Google AI Overviews and employer brands: the citation gap

A nurse in Leeds types “is [your company] a good place to work” into Google. She doesn’t scroll. The answer is already sitting at the top of the page, written before the first blue link.

Three sources built that answer. A Glassdoor page, a Reddit thread from two years ago, and a local paper covering a restructure. Your careers site wasn’t one of them.

Nobody outbid you. No job board fee changed hands. Google looked at your own website, found nothing worth quoting, and used someone else’s version of you instead.

That’s a different failure from ranking badly. And it needs a different fix.

What now sits above your careers site

Google’s AI Overview writes an answer at the top of the results page and cites a handful of sources underneath it. AI Mode goes further and runs the whole session as a conversation. Both do something worth understanding if you hire at volume: they take one question and quietly turn it into several searches behind the scenes. Google calls this query fan-out, and describes the point of it as showing “a wider and more diverse set of helpful links” than a standard results page.

So the candidate asks one thing. Google runs a dozen searches you never see as a phrase in Search Console. Pay. Shifts. Culture. Turnover. Whether the interview is any good. Your careers site is competing for all of them at once, and mostly losing to sites that answer them in plain text.

Google says there’s nothing special to do. It’s right, and that’s exactly the problem

Google’s published guidance on AI features is blunt about this:

Plenty of vendors will sell you a lever that doesn’t exist. There isn’t one. But read the guidance properly and two conditions are sitting inside it. The page has to be indexed. And it has to be eligible to show a snippet.

Careers sites fail one or both of those more often than any other page type we look at. Google names the controls that switch snippet eligibility off: nosnippet, data-nosnippet, max-snippet and noindex. A max-snippet value inherited from a template nobody’s opened since 2023 will remove an entire careers section from AI features, silently, while every page in it still ranks perfectly well.

That’s the shape of this whole problem. Nothing is broken. Nothing throws an error. You just aren’t quotable.

Known, readable, and never quoted

The SetpointHQ Index scores 100 UK employer brands every month on how AI systems describe them to candidates. Here’s the August 2026 snapshot, rubric 2026.06.5, across the five dimensions we measure.

Read the top of that table against the bottom.

Google already knows who these companies are. Knowledge graph median 88.5. It can reach and read their websites without difficulty. Crawlability median 75.0, and not a single brand in the cohort scores zero. Then ask a question a real candidate would ask, and the median brand gets named 26.3% of the time.

Composite median across all 100 brands: 53.3 out of 100.

The middle row is the one to sit with. 27% of UK employers publish no schema an AI system can read, and the median score is 30 out of 100. These are household names. They’re findable, crawlable, and structurally mute.

The sector split says something uncomfortable too. Professional services sits at 65.5 and legal at 60.2. Logistics comes in at 42.7 and healthtech at 37.5. The sectors with the least trouble hiring are the most visible. The ones running permanent vacancy gaps are the quietest. You can see the full cohort in The Index.

Why employer brands fail this in particular

This isn’t generic SEO underperformance. There are four failure modes we see again and again, and all four are specific to how recruitment sites get built.

The careers site isn’t yours. It’s a Workday, Greenhouse or Phenom subdomain. Job content renders client-side, lives on a vendor’s domain, and frequently carries noindex. Nothing to index means nothing to cite, and every scrap of authority you build points at your ATS provider.

The pages candidates ask about don’t exist. Pay bands. Shift patterns. What happens in the second interview. Whether anyone gets promoted after two years. Google will answer those questions today, from Glassdoor and Reddit, whether or not you’ve written anything.

JobPosting markup is missing or rotting. Google’s required properties are datePosted, description, hiringOrganization, title, and either jobLocation or jobLocationType. Filled roles left live with no validThrough date aren’t just untidy, they’re a policy violation that can attract a manual action.

The employer brand story is trapped in a PDF or a video. Beautiful work. Signed off by a committee. Completely unquotable.

What we’d actually do about it

1. 1Confirm the pages can show a snippet.Search your careers templates for nosnippet, data-nosnippet and max-snippet. This takes ten minutes and is the single cheapest thing on the list. One inherited meta tag can exclude an entire section.
2. 2Get job content onto a domain you own.If the ATS renders roles in JavaScript, publish a server-rendered index of live vacancies on your own site and link through to apply. The application flow can stay where it is.
3. 3Write the answer pages.One page per real candidate question. Answer it in the first 40 words, in text, above any accordion or tab. Accordions hide content from extraction more often than people expect. Doing this properly is how we’ve taken an in-house careers site to 107% application growth in twelve months.
4. 4Fix JobPosting and keep it fixed.All required properties present, validThrough populated, filled roles returning 410 rather than sitting there quietly decaying.
5. 5Put actual numbers in the copy.AI systems quote specifics and skip adjectives. “Competitive salary” gets nothing. “£13.20 an hour, £14.85 for nights, reviewed each April” gets named, because it’s the only sentence on the page that answers the question.
6. 6Claim your entity.Consistent Organisation schema, a Wikidata presence, and one legal name used identically across the site and the ATS. 8% of the Index cohort still score zero here, which means Google isn’t certain they exist as a single thing.
7. 7Measure the answer, not the position.Ranking first for your own company name tells you nothing. Every brand in the Index ranks first for its own name. They rank first and get quoted 26.3% of the time.

Items 1 to 4 are engineering. Items 5 and 6 are content and governance. Item 7 is the one most teams skip, and it’s the reason the other six keep getting deprioritised: nobody is reporting on it, so nobody is accountable for it.

This sits on top of your SEO, not instead of it

Standard search fundamentals still carry the whole thing. If the page can’t be crawled and indexed it can’t be quoted, and no amount of clever markup rescues it. What’s changed is that ranking is no longer the finish line. It’s the entry requirement.

We’ve written about the wider shift in SEO vs GEO for recruitment, the mechanics in generative engine optimisation for recruitment, and how to win the answer itself in answer engine optimisation for recruitment. When Google published its own guidance earlier this year, we covered what it left out. If you want the five minute version for your own brand, start with the self check.

Frequently asked questions