Candidates increasingly start with ChatGPT, Claude, Gemini or Perplexity instead of a search box. Those tools do not return 10 blue links. They return 1 synthesised answer naming 3 to 5 employers, and the brands not named do not exist in that moment. Across the 100 UK employer brands we score monthly for The SetpointHQ Index, the median citation score is 27.5 out of 100. The answer seats are sitting mostly empty.

AEO is not GEO’s twin, it’s the next stage

AEO and GEO sit on the same ladder, one rung apart. GEO gets your content into the material a generative engine can find, parse and trust: structured data, crawl access, a clean entity. AEO is what happens next, whether the engine, having found you, actually names you in the final answer rather than a competitor.

Think of it as a funnel. A careers site a crawler can’t reach never enters the funnel at all. A careers site the crawler reaches but can’t parse enters and stalls. A careers site that’s readable and trusted is in the running, GEO’s job is done, and AEO is what decides whether you’re the answer that gets read out or one of the sources quietly sitting behind it. SEO still matters underneath both of these, because search results are themselves part of what these engines read. The full 3-way comparison, side by side, is on our SEO vs GEO vs AEO page.

Why almost nobody is the answer today

Across the 100 UK employer brands on The SetpointHQ Index, scored monthly against rubric 2026.06.5, the median citation score is 27.5 out of 100. The median knowledge graph score, the measure of whether a machine simply knows who you are, is 89.4. Nearly every brand in the cohort is recognised. Roughly 1 in 4 is actually recommended when a candidate asks where to work.

That gap is the whole AEO opportunity, and it’s wider than most marketing teams assume, because being known and being recommended get treated as the same achievement internally, when the systems doing the recommending measure them completely separately. A model can resolve exactly who you are, read your entire Wikipedia entry, and still leave you out of an answer, because knowing who you are isn’t the same as having a reason to name you over the 3 or 4 other employers it could name instead.

Marketing instinct fights this. The version of your careers page written to sound confident, established, trusted, is often the version that scores worst on citation, because confidence is asserted rather than evidenced. “A trusted partner to employers across the UK” is a sentence any agency could paste onto its own site unchanged. It gives an assistant nothing to check and nothing to prefer you over a competitor for. The fix isn’t less confidence, it’s confidence backed by a number the assistant can verify, or at least treat as a specific, falsifiable claim rather than a mood.

Two different buyers are asking, and the content has to answer both

An assistant fields 2 distinct questions about a recruitment brand, and most sites only write for 1 of them. A candidate asks “is this a good place to work” or “is this agency good to work with if I’m looking for a job”. A client or hiring manager asks “which recruiter should I use for this role” or “is this employer worth applying to” from the other side of the desk entirely. The evidence that answers one rarely answers the other.

Candidate-facing citation material is about the experience of being placed or hired: how fast, how honestly, what happens if it doesn’t work out. Client-facing citation material is about outcomes: fill rates, time to hire, retention past probation, sector depth. A site that only has the client-facing version, common on agency sites built to win business, gives an assistant nothing to cite when a candidate asks the candidate-side question, and vice versa.

Both columns need their own specific, checkable claims, written for the audience actually asking. Most recruitment sites have built one column properly and left the other as adjectives.

6 AEO moves that work in recruitment

1. Publish pages shaped like the candidate’s question. “Is [your company] a good place to work as a nurse” beats “Our culture” every time, because it matches the actual prompt structure a candidate types, and a model retrieves what resembles the question it was asked.
2. Lead with the answer. Assistants lift opening paragraphs that resolve the question directly. A page that spends 3 paragraphs building up to the point buries the exact sentence a model would otherwise quote, and it often stops reading before it gets there.
3. Give every claim a number. In our probes the assistants consistently quote sources with figures: salaries, ratios, timelines, headcounts. A number is also easy to fact-check against other sources, which is precisely why models trust it more than an adjective.
4. Hold a comparison honestly. Answers to “where should I work” are comparative by nature. Content comparing you to the market, fairly, including where you lose, is exactly the raw material an answer needs, and the honesty itself reads as more trustworthy evidence than a page that only praises itself.
5. Keep your entity clean. Consistent naming across your site, LinkedIn and Wikidata is what lets an assistant attribute the evidence to you with confidence. A trading name that differs from your registered name, or a Wikidata entry pointing at an old domain, splits the evidence trail a model needs to connect back to you.
6. Get named somewhere you don’t control. Assistants weight independent evidence higher than a brand’s own claims about itself. Research on what AI summaries actually cite has found that Wikipedia, YouTube and Reddit alone supply roughly 15% of the links referenced. A press mention, a genuine review pattern, a forum thread where real employees weigh in, all carry more weight with an assistant than the identical claim on your own careers page, because you didn’t write it.

In an answer engine there is no page 2. There isn’t really a page 1. There is the answer, and there is everyone else.

What actually gets cited: a worked example

Take the question “is [agency] good to work with if I’m in tech”. Two answers to that question could sit on the same site.

The first: “We pride ourselves on deep sector knowledge and a consultative approach that puts candidates first.” Every recruiter’s About page says a version of this. An assistant has no way to differentiate it from the next agency’s identical claim, and no fact inside it to quote, so it gets summarised into nothing or dropped entirely.

The second: “Average time from first call to offer across our tech desk is 18 days. 71% of placements in the last 12 months came from candidates we’d already spoken to, not cold outreach.” 2 numbers, both checkable, both specific to this agency and nobody else’s. An assistant assembling an answer about which tech recruiters actually move fast has something to lift directly, attribute and quote.

The gap between those two paragraphs is most of what separates a cited brand from an invisible one. It isn’t more content. It’s the same claim, rewritten with something in it a machine can use.

The right-hand column is not better writing. It is the same claim with a number attached, and that is the entire difference between a sentence a model discards and a sentence it can attribute to you by name. The figures above are illustrative; yours have to be real, because the checkability is the point.

How we measure it

The SetpointHQ Index puts real candidate questions to all 4 major assistants every month and records who gets named, at what position, for 100 UK employer brands. Citation carries the heaviest weight of our 5 dimensions because it’s the closest thing this field has to a scoreboard. Sector by sector, the pattern holds: knowledge graph scores in the 90s, citation medians in the 20s. The machines know everyone. They recommend almost no one.

Buyers are already asking for this by name

This isn’t hypothetical demand. Recruitment buyers are already typing versions of “how do I get my agency listed as a reliable source in ChatGPT” directly into AI assistants and search, by name, asking who does this work. Some of those are full-sentence prompts naming the specific problem, which is a different kind of signal to a keyword search: nobody types a paragraph into a search box unless they’ve already decided they need to solve this and are looking for someone who does it, not just information about it.

That’s a buyer with a live problem, not a browsing prospect. If that’s specifically your question, the mechanics of winning that citation, not just being eligible for it, are covered in how to get cited by AI assistants.

Frequently asked questions

Why don’t recruitment agencies appear in ChatGPT?

Usually because there’s nothing specific enough to cite. An assistant needs a checkable fact to attribute to you: a placement time, a fill rate, a sector specialism backed by numbers. Generic positioning language, the kind every agency’s About page carries, gives it nothing to quote, so it defaults to naming whichever competitor does have something concrete on the page.

What is answer engine optimisation (AEO)?

AEO is the practice of being named in the final answer an AI assistant gives, rather than merely being part of the source material it drew on. GEO gets you into that material. AEO decides whether you’re the brand actually recommended once the answer gets written.

How does ChatGPT decide which employers or agencies to recommend?

It weighs verifiable evidence: structured data, knowledge graph presence, third-party citations and specific, checkable claims on your own site. It doesn’t take a brand’s self-description at face value the way a person skimming a homepage might. A claim with a number attached and independent corroboration outranks a well-written but generic one every time.

How do I get my recruitment agency listed as a reliable source in ChatGPT?

Start with the basics GEO covers: structured data, crawl access, a clean entity. Then write content shaped like the question a candidate or client would actually ask, with a specific, checkable claim in the first sentence. Independent mentions, press, reviews, forum threads, matter as much as anything on your own site, because an assistant trusts evidence it didn’t get from you more than evidence it did.

Which AI assistants matter most for recruitment marketing?

ChatGPT, Claude, Gemini and Perplexity account for most assistant-routed candidate and buyer research today, alongside Google’s AI Overviews sitting on top of traditional search. The SetpointHQ Index tracks all 4 assistants monthly across 100 UK employer brands for exactly this reason. A single reading on one assistant tells you very little.

Source: The SetpointHQ Index, 100 UK employer brands, scored monthly across ChatGPT, Claude, Gemini and Perplexity. Figures as at the 1 September 2026 run, rubric 2026.06.5.