43 percent of all Google searches now get an AI answer at the top of the page. A year ago that was 15 percent.
That figure comes from Similarweb’s report “The 2026 Generative AI Landscape”, published on 27 July. Almost a tripling in twelve months. In that same period, visits to Google’s AI Mode grew from 126 million to 279 million per month.
The blue link is drying up
Check this: if nearly half of search queries first return a summarised answer, the place where the candidate looks shifts.
In the past, a jobseeker would scroll through ten vacancies on your site. Now, she reads one paragraph that a language model wrote for her, about you, without you in it.
She asks what it's like to work with you. She asks which employers in her sector take their people seriously. She asks where she can really progress. The model answers. And that answer gets its words from somewhere.
The search query has become a sentence
The questions have also become longer. Similarweb sees the average search query growing: from individual keywords to entire sentences. Kim no longer types “accountant vacancy Utrecht”. Instead, she asks which accountancy firm in Utrecht lets you work part-time and still allows you to progress. A keyword page doesn't have an answer to that. A clearly written story does.
The model does forward. Just not to everyone.
There is an angle to this story that employer branding people often overlook. Following a search update on 7 May, referrals to external sites more than doubled: from 25 per cent of searches in March to nearly 60 per cent by the end of May. Citations in AI answers increased more than fivefold in a year.
So the model does forward after all. Just not to everyone. It forwards to the source it was able to cite.
Ranking and being citable are not the same thing
There lies the work. Your employer branding has been built for years to rank: the right words on the careers page, a slick video, a culture block with smiling colleagues. That system rewarded the site that was at the top of Google. The new system rewards the source that a language model dares to cite.
Being citable and being discoverable are two different things. A model does not quote marketing language. It quotes an answer that is correct, that is concrete enough to adopt, and that says the same thing in multiple places.
So it's about your employee benefits in plain English. Your career paths with real examples. Your culture described the way an employee would retell it, the way things actually go on the work floor.
Three names or nothing
A model does not name everyone. Ask it for the best employers in an industry and you get a list of three, maybe five. The rest do not exist for that candidate. Being visible in Google used to mean being somewhere on page one. Being visible in a language model means being mentioned in that short list, or nowhere.
The click only comes after the judgement has already been passed
The first thing that breaks is the click through to your application page. For years, that was the entire funnel: being findable, securing the click, guiding the candidate through the form. If the answer has already been given before the click happens, the application begins with what the model said about you. Your own words only come afterwards, once the candidate is already in.
And this is where it gets uncomfortable. If you don't provide those answers, someone else will. Glassdoor. Reddit. A former employee with an opinion. The model takes what's available.
What is missing is filled in. What is unclear is simplified. And what is nowhere to be found does not exist in the conversation the candidate has with her AI.
Check it in one afternoon
For the recruiter, the work is changing as early as this week. Your real shop window is now the answer that a model gives when someone types in your company. You can check that answer today: which sources it cites, where it gets its image from, and which part of your story is missing from it.
Work it out based on your own figures. Every open role starts with orientation, and that orientation increasingly runs through a model. The scarcer your field, the sooner a candidate drops out over an answer that is incorrect or missing.
This is happening now. 43 percent is today. There are 279 million AI Mode visits this month. The candidate looking up your company tonight will most likely do so via an answer you didn't write and have never seen.
The infrastructure is already there, the story isn't yet
The silver lining is in the numbers themselves. The model refers back to sources more often than a year ago, and it cites those sources verbatim. The infrastructure to be found therefore already exists. Consistent information across your channels. Content that a model can adopt without having to guess. A narrative that sounds the same on your own site as it does in the places where others talk about you.
Any employer can test this today. Ask ChatGPT or Google what it is like to work at your company, and read what comes back. Either your story is there, or the model has filled in the gaps for you.
My name is Bryan Peereboom, founder of AI Rebels. We make employers visible in language models.
Sources: Similarweb, The 2026 Generative AI Landscape (27 July 2026) via Tekedia; TechCrunch, “Google’s AI search is rapidly becoming the default, new data shows” (27 July 2026).