How AI answers are reshaping employer-brand discovery, and what GEO looks like for hiring teams

A candidate who wants to know whether a company is a good place to work used to type the name into Google, open the careers page, skim a few reviews and make up their mind. More and more of that research now happens inside an AI assistant. They ask ChatGPT, Claude, Perplexity or Gemini something like "which fintechs in London are good employers for software engineers?" and read the answer.

That answer is short. It names a handful of companies, describes each in a sentence or two, and sometimes cites a source. If your company isn't named, or is described inaccurately, the candidate may never reach your careers site at all. This is where generative engine optimisation (GEO) stops being a marketing buzzword and becomes a hiring problem.

This article looks at what assistants draw on when they describe an employer, what we've seen from measuring 100 UK employer brands, and a practical checklist talent and employer-brand teams can work through.

What assistants draw on when they describe an employer

Each provider publishes some documentation about how its systems find web content. None of it explains exactly how an answer is put together, but it does tell you what has to be true for your pages to be in the running.

Google AI Overviews and AI Mode use Google's normal index. Google says that to appear as a supporting link, a page "must be indexed and eligible to be shown in Google Search with a snippet," and that "there are no additional requirements" beyond standard SEO (Google Search Central: AI features and your website). For employers, that means the basics of careers-site SEO still decide whether Google's AI features can use your pages.

ChatGPT search relies on OpenAI's OAI-SearchBot crawler. OpenAI's documentation says sites can allow OAI-SearchBot "in order to appear in search results" independently of GPTBot, which relates to training (OpenAI: Overview of OpenAI crawlers).

Perplexity uses PerplexityBot, which its documentation describes as "designed to surface and link websites in search results on Perplexity," and recommends allowing if you want to appear (Perplexity crawlers).

Claude uses Claude-SearchBot for search and Claude-User to fetch pages when someone asks, and Anthropic notes that blocking these can reduce a site's visibility in user-directed search (Anthropic: crawler documentation).

Structured data helps machines understand a page. Google describes it as "a standardized format for providing information about a page and classifying the page content" (Google: Introduction to structured data). For hiring, the two types that matter most are JobPosting on job pages and Organization on the homepage, including sameAs links to your official profiles.

Beyond what's documented, our own probing suggests two more inputs matter: knowledge-graph entities such as a Wikipedia article and a Wikidata entry, and third-party employer signals such as review sites. That's an observation from our Index work, not something the providers have confirmed.

What our Index shows

Disclosure first: the SetpointHQ Index is run by my company, SetpointHQ. Each month it scores 100 UK employer brands across 14 sectors against a published rubric. For each brand we send a standard set of candidate-research questions to four assistants (Anthropic Claude, OpenAI GPT, Perplexity Sonar and Google Gemini) and record whether the employer is named and how prominently, alongside structural checks of their sites and external profiles. The full methodology and weightings are public in the SetpointHQ Index, and the rubric version for the 1 October 2026 cohort is 2026.06.5.

Each brand is scored from 0 to 100 on five dimensions:

1. Citation likelihood (35% of the composite). Do the four assistants name the employer when asked candidate-style questions, and how high in the answer?
2. Schema completeness (20%). Does the careers site expose valid JSON-LD for types such as Organization, JobPosting, BreadcrumbList and WebSite?
3. Knowledge-graph presence (15%). Does the company have a Wikipedia entry, a Wikidata item and complete sameAs cross-references?
4. LLM crawlability (15%). Does robots.txt allow AI crawlers such as GPTBot, ClaudeBot, PerplexityBot and Google-Extended, is there an llms.txt file, and is careers content in raw HTML rather than rendered only by JavaScript?
5. Social-signal density (15%). Is the employer present where assistants pick up employer signal? In the current version, this dimension covers Glassdoor only (review count, recency and average rating). Other platforms are planned but not yet scored.

In the 1 October 2026 cohort, the median composite score was 55.2. The dimension scores tell a clearer story:

• Knowledge graph: 89.3, with 8% of brands scoring zero
• Crawlability: 75.0, with no brands scoring zero
• Social (Glassdoor): 74.0, with no brands scoring zero
• Schema: 35.0, with 19% of brands scoring zero
• Citation: 26.3, with no brands scoring zero

So most large UK employers are known entities and can be crawled. Where they fall down is in the two places they control most directly: the structured data on their own careers sites, and actually being named when a candidate asks an assistant. A brand with a Wikipedia page and an open robots.txt can still be left out of the answer.

Sector differences are large too. Professional services brands scored 63.8 as a sector and media and telecoms 43.9. My reading, and it is a judgement rather than a finding, is that sectors with strong graduate-recruitment programmes have spent years producing the kind of specific, well-structured careers content that assistants find easy to use.

A practical GEO checklist for talent teams

These steps come from the documentation above and from what we've seen in the Index. Where something is my own judgement, I've said so.

1. Check that AI crawlers can reach your careers site. Review robots.txt against each provider's published crawler list. Many careers sites sit on a separate applicant-tracking platform with its own robots.txt, so check that too. Make sure job and culture content is present in the HTML, not only rendered by JavaScript.

2. Add JobPosting and Organization schema. Follow Google's documentation for JobPosting and Organization. Include salary, location and employment type where you can. Validate with the Rich Results Test, and re-test after any careers-site or ATS change.

3. Keep your entity data consistent. Use the same legal name, description, headquarters and official profile links everywhere: your site, your sameAs list, LinkedIn, Wikidata and directories. In my judgement, inconsistency is one of the most common reasons an assistant describes an employer vaguely or confuses it with a similarly named company.

4. Publish careers content worth citing. Assistants have to say something specific about you. Give them specific, checkable facts: where you hire, what roles you hire for, how training and progression work, and what your benefits actually are. Generic employer-value-proposition copy gives them little to quote. This is my judgement, informed by which brands score well on citation in the Index.

5. Earn mentions on sources assistants already use. When Perplexity and ChatGPT cite sources, look at which ones they choose for your sector's questions. Those publications, directories and review platforms are where coverage is most likely to feed back into answers. This is an observation from our probing, not a documented rule.

6. Look after your Glassdoor presence. It's the one external review platform our Social dimension currently measures. In my judgement, recent, honest reviews with visible employer responses give assistants a fairer picture than a stale profile.

How to measure progress month to month

GEO for employer brands needs its own measurement, separate from careers-site traffic. A simple approach any team can run:

1. Write 20 to 30 questions a candidate in your key talent pools would actually ask, such as "best graduate schemes in engineering in the UK" or "good employers for nurses in Manchester."
2. Ask each question in ChatGPT, Claude, Perplexity and Gemini on the same day each month.
3. Record whether you're named, where you appear in the answer, how you're described and which sources are cited.
4. Note anything inaccurate, such as old office locations, discontinued schemes or wrong benefits, and trace it back to the page or profile it came from.
5. Repeat after each fix, remembering that answers vary between runs, so look at trends rather than single results.

The candidate's first impression of your company is increasingly a paragraph written by an AI assistant. You can't write that paragraph yourself, but you can make sure the material it's built from is accurate, structured and easy to find.