Your agency is not showing in ChatGPT.
6 reasons it happens, in the order to check them. Most agencies fail on the first 2 and never get far enough to find out.
Book the £999 audit → Read the full picture →You typed your own agency name into ChatGPT and got something vague. Or you asked it to recommend recruiters in your sector and your region, and it named 3 firms that are not you.
Neither result is random. Assistants are not withholding you out of preference. They are working from what they could reach, parse and corroborate, and one of the 6 things below broke that chain. They are ordered by dependency, so work down and stop at the first one that applies. Fixing number 5 while number 1 is broken changes nothing.
A model cannot cite what it cannot reach. Everything else is downstream of that.
You are blocking the crawlers.
The commonest cause and the easiest to miss. Many agency sites were built or migrated by a developer who added a blanket disallow, or turned on a bot protection rule that treats AI crawlers as scrapers. Cloudflare added AI crawler blocking as a default on some plans, so this can happen without anyone deciding it.
How to check. Open your robots file at yourdomain plus /robots.txt and look for GPTBot, ClaudeBot, PerplexityBot, Google-Extended and CCBot. A disallow against any of those is a decision, and it needs to be a deliberate one.
What it looks like when it is the culprit. The assistant knows your company exists, often from a directory or LinkedIn, but cannot say anything specific about what you recruit. Vague, hedged, mostly right about the name and wrong about everything else.
Your jobs render in JavaScript.
Your vacancies are the single strongest evidence of what you actually recruit, at what level, in which places. On most agency sites they are pulled from an ATS and rendered client side, sometimes on a separate subdomain entirely.
Crawlers that do not execute JavaScript see an empty container. So the page that proves you place warehouse managers in Warrington reads as a heading and nothing else.
How to check. Load a job listing page, view source rather than inspect, and search the raw HTML for the job title. If it is not in the source it is not being read.
What it looks like. The assistant describes you generically as a recruitment agency and cannot name a single specialism, even though your site has 200 live roles in one sector.
Your entity is ambiguous.
You trade as one name. Companies House has another. LinkedIn shows a third with Ltd on the end. An old brand from before the rebrand still has a live site. A model trying to resolve which of those is a single organisation does the safest thing available to it, which is to stay quiet.
How to check. Search your trading name and your registered name separately in the assistant. Different answers, or one that returns a different company altogether, is the tell. Then check whether a claimed Wikidata item and a Google Business Profile exist and agree with each other.
What it looks like. The assistant confuses you with a similarly named firm, or hedges with a phrase like there appear to be several companies operating under this name.
Nobody but you says you are good.
This is the one that hurts, and it is the one that carries the heaviest weighting in our rubric at 35%.
Assistants corroborate. Your own site claiming sector expertise is an assertion. A trade publication, a client, an award listing, a podcast, a directory with real editorial, all of those are corroboration. Without any of it, the model has one source for a commercial claim, and the safest thing is to name someone else.
How to check. Search your agency name plus your main sector, excluding your own domain. Count the results that are not your own site, not a job aggregator republishing your ads, and not a press release you paid for. If that number is 0, this is your problem.
What it looks like. The assistant will name you if asked directly, accurately and dully, but will never volunteer you in answer to a question about who is good at something.
Your best proof is behind a form.
The case study with the fill rate in it is a gated PDF. The salary guide that would earn a citation asks for an email first. The testimonial page loads after a cookie wall that a crawler never clears.
Gating is a legitimate commercial choice, and it has a cost that nobody priced in when the choice was made. Content a model cannot read cannot be quoted, and a citation is worth considerably more than a name on a list you will email twice and give up on.
How to check. List the 5 pages you would send a prospective client. Count how many are reachable without a form, a login or a download.
What it looks like. Competitors with weaker results and worse websites get named ahead of you, because they published theirs and you gated yours.
Everything you publish is interchangeable.
Your sector pages would still read correctly if you swapped the sector. Your about page could belong to any firm of your size. There are no numbers, no named clients, no dates, no losses, no positions anyone could disagree with.
Models repeat specifics. Adjectives get dropped, because there is nothing in them to relay. A page that says we deliver exceptional talent solutions gives an assistant nothing it can pass on with your name attached. A page that says we filled 340 warehouse roles across 6 North West sites in 11 weeks, with a 91% retention rate at 12 weeks, gives it a sentence.
How to check. Take your main sector page, replace your agency name with a competitor name, and read it. If it still works, so does the problem.
Cut any sentence a competitor could put on their own site unchanged. Whatever survives is the only part a model will ever quote.
Work down, not across.
Reasons 1 and 2 are access. Reason 3 is identity. Reasons 4, 5 and 6 are evidence. There is no point commissioning content while a crawler cannot reach the site, and no point unblocking crawlers if everything they then read is interchangeable.
Access and identity are a fortnight of unglamorous technical work. Evidence takes a year and is the part competitors cannot copy quickly. Start the first, but start the second in the same week, because it is the one with the lead time.
The full version of this, with the 5 signals and the weightings behind them, is in how your agency shows up in AI search.
Find out which of the 6 is yours.
The £999 Recruitment Marketing Audit runs your sector and region questions across ChatGPT, Claude, Perplexity and Gemini, scores all 5 dimensions with the evidence behind each one, and tells you which link in the chain is broken. Back in 7 working days with a prioritised fix list and a 60 minute debrief. If the honest answer is that you have bigger problems than AI visibility, we will tell you that instead.
Book the £999 audit → Explore Studio, £149 per seat →