Your site is in good shape. The tech is sound, the copy is right, the load time is fine. And still, when someone asks an AI assistant who the best provider in your market is, it names your competitor. Your company does not come up. We hear this question every week from IT companies. The answer is not on your website. It is in four mechanisms that decide which companies an assistant names.
Sound familiar? Everything is in order, yet the visibility does not follow. It helps to understand how an assistant makes its choice. We explain it through four mechanisms, and we give you a checklist to work out for yourself why you are missing.
An AI assistant works differently from a search engine. Google shows a list of results. An assistant gives one answer, summarised from a handful of sources. There is no second page. Whoever is not in that selection does not exist for the person asking.
That selection does not come from a single database. It is a combination of three signals: the authority of your domain, how well your entity is recognised, and whether you appear in the model's training data. A fourth signal sits on top: how often other trusted sources point to you.
It has nothing to do with how good you are. It is about how visible and how recognisable you are online. We regularly see a weaker provider win because its digital footprint is stronger. That is not a judgement on quality, it is an observation from practice.
Assistants have preferred sources. Wikipedia, large publications, official company websites: those domains keep coming back. Not because they are always right, but because they are reliable and stable. An assistant would rather pick a known source than an unknown one.
Your competitor may sit on an industry comparison that gets cited often. Or on a trade magazine page that has been around for years. Your domain is technically better, but it has not built the same authority. In that case the known source wins.
You do not build domain authority with a nice website. It comes from outside: mentions on other sites, links from trusted parties, reviews on platforms that count. An assistant weighs those signals even though your own copy says nothing about them. It is not a popularity contest, it is a trust score you have to earn.
An AI has to understand who you are. That sounds simple, but many IT companies make it hard for the models. Shifting names, no clear core activity, different descriptions on your site, on LinkedIn and in directories. A model then fails to recognise you as one entity.
Among Dutch IT companies that do not get named, we often see the same thing: no consistent name and address across the relevant sources. The competitor that does get named has that recognisability. Its products, audience and specialism are described the same way everywhere.
Entity recognition is about consistency. Use the same name everywhere, the same description, the same core activities. An assistant has to be able to place you. If you do not make that clear, it picks someone who has arranged it. Put plainly: if you do not make clear who you are, someone else does it for you.
Models have a knowledge cutoff. Anything that happens after that date is not automatically visible to them. Your new service, your new customer case, your rebrand. It is on your site, but the model does not know it.
This explains why an outdated mention keeps getting used. The information is old, but it is present in the training data. Your current information is better, but not available to the model. An assistant can only draw on what it knows.
Live web search helps, but not always. Some assistants search the web at the moment of the question. Others rely on their fixed knowledge. And even with live search: only findable, well-structured pages get picked up. A recent change is no guarantee. It is an opportunity.
Assistants like to check whether a source is trustworthy. They look at the context in which your company is mentioned. Are you only mentioned on your own site? That is a weak signal. Are you also mentioned by customers, trade publications or independent parties? Then it gets stronger.
Reviews play a part here. On platforms such as G2, Capterra or Google Business Profile you build up evidence an assistant can use. Consistency is the key word here too. A few good reviews a month, just after a moment of success, builds more trust than collecting a pile of reviews once.
Think about how you pick an adviser yourself. You do not only ask what they say about themselves. You look at what others say. An AI does the same. The difference: an AI does it in milliseconds, based on what is findable online. If that picture is incomplete, it picks someone else.
You do not need to be an expert to see where you are missing. Use this checklist. It costs you an afternoon, and you will know exactly where it hurts.
The outcome is often confronting. Not because your site is bad, but because the visibility outside your own domain is missing. That is exactly what AI assistants look at. Your website is only the start. The work sits in everything around it.
Want to know where you stand? We measure every month how often you and your competitors are named in ChatGPT, Gemini, Claude and Perplexity, including the source underneath. That way you see not only that you are missing, but why. Request a no-obligation scan through our AI visibility scan.
Because the model draws on what it knows, not on what is best. If a company has been consistently present online for years, that information stays available, even when better options exist. The knowledge cutoff means recent improvements by others are not picked up in time.
Yes, and it is an important step. Make sure your content is readable for AI crawlers and that your entity is described consistently. But your website alone is not enough. Mentions outside your domain, such as reviews and press coverage, weigh at least as heavily. An assistant relies on outside confirmation.
Yes. SEO aims at position in search results, GEO aims at appearing in AI answers. With SEO you pull traffic to your site, with GEO you want to be cited in an answer that summarises. The techniques overlap, but the goal and the way you measure differ.
The four mechanisms explain a lot, but they do not solve the problem. That takes work, and not only on your own domain. Whoever starts now on building a consistent, citable presence is the one who gets ahead. Call us if you want to know where you stand. We are happy to take a look.