thing connects them, though: distribution. The method follows from that, in three steps.

1. Create proprietary data

Generic information already exists in thousands of copies in the model’s training data. It can reconstruct it without you, and it has no reason to cite you. Yet for years, SEO churned out interchangeable content that could sit on any site in the industry. Those days are over.

Only information that exists nowhere else forces AI to go through a source. That means data only you own. A ranking of the easiest coffee machines to descale, built from your repair tickets. The questions your support team hears every day. A survey of your customer base. Real estate portals figured this out long ago. Some publish an annual list, drawn from their own listings, of the cities where rents are rising fastest. The press picks it up. The backlinks follow.

One risk remains: AI reusing your numbers without naming you. Models already do this at scale. Their makers say little about the texts they were trained on, books and copyrighted content included, and now sell access to the knowledge they drew from them. The fix is to brand every data point. “According to a 2026 study by your brand.” When the name sits in the same sentence as the number, it travels with it every time it is repeated. Skip that attribution, and your numbers will circulate without you.

A study does not have to be rocket science. A polling firm can survey a 1,000-person panel on a reasonable budget. One SEO agency in France did exactly that on how people there use AI. Its study now shows up in ChatGPT’s answers on the subject. With a customer base, you can even become your own polling firm and have AI help you analyze the results.

Proprietary data covers your product feeds too. Tomorrow, an AI agent may read your product pages instead of a human. But it starts from a prompt, and a prompt describes a need, not a spec sheet. It will not look for a 22 by 14 by 9 inch suitcase. It will look for a carry-on that budget airlines accept and that rolls quietly over cobblestones. If your descriptions only list dimensions, the agent will not make the connection. Describe the use, not just the spec. Agentic commerce, AI agents that shop on the user’s behalf, is still in its infancy. No need to overhaul everything, but start polishing titles and descriptions now.

2. Distribute even more than you create

The rule fits in a deliberately excessive formula: 90 percent creation, 90 percent distribution. It follows directly from the probabilistic mechanism. Data published once, on your own site alone, is a weak signal. Picked up by twenty different sources, it becomes a strong association.

Every mention counts twice. It adds a document to another index, and it brings an outside voice. A brand talking about itself carries little weight. A brand others talk about carries a lot. It is the link-as-vote logic, applied to AI.

So mention your study in your podcasts. Turn it into LinkedIn posts, short videos, Instagram content. Get creators to pick it up, invite them on, let them interview you. To the AI, you become the one talking most about the topic.

Borrow other people’s authority, too. The psychology of persuasion has shown it for decades. We trust someone more readily once recognized figures have vouched for them. Interview the most-followed voices in your industry. They share the conversation, and their audience discovers you.

Organic reach, meanwhile, is hitting its limits. Platforms throttle the free reach of brand content. A one-hour podcast might cost a few hundred dollars in studio time and editing. Getting it everywhere can take a hundred times that. The question becomes: who do you pay, and how?

Clipping, having dozens of accounts cut your content into short clips and repost them, is part of the answer. Pay clippers per thousand views, add affiliate links that pay on sales, and hundreds of accounts will circulate your clips. Round it out with Google and Meta ads, plus affiliates on LinkedIn, Instagram, TikTok or Reddit.

3. Lock down consistency

AI hates inconsistency, for a reason we have already seen. When two sources contradict each other, its confidence drops. One directory listing you as closed on Mondays when you are open seven days a week is enough to sow doubt. And doubt can make you disappear from answers.

Check that every piece of information about your brand, products and opening hours matches everywhere: your site, local listings, directories, marketplaces, social media. SEO tools now monitor this consistency and flag sites publishing contradictory data.

Measuring when the click disappears

Nobody ever asked a billboard company how many website visits a billboard drove. Tomorrow’s AI answer works the same way. It puts your brand in front of people without generating a click. Yet billboards, TV and radio have never stopped paying off. The issue is attribution.

Why is it getting harder? Follow a buyer. They open your email, visit your site, see a retargeting ad, come across your brand on TikTok, hear it mentioned on a podcast, then find it again in a ChatGPT answer. The last click, often a branded Google search, takes all the credit. The earlier steps, the ones that built trust, stay invisible.

That trust takes time. A prospect may follow you for two or three years, listen to dozens of hours of your content, then message you one day: “So how do we work together?” No tracking tool can rebuild that path.

Is your performance still judged on traffic? Change the metric, fast. Traffic will soon tell you very little.

Large companies have advanced attribution systems. For everyone else, a few simple methods will do.

• Switch off a channel and see if revenue drops. It is the most direct way to measure what that channel really brings in.
• Ask customers how they found you: on the phone, in a post-purchase form, by email. Be specific. ChatGPT, or Google’s AI answer?
• Give each channel its own promo code.

One major blind spot remains. Nobody knows what users actually type into assistants. Search Console shows you the queries that lead to your site. No equivalent tool shows you the prompts. You will need to watch your customers at their screens, the way sociologists once followed families around their kitchens.

Start with your personas, in the copywriter’s sense. Not “35-year-old urban professional, high income,” but their deepest desire, what keeps them up at night, what stresses them out in the morning. Call twenty of your best customers. Ask them why they choose you, what phone they use, whether they live on WhatsApp or Instagram. Focus on the two or three channels they mention. You do not need to be everywhere. A consistent presence on a few channels is enough to feed the AI.

Then query ChatGPT and AI Overviews yourself. Look at which sources they cite, and ask yourself why. Often it is neither your site nor your competitor’s. It is a third-party comparison page. That makes sense. It answers the question “which one should I pick?” and gathers several brands on a single page. Call its publisher and ask what it would take to get listed.

Acquisition without silos

It all comes back to the same mechanism. AI cites what it finds often, across varied sources, phrased the same way. So the SEO team cannot work alone anymore. Influencers, PR, ads, email, communities: every channel now feeds AI answers. The real job is organizing all of it around a single piece of content, adapted and distributed everywhere.

AI assistants will cite a brand in your industry. The only question is which one.