For every visitor that ChatGPT sends your way, OpenAI’s crawler has read your pages 217 times.
That figure comes from Cloudflare, measured over 28 days up to 21 July 2026. GPTBot reads 217 pages for every click it returns. Perplexity sits at 225 to 1. Anthropic's ClaudeBot at 2,237 to 1. Google with its AI Overviews at 4.6 to 1.
That relationship is called the crawl-to-refer ratio, and for employers it tells the whole story. The language models read everything you publish about yourself. Your careers site, your vacancies, your culture page, your press releases. They read it hundreds of times. And they send almost nobody back.
Your candidate is no longer coming over to take a look for themselves. They ask the machine and the machine answers with what it has read.
The conversation does not take place on your dashboard
If you look at your visitor figures, everything seems fine. Ahrefs analysed 76,000 sites: ChatGPT accounts for 0.21% of web traffic, whilst Google accounts for nearly 40%. On your dashboard, AI is a rounding error, so it gets pushed into the “to do later” pile.
But the conversation doesn't happen on your dashboard. It happens in the reply.
Imagine the question thousands of jobseekers are typing right now. “Which employers in my field have a good culture and room for career progression?” The model names three, maybe five. Are you among them, or do you not exist in that answer?
That is a different question than whether you rank well on Google. In the past, a candidate would read your Glassdoor reviews, your LinkedIn page and your careers site, and draw their own conclusion. Now the model draws the conclusion, and the candidate only reads that.
What a model says about you is up to you
Here lies the core. A language model builds its response from what it finds out about you. What is missing, it fills in. What is unclear, it simplifies. And what is nowhere to be found, does not exist.
It works in two steps. First, the model determines whether you count as a source at all: are you findable, recognisable and consistent enough to be mentioned. Only after that does it decide whether it actually uses your words in the answer. You can be read and still not turn up anywhere. That is not bad luck; that is a gap in your narrative that you can plug.
If your employment conditions tell three different stories in three places, the model chooses one, or it drops you. If your culture is only in an atmospheric video and nowhere in readable text, the crawler reads nothing. If the most recent thing written about you is a reorganisation from two years ago, then that is your story, whether you recognise it or not.
The click shrinks. The answer grows.
For fifteen years, employer branding was all about the click. You built a nice careers site, put some budget into ads, and waited for the candidate to turn up. That model still works, but it is shrinking by the day. Google itself increasingly pushes an AI summary above the blue links; according to research from this year, nearly half of all search queries now have one.
The click is becoming rarer. The answer is becoming the place where the candidate chooses, compares and rejects.
217 chances to say the right thing
I understand that this feels uncomfortable for a Head of Employer Branding. You've just launched your new careers site, the photos are right, the EVP is in place. And then it turns out that the most important reader of that site isn't a human, but a crawler that reduces your text to a single paragraph that you have no control over.
Yet that is precisely the point where you can do something about it. A model that you read 217 times also gives you 217 chances to say the right thing.
What seems like a peripheral phenomenon today will be standard practice in six months. What is still optional today will be the baseline in a year. And what is still invisible in those answers today will soon be undiscoverable to the very people you need.
What you can do now
What you can do now is less complicated than it sounds. Make sure your story says the same thing everywhere, so the model doesn't have to guess. Put your culture, your terms and conditions, and your career progression opportunities into text that a machine can read, not just visuals. And check what the models are actually saying about you today, because that is the only honest measurement of where you stand.
That last part takes an afternoon. Most employers don't do it and that is precisely why the few that do are now building an advantage you won't catch up with later.
Candidates no longer ask Google. They ask a model that has read your pages hundreds of times and turns them into a single sentence. The only question that matters is whether that sentence is correct and whether you are in it.
My name is Bryan Peereboom, founder of AI Rebels. We make employers visible in language models.
Sources: Crawl-to-refer ratios (Cloudflare Radar, July 2026) via SEOmator; ChatGPT traffic data (Ahrefs / ALM Corp); Google AI Overviews statistics 2026 (The Stacc).