We paid a UK newswire to distribute our AI visibility study. Then we tested whether AI can actually read it.
By Alex Golombeck — The Happy Cat, Manchester. Published August 2026.
Yesterday we did a very normal piece of PR: we paid a UK newswire to distribute a press release. Today we discovered the release page returns "403 Forbidden" to AI assistant crawlers — while its own robots.txt says they're welcome.
If you spend money on press distribution and you care about AI visibility, what follows is worth five minutes. It ends with a test you can run yourself, in a terminal, before spending a penny.
Why we used a newswire at all
In our baseline study of how AI assistants recommend UK businesses, one finding stood out: Perplexity rarely cites a company's own website. It cites third parties — comparison articles, roundups, and in one logged case, a company that entered AI answers through a press release on a newswire.
That made a newswire submission an evidence-based move, not a hopeful one: an indexed release page on a trusted domain is exactly the kind of third-party source AI assistants draw on. So we distributed our study announcement through Pressat, a well-regarded UK service, and logged the live release page as our first placement.
Then, because checking is the whole methodology, we checked.
The discovery, step by step
Step 1: An AI assistant tried to read the page — and couldn't. A fetch of our release page by a genuine AI crawler (Anthropic's, from Anthropic's own infrastructure — not a simulation) was refused: the site disallows automated access.
Step 2: We read the robots.txt. A website's robots.txt is its published policy on who may crawl what. Pressat's is permissive: release pages are open to all user-agents; only housekeeping paths (search, dashboards, PDFs, embargoes) are disallowed. Google's news crawler is explicitly welcomed. According to the file, AI crawlers should walk straight in.
Step 3: We knocked on the door directly. From an ordinary Mac terminal:
curl -A "GPTBot" -o /dev/null -s -w "%{http_code}\n" https://pressat.co.uk/releases/study-chatgpt-perplexity-gemini-cited-104-sources-in-30-answers-and-cant-agree-on-any-of-them-1e3f27038c139ff0cacef81fa3511f93/
403A request identifying itself as GPTBot — OpenAI's crawler — receives 403 Forbidden.
Step 4: We knocked again, wearing a different hat. The same request, from the same machine, seconds later, identifying as an ordinary Chrome browser:
curl -A "Mozilla/5.0 ... Chrome/127.0 Safari/537.36" -o /dev/null -s -w "%{http_code}\n" https://pressat.co.uk/releases/study-chatgpt-perplexity-gemini-cited-104-sources-in-30-answers-and-cant-agree-on-any-of-them-1e3f27038c139ff0cacef81fa3511f93/
200200 OK. Same page, same second, same computer. The only difference was the visitor's name.
The conclusion writes itself: the pages are blocked to AI crawlers specifically, and the block sits above the robots.txt — at the firewall or bot-management layer — in contradiction of the site's own published crawl policy. The front door sign says "welcome"; the bouncer says otherwise.
Why this matters (and to whom)
A growing share of what press distribution buys in 2026 is machine readership. Human journalists still matter, but the durable asset a newswire sells is an indexed page on a trusted domain — and increasingly, the readers of trusted pages are AI assistants deciding what to cite when someone asks them a question.
A release page that AI crawlers cannot read can still work through side doors: it can rank in Google and Bing (whose crawlers may be permitted), it gets syndicated to aggregator networks on other domains with their own access policies, and some AI systems draw on search-engine indexes rather than crawling directly. Our release isn't wasted. But the most direct route — an AI assistant reading and citing the release page itself — appears to be closed, and nothing in the product description mentions it.
We don't think this is malice. Aggressive bot protection is usually a side effect: firewalls ship with AI-crawler blocklists switched on, scrapers get blocked wholesale, and nobody re-checks the settings against the robots.txt they published years earlier. We've asked Pressat directly whether the block is intentional and will update this post with their answer.
But intent doesn't change the outcome for buyers. If part of your reason for paying a newswire is AI visibility, "can AI actually read the page?" is now a purchasing question — and as far as we can tell, nobody is asking it.
The same bug, three sites in a row
Here's the pattern that made us check at all. In three weeks of doing this work we've now seen the identical failure shape three times:
- Our own site blocked AI crawlers when it launched — default platform behaviour, fixed in an afternoon.
- A client's site had a robots.txt that looked fine in a browser but was corrupted underneath, with a security layer blocking above it.
- A national newswire's robots.txt permits AI crawlers that its server then refuses.
Three different organisations, same lesson: what a site says about crawler access and what its server actually does are two different facts, and only one of them is visible without testing. The robots.txt is the policy; the HTTP response is the truth. Every AI visibility audit we run now tests both, because any site — including ones you pay to publish on — can fail the second test while passing the first.
Test it yourself before you spend
You don't need us for this. Before paying any distribution service (or after fixing your own site), open a terminal and run:
curl -A "GPTBot" -o /dev/null -s -w "%{http_code}\n" https://example.com/a-real-release-page/Then the same command with a browser user-agent string instead of "GPTBot". Two numbers come back:
- 200 / 200 — open to both. Good.
- 403 / 200 — AI crawlers specifically blocked. The AI-citation route through that domain is closed, whatever the robots.txt says.
- 403 / 403 — the page may be blocking all automated requests including your curl; test from the target's robots.txt and other pages before concluding anything.
Swap "GPTBot" for "ClaudeBot" or "PerplexityBot" to test other engines. Sixty seconds per vendor, and you'll know more about their product than their sales page tells you.
One honest caveat about the method: a curl faking a bot's name isn't identical to the real crawler, which arrives from the AI company's own verified servers — sophisticated bot management can tell the difference and treat them differently. In our case that distinction doesn't rescue the result, because a genuine AI crawler was refused too (that's what started this). But if your own test shows a 403, treat it as a strong signal worth confirming, not an absolute verdict.
What we're doing with this
Our release stays where it is — the search-index and syndication routes remain live, and we'll be watching which copies of the story, on which domains, actually earn AI citations when we re-run our 30 baseline queries on 16 September. That's now a more interesting experiment than the one we paid for: not just "does a newswire placement get cited," but "which links in the distribution chain are actually readable by the systems doing the citing."
Whatever the answer, it'll be published here, with the log to back it up.
The Happy Cat is an AI visibility (GEO) agency in Manchester. We document our methodology in public — including the parts where we discover our £-spend works differently than advertised. If you'd like us to test what AI assistants can see of your business, that's exactly what our audit does.