In 2026, AI answers stopped being a novelty and became the place a lot of customers actually make decisions. Getting named inside a ChatGPT reply, a Gemini shopping comparison, or a Google AI Mode answer, what we call AI citations, now matters as much as ranking on page one used to. But it does not work the way classic SEO does. Here are the six things a year of doing this taught us, and what we would tell any store owner or marketing lead heading into next year.
We build and market webshops for a living, so we watched this from the inside all year: clients getting pulled into AI answers, clients quietly left out of them, and a lot of dashboards that said everything was fine while the real action moved somewhere the dashboard could not see. These are the lessons that stuck.
You measure AI citations by running the real questions your customers ask through ChatGPT, Gemini, and Google AI Mode every month and logging who gets named. Your rank tracker cannot do this, because a strong Google ranking does not mean you appear in AI answers.
The biggest early mistake, ours included, was assuming a good Google ranking meant good AI visibility. It does not. A page can sit at position two for a category and never once get named in an AI answer, while a competitor who ranks lower gets recommended in every reply.
Rankings and citations are different signals. Your rank tracker cannot see whether ChatGPT recommended you, because that conversation never touches a results page. So you have to go and look on purpose.
The practical version is unglamorous. You build a set of the real questions your customers ask, the buying questions, not the keyword phrases. Things like “best winter tires for a small SUV in Croatia” or “which webshop platform is easiest for a first store.” Then you run those prompts across the engines every month and log who gets named, how often, and in what order. That log becomes your baseline. Without it you are optimizing blind.
What we would tell you: treat AI citations as a metric with its own tracking, not a footnote under SEO. If you are not measuring it, you cannot claim you are winning it.
They cite different sources because they pull from different data. Gemini and Google AI Mode rely on Google’s index and shopping data, while ChatGPT leans on its own browsing, data partners, and third-party sites like marketplaces and roundups.
This is the lesson that changed how we work. It is tempting to bundle everything into one “AI” line in a report. That hides almost everything that matters, because the three big engines pull from different places, cite differently, and reach different people.
We saw this clearly with an automotive parts client. Their product feed was clean, so Gemini surfaced them in comparisons without much trouble. In ChatGPT they were nowhere, because ChatGPT was leaning on a handful of enthusiast sites and marketplaces where the brand barely appeared. Same company, same month, two completely different outcomes. If we had averaged those into one score we would have missed the whole problem.
What we would tell you: track each engine on its own. Different sources feed them, different audiences use them, and your fixes will be different for each.
Yes, but only when it marks up content that directly answers a specific customer question. Adding more schema alone does not get you cited. Clear answers, structured so a machine can find them, do.
We have written a lot this year about schema markup, FAQ sections, and clean product data. That work paid off, and in 2026 it stopped being optional. Machines read structure, not vibes. If your content is a wall of text with the answer buried in paragraph nine, it will not get lifted into an answer.
The surprise was that it was never really about adding more schema. It was about matching a piece of content to the exact question a person asks, then marking it up so the engine can grab it cleanly. FAQ blocks written as genuine customer questions, with a direct answer in the first sentence, got quoted almost word for word. Vague marketing copy got ignored. The same lesson applied to product pages, where descriptions that answered real buyer questions got picked up and filler did not.
A simple example: one store had a returns and delivery FAQ written the way a customer would actually ask it, with the answer stated plainly up top. That answer started showing up inside AI Overviews for delivery questions in their category. Nothing exotic happened. The content just answered the question directly and was structured so a machine could trust which part was the answer.
What we would tell you: write for the question, put the answer first, keep each answer factual and self-contained, then mark it up. That combination is what gets cited.
Yes. If AI crawlers cannot reach your site, you will not be cited, no matter how good your content or schema is. The most common causes are AI user-agents blocked in robots.txt, JavaScript-heavy rendering, login or cookie walls, and slow servers.
More than once this year, the entire problem turned out to be access. The content was fine. The schema was fine. The site simply was not reachable by the crawlers that feed AI answers.
We watched a store drop out of ChatGPT citations after a migration disallowed an AI crawler by mistake. Google traffic barely moved, so the alarm never went off. It took someone actually asking ChatGPT about the category to spot that the brand had vanished from it. A newer piece of the puzzle is llms.txt, an emerging file some sites use to guide AI crawlers toward the content they want surfaced, which is worth understanding even if you decide not to use it yet.
What we would tell you: audit your robots.txt for AI user-agents specifically, confirm your important content renders without heavy scripting, and make blocking or allowing AI bots a deliberate decision, not an accident.
Mostly brand awareness first, traffic second. Many AI answers are zero-click, so a citation usually pays off through branded searches, direct visits, and assisted conversions rather than referral clicks.
Here is the part that breaks a lot of reporting. AI answers are often zero-click, and AI Overviews have been pulling clicks away from webshops all year. The person gets what they need inside the answer and never visits your site. If you judge AI visibility only by referral traffic, you will conclude it is not working, and you will be wrong.
Being named still does real work. It shapes which brands a buyer considers before they ever type your name into Google. So the win shows up downstream, not in a referral column. We started watching branded search lift, direct traffic, “how did you hear about us” answers on forms, and assisted conversions, and that is where the value of a citation actually appeared.
One brand we work with saw branded searches climb steadily while their direct AI referral clicks stayed tiny. The citation was not sending clicks. It was doing pre-sale persuasion, and the payoff arrived later through other channels. If we had only counted AI referral traffic, we would have killed a channel that was quietly feeding the whole funnel.
What we would tell you: measure AI visibility with brand-lift and assisted signals, not just last-click referral traffic. A citation is often an awareness win before it is a traffic win.
AI cites brands whose information is consistent and verifiable across the web. When your prices, specs, or business details conflict between your site and other sources, engines play it safe and cite a competitor instead.
Engines are cautious. When the facts about a brand are consistent everywhere, they cite it with confidence.
That means the boring work matters more than ever. The same product specs, structured for AI shopping agents, on your site and on the marketplaces you sell through. Consistent business details across your site, your profiles, and directories. Original information that only you publish. Being referenced by other credible sites. All of it tells an engine that you are a source it can stand behind.
We saw the opposite cost real citations. A store had one price and spec set on its own site and a slightly different one on a marketplace. Faced with that conflict, the engine did the safe thing and cited a competitor whose information lined up cleanly. The store was not beaten on quality. It was beaten on consistency.
What we would tell you: keep your brand, product, and entity information identical across the web, publish things only you can publish, and earn mentions elsewhere. Trust is what gets you cited when the engine has to choose.
AI engines cite stores they can reach, read, trust, and verify. Classic SEO asked whether a page can rank. AI citations ask whether an engine can choose you over the alternatives, and whether you can tell when it did.
The stores winning this in 2026 are not the ones with the most content. They are the ones who made themselves easy for a machine to verify and then actually measured the result.
How long does it take to start getting cited by AI? It depends on what is holding you back. Fixing a blocked crawler or a data conflict can show results within weeks, once engines recrawl your site. Building the trust signals that get you cited consistently, like third-party mentions and original content, usually takes months.
Is optimizing for AI search different from SEO? It overlaps but it is not the same. Good SEO foundations help, but AI citations also depend on whether crawlers can reach you, whether your facts are consistent across the web, and how directly your content answers a specific question. You can rank well and still never get cited.
Should I block AI crawlers? That is a deliberate business decision, not a default. Blocking them protects content but removes you from the answers customers increasingly rely on. Most stores that want to be discovered should allow the major AI crawlers, and should check they have not blocked them by accident.
What is llms.txt and do I need it? llms.txt is an emerging file that tells AI crawlers which content on your site matters most. It is not a standard yet and no major engine has confirmed it affects citations. It is worth understanding, but fixing crawler access and consistent product data will have a bigger impact first.
Which matters most, ChatGPT, Gemini, or Google AI Mode? It depends on your audience and your category, which is exactly why you should track them separately. They pull from different sources and reach different people, so treating them as one channel hides where you are actually winning or losing.
