In short: In an AI answer you can be visible in two ways: Your brand can be mentioned in the text itself, and your website can be linked as a source. Being mentioned is built on your reputation across the whole web. Being cited is built on the quality and readability of your own pages. They are two different competitions with two different rulebooks, and most businesses have no idea which one they are losing. This guide teaches you to read an AI answer the way we do when we measure.
The anatomy of an AI answer
Ask ChatGPT or Google AI Mode a buying question, for example "what is the best mobile plan in Norway", and look closely at what comes back. The answer has a fixed anatomy, regardless of platform:
- The text itself, where a few brands are singled out by name, usually with a reason: One is recommended for unlimited data, another for low prices, a third for coverage.
- The source references, small links inside the text or gathered at the end, showing which websites the model built the answer on. They are often review sites, expert articles and comparison services, not the vendors themselves.
We use exactly this example in the visualisation on our front page, because it shows something most people miss when they skim AI answers: The names in the text and the links in the source list are rarely the same. The plans being recommended are mentioned. The comparison sites writing about them are cited. Two forms of visibility, two completely different positions in the customer's decision.
Whoever gets mentioned wins the recommendation. Whoever gets cited wins the click and the authority of being the source. Both positions are valuable, but they are earned in entirely different ways. To understand why, we need to look at where the answer actually comes from.
Where the answer comes from: Memory and research
When an AI platform answers a question, two things typically happen, and the split between them is the key to this whole guide:
The model uses what it has learned. Language models are trained on enormous amounts of text, and brands that are well described, frequently covered and consistently presented in that material sit in the model's muscle memory. When the model writes that a certain company "is highlighted as a test winner", it is in effect repeating a consensus it has read across countless sources over time.
The model searches and reads in real time. Most platforms also fetch fresh web pages as they answer, pick a handful of them and use them as sources, with links. Here your website competes to be chosen, read and quoted, against every other page about the same subject.
Mentions mainly come from the first mechanism: The model's learned picture of the market, supplemented by what the sources it reads say about you. Citations come from the second: The real time selection of which pages are good enough to build an answer on. That is why the two forms of visibility can diverge completely, and why they demand different work.
Being mentioned: The reputation competition
When a model recommends a brand in the text, it is not because the brand has a nice website. It is because the name keeps appearing, with the same message, in the sources the model trusts: Reviews, articles, trade press, forums, directories and reference works. The model essentially does what a thorough journalist would do: It checks what many independent sources say, and repeats the pattern it finds.
That means mentions are mostly built outside your own website:
- Third party coverage. Reviews, industry roundups and articles where you are evaluated by others. A spot on a credible "best in category" list is among the most valuable assets in AI search, because it is exactly the kind of source models lean on when they recommend.
- Consistency. The same name, the same description and the same facts everywhere you are mentioned, from your own profiles to directories and press coverage. Conflicting information makes the model uncertain, and an uncertain model leaves a name out rather than gamble on it.
- A sharp position. Models love attaching a name to a concrete strength: "best for unlimited data", "fastest delivery in the region", "specialists in small businesses". A clear position gives the coverage something to say about you, and gives the model a reason to pick you for the answer.
Notice how little of this is classic SEO. It looks more like PR and brand building, except the reader is a machine that reads everything, remembers everything and cannot be charmed in a meeting. What is written about you is all it has.
Being cited: The craftsmanship competition
The source links follow a different logic. There, the model picks pages it can build an answer on, in practice while the user waits. The pages that win are the ones that are easy to fetch, easy to understand and easy to quote:
- Technical accessibility. The page must be crawlable, load fast and deliver its content as plain HTML. Fail here and nothing else matters. The whole foundation is covered in our guide on technical SEO for AI search.
- Answerable structure. Clear questions as headings, concrete answers right below, numbers, tables and summaries. Pages that have already done the work of formulating the answer get preferred as sources, simply because they are easiest to cite. The craft is described in our guide on AI friendly content.
- Documented credibility. A named author with a real background, a visible update date, sources and structured data. Models are trained to prefer pages that look accountable, because those are the pages that make their answers correct.
The interesting thing about the citation competition is that you can win it even if you are small. The model evaluates the page, not the company name. A thorough expert article from a local player beats a thin page from a big competitor, every time. That makes citations the part of AI visibility where effort pays off fastest and most predictably.
Four positions: Where do you stand?
Put the two measures together and every business lands in one of four positions, each with its own diagnosis:
| Position | What it means | What to do |
|---|---|---|
| Mentioned and cited | The models recommend you and use your pages as sources. This is the position everyone wants. | Defend it. Monitor the trend, keep the content fresh and watch who is climbing behind you. |
| Mentioned, not cited | The brand carries weight in the model's picture of the market, but your own pages are not chosen as sources. | A craftsmanship problem. Make your pages more answerable and technically accessible, so you own the source spot others currently fill. |
| Cited, not mentioned | Your content is good enough to build answers on, but the name lacks the weight to be recommended. | A reputation problem. Work on third party coverage, consistency and a sharp position that gives models a reason to name you. |
| Neither | You are not part of the AI answers in your own category at all. | Start with the technical layer and answerable content, because you control those yourself, and build coverage in parallel. |
The point of the matrix is that "more content" is not the answer to everything. A business that is cited but not mentioned can publish itself into the ground without it helping, because its problem lives in the third party sources. And one that is mentioned but not cited does not need more PR, it needs better pages. Without splitting the two numbers apart, there is no way to know which medicine applies.
The shortlist effect: Why this is urgent
The reason the split matters so much is the arithmetic of the answer. A Google results page has ten blue links, with page after page behind them. An AI answer names two or three brands and links a handful of sources. There is no page two, and no eleventh spot to cling to.
The customer who asks gets a finished shortlist, phrased as a recommendation from an assistant they trust. We know from decades of research on buying behaviour that people rarely look beyond the first list they are given, and an AI answer feels more like advice than advertising. If you are not in the answer, you are effectively not competing for that customer, and you never even find out the competition took place.
That is what makes mentioned and cited the two numbers every marketing team should follow from here on. They do not describe a future channel. They describe how many of today's buying decisions you are quietly excluded from.
How to work on both at once
In practice we recommend attacking the two competitions in parallel, with a clear division of labour:
- Establish the baseline. Find out how often you are mentioned and how often you are cited within your topics, and who holds the spots today. Without that number, everything else is guesswork.
- Take the citations first. They are built on your own pages, so this is where initiatives work fastest: Technical foundation, answerable content, named authors, structured data. This is weeks of work, not months.
- Build the mentions systematically. Make your information consistent everywhere, earn third party coverage and reviews where your customers compare options, and sharpen your position so the coverage has something concrete to say. This is months of work, which is exactly why you start now.
- Measure both, separately. Follow the two numbers over time and see which initiatives move which number. That is how you learn what works in your specific category, instead of guessing.
Read the answer twice
So the next time you see an AI answer in your own industry, read it twice. Once for the names in the text: Who gets recommended, and with what reasoning? Once for the links in the source list: Which pages got to define the answer? Those two lists tell you exactly where your work lies, and our pillar guide on AI visibility shows you the road from there.