Candidates ask assistants which employers to consider, what a role pays, and whether a company is worth their time, before they ever land on your careers site. The assistant’s answer is assembled from whatever it can find, parse and trust about you. GEO determines whether your brand is in that working set at all. For the full breakdown of how GEO differs from SEO and its sibling discipline AEO, side by side, that comparison lives on its own page. This one is the operational guide: what GEO actually optimises for, where recruitment brands lose the most points, and what to fix first.

What GEO actually optimises for

Generative engine optimisation is not a rebrand of SEO for a new algorithm. It optimises for a different event entirely.

SEO gets a page in front of a person, who then decides whether to click. GEO gets a fact in front of a model, which decides whether to reuse it. Nobody clicks through to your careers site from inside a ChatGPT answer. The model has already extracted what it needs before the candidate sees a link, if it shows one at all.

That changes what “good content” means. A careers page written to persuade a human reader, long paragraphs, brand voice, a hero video, gives a language model almost nothing to extract. A careers page that states facts plainly, the role, the pay band, the location, the team size, the interview stages, gives it plenty. Recruitment content has spent twenty years optimising for persuasion. GEO asks it to optimise for extraction too, without losing the persuasion that still has to land once a candidate actually arrives.

3 things determine whether your content makes it into that working set: can a crawler reach it, can a machine parse what it says, and does the machine trust the entity saying it. Miss any one and the rest doesn’t matter. A beautifully marked-up page a crawler can’t reach is invisible. A reachable page with no structured data is unreadable. A readable page attached to an entity the machine can’t verify gets discounted. Fix them in that order, each one is a precondition for the next.

Put two sentences side by side and the difference is obvious. “We believe in empowering our people to do their best work” gives a model a feeling with nothing underneath it to extract or check. “Average time from application to offer is 12 days, and 68% of our current managers were promoted internally” gives it 2 facts it can quote, compare and attribute to you specifically. Most recruitment marketing was written by people whose job was to make a human feel something. GEO needs a second pass that hands the machine something to work with as well, without deleting the first version, both jobs still need doing.

The 7-point GEO checklist for recruitment

1. Organization markup on your corporate root. The machine’s anchor for who you are. Run the Rich Results Test against your homepage; if Organization doesn’t appear, nothing downstream resolves to your brand.
2. JobPosting markup on every live role. Without it your vacancies are pictures of jobs, not data about jobs. This is the schema type recruitment sites skip most, because it lives on the page a developer builds once and forgets, not the marketing site everyone keeps an eye on.
3. Crawler access for GPTBot, ClaudeBot, PerplexityBot and Google-Extended. Blocked crawlers mean an empty working set, however good the content behind them. Check robots.txt for these 4 by name, not by assumption; a blanket rule written for something else years ago can catch all 4 without anyone noticing, and there’s no error message when it does.
4. Server-rendered careers content. Roles that only exist after JavaScript runs are invisible to most retrieval. Load your careers page with JavaScript disabled and see what’s left; that’s closer to what a non-rendering crawler sees than the polished version in your own browser.
5. A correct Wikidata entity with your official site attached, so evidence attributes to you and not a namesake or a subsidiary. Search your own company name on Wikidata before assuming this is done. A surprising number of UK employers don’t have one, or have one pointing at the wrong URL.
6. Concrete facts on career pages. Pay, progression timelines, headcounts, locations. Engines reuse specifics and skip generalities, the same way a journalist would.
7. Measurement on a cadence. Models and indexes shift constantly, so a single reading tells you almost nothing. The SetpointHQ Index rescores its cohort monthly against a published rubric. GEO is a discipline with a scoreboard, not a one-off project.

Points 1 to 6 you can audit yourself in an afternoon. Point 7 needs a baseline: check what the assistants say about you today before you change anything, so you can tell later whether the work moved the number. If you want to know why one employer gets named and a larger one does not, that’s how ChatGPT decides who to name.

Where recruitment brands lose the most points

Score enough employer brands and a pattern holds sector after sector. Knowledge graph scores sit in the 90s. Schema scores don’t.

Across the 100 UK employer brands on The SetpointHQ Index, the median schema score is 35.0 out of 100, and 21% of the cohort scores zero. The median knowledge graph score, by contrast, is 89.4, with only 6% at zero. Nearly every brand in the cohort is a known entity. Barely a third of them have marked up their own site well enough for a machine to read it.

Read the first and last columns together and the problem states itself. The 2 dimensions the cohort performs worst on, citation at 27.5 and schema at 35.0, are the 2 carrying the heaviest weight, 55% of the score between them. The 3 dimensions most brands are comfortably passing are worth 15% each. The median composite lands at 54.9 not because brands are failing everywhere, but because they are failing precisely where it counts.

That gap is the whole opportunity. A brand a model already knows about but can’t extract facts from is the easiest fix on this list, the entity work is done, only the markup is missing. A brand genuinely unknown to the knowledge graph has a longer road.

The UK retail sector illustrates it cleanly, and the 2 brands at either end of it make the point better than any average can.

The machine can recite Sainsbury’s store count from memory and can’t read a single one of its live vacancies. Retailers spend millions making people feel something about working there. The markup that lets a machine read the jobs costs almost nothing, and most of the sector hasn’t done it.

How to run this yourself, this week

5 checks, roughly in order of effort:

1. Run the Rich Results Test against your corporate homepage and one live job page. Confirm Organization appears on the first and JobPosting on the second. If either is missing, that’s the first fix, before anything else on this list.
2. Check robots.txt for GPTBot, ClaudeBot, PerplexityBot and Google-Extended by name. A blanket disallow written for a different reason, stopping scraper spam, say, can catch all 4 without anyone noticing.
3. Load your careers page with JavaScript disabled. Whatever’s missing is what a non-rendering crawler sees.
4. Search your company name on Wikidata. If there’s no entry, or the entry points at the wrong URL, that’s a fixable half-hour, not a project.
5. Ask ChatGPT, Claude, Gemini and Perplexity a question a candidate would actually ask about you, not “tell me about [company]” but something closer to “is [company] a good employer for [a specific role]”. Note whether you’re named, and what’s said if you are. That’s your baseline. Everything above this line is aimed at moving it.

If you’re an agency, not an employer

Everything above still applies, with one caveat worth knowing before you benchmark yourself against it. The Index’s citation questions are written from an employer’s perspective, is this a good place to work, because that’s the buyer the underlying research is built for. An agency asking the same questions of itself will score lower than the true picture, because the probe is measuring the wrong entity’s recommendability.

That doesn’t mean GEO doesn’t matter for agencies. It means the citation number specifically should be read as a floor, not a verdict, until the probe set catches up. The other 6 checklist items, schema, crawl access, server rendering, Wikidata, concrete facts, measurement on a cadence, measure exactly the same thing for an agency’s own site as they do for an employer’s. An agency invisible to GPTBot is exactly as invisible as an employer would be. Fix those 6 regardless of which side of the desk you sit on.

GEO gets you in. AEO gets you named.

None of the above guarantees you get recommended. It guarantees you’re eligible to be. GEO is entry to the machine’s working material, the facts it can find, parse and trust about you. What happens after that, whether the machine actually names you in the answer a candidate reads, is a separate discipline with its own mechanics. Its sibling discipline, answer engine optimisation, covers the content shape and proof density that turns a readable brand into a cited one. If you’re specifically trying to get named as a source agencies and employers can point to, here’s how to get cited by AI assistants directly.

Frequently asked questions

What is generative engine optimisation (GEO)?

GEO is the discipline of making your employer brand and careers content readable, retrievable and reusable by AI assistants including ChatGPT, Claude, Gemini and Perplexity. Where SEO earns a ranking, GEO earns a place in the raw material a generative engine draws its answers from.

Do I need JobPosting schema for AI assistants to read my job ads?

Yes. Organization markup tells a machine who you are. JobPosting markup tells it what you’re hiring for. Without JobPosting on your live roles, your vacancies exist as pictures of jobs rather than structured data about jobs, and most retrieval can’t extract detail it can’t parse.

How often should I check my GEO score?

On a cadence, not once. Models and their indexes shift constantly, so a single reading only tells you where you stood on that day. The SetpointHQ Index rescores its cohort of 100 UK employer brands monthly against a published rubric for exactly this reason.

Is our employer brand ready for AI search?

Most aren’t, and the reason is usually schema, not reputation. Across the Index cohort, the median knowledge graph score is 89.4, nearly every brand is a known entity, against a median schema score of 35.0, with 21% scoring zero. Brands are known. Far fewer are actually readable.

What’s the difference between GEO and AEO?

GEO gets your content into the material a generative engine can draw on. AEO is about winning the citation itself, being named in the final answer rather than merely present in the source data. The full comparison with SEO is on our SEO vs GEO vs AEO page.

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. Retail sector figures as at 13 July 2026.