.png)
You are no longer only writing for candidates. You are writing for the thing that explains you to them.
Here is a sequence available to anyone considering your company, right now, for free, in under four minutes.
What is it actually like to work at [your company]? How does it compare to [your closest competitor] as an employer? Is the company financially stable? What do employees complain about most? Is it a good place for a senior engineer specifically? What should I ask the recruiter that they will not expect?
Six questions, one interface, a synthesized answer to each, complete with comparisons and caveats. No visit to your careers site. No Glassdoor scroll. No conversation with anyone who works for you.
By the time a recruiter reaches this person, a summary of your company has already been produced and read. Somebody has already answered the question of why they should consider you. It just was not you.
First, a caution about the numbers
The recruiting industry has produced a rapid supply of statistics about this, and most of them should be handled carefully.
Greenhouse's 2025 report put AI search somewhere in the job hunt for roughly three in four candidates. Other industry surveys go considerably higher: one widely circulated figure claims 96% have used AI to research employers, but read the base rate and it applies to job seekers who already use AI tools, which is a very different population from job seekers generally. That kind of conditional denominator gets stripped off in the retelling, and then a real finding becomes a misleading one.
These are vendor and industry surveys, often with self-selected samples and an interest in the answer. Treat the specific percentages as directional at best.
The argument does not need them. The durable evidence is infrastructural and comes from outside recruiting entirely. Google's AI Overviews went from about 1.5 billion monthly users in early 2025 to more than 2.5 billion by 2026, across 200-plus countries. OpenAI reported ChatGPT at 900 million weekly active users in February 2026, up from 400 million a year earlier, and it crossed a billion monthly app users in June.
Whatever proportion of your candidates does this today, three things are not in dispute: the behavior is trivially easy, the tools are already in everyone's hands for other purposes, and the direction of travel is one way. Planning around the current percentage is a mistake regardless of what the percentage is.
You used to own more of the path
The old candidate journey had nodes you could influence and, in places, control.
Someone searched. They landed on your careers site, which you owned outright. They checked Glassdoor, which you did not own but could respond on. They talked to a recruiter, who was your employee, delivering your message. They read a news article or two. Each surface was distinct, each was visited deliberately, and the candidate assembled their own impression from the pieces.
That assembly job has now been partially outsourced. A layer sits on top of all those sources, reads them faster than any candidate would, resolves the contradictions between them according to its own logic, and returns something that reads like a verdict.
You still own your careers site. You just no longer own the summary, and the summary is what gets read.
The shift is from discovery to interpretation
This is the part most employer brand teams have not absorbed yet, and it is not a variation on search engine optimization.
SEO was a contest to be found. You competed for position on a results page, and winning meant the candidate arrived at your content and read it in your words, in your order, with your framing.
This is a contest to be explained accurately. The candidate is not asking where information about you lives. They are asking what it all means and whether you are worth their time. The answer is generated by something that has read your careers page, your Glassdoor reviews, your news coverage, your Reddit mentions, and your competitors' equivalents, and has formed a compressed view.
You are not competing for a click anymore. You are competing to be summarized in a way you would recognize. The nature of the work changes accordingly, and so does what counts as good content.
Your employer brand now has two audiences
The human audience has not changed. People still respond to narrative, emotion, specificity, and context. They want to picture themselves in the work, and they decide with feeling as much as logic.
The machine audience wants something else: claims that are explicit rather than implied, evidence that is attributable, information that is structured, and positioning that stays consistent across every source it encounters.
The mistake would be concluding that you should write for the machine, producing flat, hedge-free, keyword-shaped prose that no human enjoys reading. That solves nothing, because the human still has to be persuaded at the end.
The good news is that the two requirements overlap far more than they conflict. Specificity serves both. Evidence serves both. Consistency serves both. What fails with both audiences is exactly what employer branding has produced most of: fluent, warm, unfalsifiable copy that means nothing in particular.
Generic branding degrades fastest here
Imagine a model asked to compare ten companies in your sector as employers, and every one of them describes itself as innovative, collaborative, purpose-driven, and people-first.
It cannot differentiate on that material, because there is nothing there to differentiate on. So it does what any reasonable summarizer does: it reaches for whatever else it can find that is specific. Glassdoor complaints are specific. Layoff coverage is specific. Salary data is specific. A Reddit thread about a bad interview process is extremely specific.
This is the part worth sitting with. Saying nothing distinctive is not a neutral position. It is a decision to let third-party sources supply the substance of your summary, because they are the only ones providing anything concrete enough to use.
The companies that come out of this well will not be the loudest or the most tastefully designed. They will be the most legible: specific claims, real proof, named tradeoffs, published where both people and machines can find them. Which is the same standard that made a company choosable before any of this existed, now enforced by a new intermediary.
Subscribe for more ideas and insights
Give it something it can confidently repeat
If you want to be described accurately, provide material that is safe to restate. In practice that means a few specific things.
Named programs rather than described intentions. "A mentorship program" is a claim. "The Bridge program, which pairs every new engineer with a senior engineer for their first six months" is a fact with a name attached.
Numbers, stated plainly and in text. Percentages of internal promotions, headcount by function, tenure figures, how many people moved between teams last year.
Policies written explicitly rather than gestured at. If the working model is three days in office, say three days in office. Vague flexibility language reads as evasion to a model, and to a person.
Role-specific information, because "what is it like to work here" is nearly always asked as "what is it like to work here as a nurse, an engineer, a field technician." Company-level content cannot answer a role-level question, and generic answers to specific questions are unsatisfying to everyone.
Employee evidence with attribution: named people, real quotes, specifics that could not have been invented by a copywriter.
And consistency across sources, which is more important than most teams realize. When your careers site, your LinkedIn page, your job postings, and your review profiles tell meaningfully different stories, that contradiction is visible to something reading all of them simultaneously. Contradiction reduces confidence, and lower confidence produces hedged, unhelpful summaries.
The technical layer, structured data and clean markup, matters and is worth doing. But it is a multiplier on substance, not a substitute for it. Perfectly structured nothing is still nothing.
Your career site is evidence, not a destination
The mental model to retire is the career site as a place people go.
Increasingly it is a source that gets read on the candidate's behalf, by a tool, and then compressed. A person may make a decision materially influenced by your careers content while never seeing your careers page, your photography, or the video you spent a quarter producing.
That does not make the site less important. It changes what makes it good. Beautiful and vague loses to plain and specific, because only one of those survives compression.
The test that takes ten minutes
Stop theorizing and go run it.
Open the tool your candidates are most likely to use. Ask the six questions from the top of this article about your own company. Then ask them about your two closest competitors. Read all of it carefully.
You will learn several things quickly. Whether the answer about you is accurate. Whether it is flattering. Which sources it appears to be leaning on, which is often the most alarming part. Whether a reasonable person reading it would put you on their list. And whether anything in it is drawn from material you actually produced.
Then do the version that matters most: ask it directly why someone should choose your company over your main competitor, and see whether it can construct an answer at all.
If it cannot, that is not a technology problem. It is the same problem you had before, now being read back to you by a machine that has no incentive to be polite about it.
The old question was whether your careers site sounds good. The new one is sharper: if something had to explain, in three sentences, why the right person should choose you, have you given it enough to work with?
