People searching for a recruitment agency or a marketing partner increasingly don’t type a keyword any more, they describe their problem to an assistant in a full sentence and take whoever it names. We see the pattern directly in our own search data: recruitment buyers typing questions like “how do I get a recruitment agency listed as a reliable source in ChatGPT” and “we’re missing from almost all the best-of lists and industry directories in our sector”, in full, as prompts. Nobody types a paragraph into a search box by accident. That’s someone who already knows they have this problem and is looking for whoever solves it.

What “getting cited” actually requires

Citation sits downstream of 2 things that have to be true first. Our GEO guide covers the first: can a crawler reach your content and can a machine parse it, structured data, crawl access, a clean entity. Our AEO guide covers the second: does your content give a model something specific enough to quote and attribute to you by name. Get both right and you’re eligible to be cited. This page is about the part neither of those covers: actively earning the citation once you’re eligible, on your own site and beyond it.

The best-of-list problem

A lot of what an assistant cites when it answers “who’s good at this” isn’t your site at all. It’s a roundup, a directory, a trade-press “best of” list, a comparison site, sometimes a Reddit thread where someone asked the same question 2 years ago and got real answers. If you’re missing from “almost all the best-of lists and industry directories in your sector”, fixing your own site’s schema doesn’t touch that problem, because the assistant is drawing on a source you don’t control and can’t edit.

That means citation strategy has 2 tracks running at once, not one. Track 1 is your own site: the GEO and AEO work. Track 2 is everywhere else an assistant looks for corroboration: trade press, industry directories, comparison and review sites, Wikipedia, YouTube, Reddit. Research on what AI summaries actually cite has found Wikipedia, YouTube and Reddit alone supply roughly 15% of the links referenced, a share no amount of on-site schema work will ever capture, because it isn’t about your site at all.

Most recruitment marketing budget goes entirely to track 1. Track 2 is cheaper and mostly untouched, because nobody thinks of a trade-press mention or a directory listing as SEO work any more, even though it now does more for citation than most of what track 1 spends on.

Which third-party sources actually matter

Not every mention is worth the same to a model, and the right target depends on which side of the desk you're on. An agency chasing citations should care about trade press covering recruitment specifically, sector directories a client would actually browse, and comparison content that names agencies by specialism and region. A generic “top 100 UK companies” list rarely gets pulled into a question about who fills warehouse roles in the North West, because the question and the list are matched on different criteria.

An employer chasing citations for candidate-side questions cares about a different set: genuine review platforms, “best places to work” style coverage where the methodology is real, and press that covers the specific thing candidates ask about, pay, flexibility, progression, rather than general company news.

The test either way is simple: would this source plausibly get pulled up if someone asked an assistant the exact question your buyer or candidate is asking. If yes, it’s worth pursuing. If it only proves you exist, it’s a nice mention and not a citation asset.

A 30-day plan

1. Week 1, baseline and entity cleanup. Ask ChatGPT, Claude, Gemini and Perplexity the exact question your buyer would ask, and write down whether you’re named and who is instead. Fix any mismatch between your trading name, your registered name, your LinkedIn page and your Wikidata entry while you’re looking, this is the cheapest fix on the whole list and it undermines everything else until it’s done.
2. Week 2, machine-readable layer. Confirm Organization markup on your root and JobPosting or Service markup wherever relevant. Check robots.txt allows GPTBot, ClaudeBot, PerplexityBot and Google-Extended by name. This is the GEO checklist in full, done in a week rather than left open indefinitely.
3. Week 3, publish 2 to 3 citable pages. Not more content, more specific content. Each page should carry at least one number a competitor couldn’t also claim: a fill rate, a time-to-hire figure, a retention rate, a sector specialism backed by a placement count. Genuinely comparative content, where you honestly place yourself against the market rather than only praising yourself, performs disproportionately well here.
4. Week 4, pursue 1 or 2 third-party mentions. A trade-press quote, a directory listing, a genuine case study a client will let you name. Pick targets that already rank or get cited themselves, being mentioned by a source a model already trusts transfers some of that trust to you.

What “reliable source” means to a model, not to you

“Reliable” isn’t a badge an assistant hands out once. It’s a pattern it re-derives every time it’s asked, from whatever evidence exists at that moment: is the entity clean and consistent, is the claim specific and checkable, does independent corroboration exist, has this held up over repeated queries rather than appearing once and vanishing. The SetpointHQ Index rescores its cohort of 100 UK employer brands monthly against a published rubric for the same reason a model re-derives reliability every time, a single good answer on one date proves very little about the next one.

That has a practical consequence most people miss: consistency over months beats a single strong push. A brand that quietly keeps its entity clean, its claims specific and its third-party mentions current will out-compound a brand that does one big content push and leaves it untouched for a year, even if the one-off push was more impressive on the day it launched.

Where this goes wrong

3 mistakes we see agencies and employers make once they understand citation matters and start chasing it directly.

1. Stuffing FAQ pages with every question a competitor might rank for. A page that answers 40 loosely related questions with a paragraph each reads to a model exactly as it reads to a person: nothing on it is specific enough to trust over a page that answers one question properly.
2. Inventing or rounding statistics to sound more impressive. A model that cites a figure and later finds it contradicted elsewhere, by you or by someone else, doesn’t just discount that claim, it discounts the entity making it. The honest, checkable number beats the impressive, unverifiable one every time this gets tested.
3. Gaming schema with markup that doesn’t match the visible page. Structured data that claims something the rendered content doesn’t support is exactly the kind of inconsistency the knowledge-graph and citation dimensions are built to catch, and it tends to cost more trust than the missing markup would have.

How to check whether it’s working

Traffic and impressions won’t tell you this. A page can gain visits from people who found it on Google while still never getting mentioned by name inside a ChatGPT or Perplexity answer, and the 2 numbers move independently of each other. The only real test is the blunt one: open each of the 4 assistants, type the exact question your buyer would type, and note who gets named. Do it once as a baseline, then repeat monthly on the same wording, because the answer changes as the underlying evidence changes and a single check tells you almost nothing about the trend.

Keep the log simple. Date, assistant, exact prompt, who was named, where you ranked in the list if you appeared at all. After 3 or 4 months a pattern shows up that a single snapshot can’t: whether you’re gaining ground, holding steady while a competitor gains, or invisible regardless of what you’ve shipped. That last result usually means the machine-readable layer, not the content, is still the blocker, worth ruling out before writing more pages.

This is the same method the SetpointHQ Index runs at scale, monthly, across 100 employer brands and 4 assistants, because a one-off score tells a brand almost nothing about direction. Running it by hand on your own brand costs nothing but 20 minutes a month. Running it properly, tracked, benchmarked against a named cohort, with the drift explained rather than just logged, is the job Pro does for brands who’d rather not do it by hand every month.

Frequently asked questions

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

Fix the machine-readable layer first, structured data, crawl access, a clean and consistent entity. Then publish content with specific, checkable claims rather than generic positioning, and pursue genuine third-party mentions, press, directories, reviews, since independent corroboration carries more weight than anything on your own site.

Why is my recruitment agency not visible in Gemini specifically?

The same evidence gaps that hide you from one assistant usually hide you from all 4, blocked crawlers, missing structured data, an inconsistent entity, no specific claims to quote. It’s rare for a brand to be well cited on 3 assistants and invisible on a 4th; check whether Gemini’s crawler specifically is blocked before assuming the content itself is the problem.

Why aren’t we in the best-of lists and directories our sector gets cited from?

Because those lists are compiled by people or processes you haven’t engaged, not generated automatically from your website. Getting onto them is closer to PR and outreach than to SEO: pitch trade press, apply to relevant directories, and build the kind of genuinely comparative, checkable content that makes a journalist or list compiler want to include you.

Can I pay to be cited by an AI assistant?

No. There’s no paid placement inside a synthesised answer the way there is in a sponsored search result. The assistant weighs evidence about how recommendable you are, and evidence has to be earned across your own site and independent sources, not bought.

How long does it take to start showing up in AI answers?

The machine-readable fixes, crawl access, structured data, entity cleanup, can be reflected within days once the crawlers next visit. Being named consistently for a specific buyer question usually takes longer. Most brands we’ve tracked see a real shift inside 2 to 3 months of consistent work, not one page and a wait. Anyone promising overnight citation results is describing a system that doesn’t exist.

Source: The SetpointHQ Index, 100 UK employer brands, scored monthly across ChatGPT, Claude, Gemini and Perplexity. Cited research on AI summary link sources is third-party (Pew Research), not Index data. Need the scoreboard running continuously rather than a single read? Talk to us about Pro.