CC
AI Visibility Audit
This report is confidential. Enter the access code provided by Saigon Digital to continue.
Incorrect code. Please try again.
Prepared by saigon.digital · September 2026
This report is confidential. Enter the access code provided by Saigon Digital to continue.
Parents no longer just search Google. They ask ChatGPT, Gemini and Perplexity which schools to shortlist, and the answer they get is built from whatever those engines are allowed to read.
Saigon Digital helped us increase awareness of BVIS and support admissions for the opening of our new campus. Through a focused SEO and AI visibility strategy, they helped more parents and families discover BVIS when searching online for international school options. Their work strengthened our digital presence and helped us reach prospective families at an important time for the school.
See how we've helped brands grow. Read our case studies and learn more about what we do.
View Case StudiesA snapshot of where Clifton College stands in the AI search era, and what the current gap is costing.
Clifton College has done something almost no other UK independent school has done. You publish an llms.txt file and EducationalOrganization schema written specifically so that AI models describe the school accurately: independent, coeducational, boarding and day, ages 3 to 18, Pre-Prep through Sixth Form. On 9 September 2026 we asked those AI models to go and read it. Every single one was refused. Fourteen AI user agents, including GPTBot, OAI-SearchBot, ChatGPT-User, PerplexityBot, ClaudeBot and Google-Extended, receive a 403 Forbidden from your edge firewall. Googlebot, Bingbot and every browser receive 200. Badminton, Bristol Grammar, QEH and Redmaids’ High publish no llms.txt at all, and all four are completely readable. The one school in Bristol that did the work is the only one the answer engines are locked out of.
User-Agent: * / Allow: /, with a single Disallow for /private/. The block sits in front of it, at the Vercel edge, which returns 403 with the header x-vercel-mitigated: deny to AI user agents only. It is a targeted AI-bot rule, not general bot protection: curl, python-requests, AhrefsBot, SemrushBot, Twitterbot and facebookexternalhit all pass through at 200. That is why no SEO tool, Search Console report or site crawl has ever flagged it.
We tested whether the four major answer engines can reach cliftoncollege.com at all. This is a request-level test, not an opinion about what any model says on a given day: same IP, same minute, only the user agent changed.
GPTBot, OAI-SearchBot and ChatGPT-User all return 403. Both the training crawler and the live-fetch crawlers are refused, so ChatGPT cannot verify anything on cliftoncollege.com.
Googlebot is allowed (200), so AI Overviews can still draw on the Google index. Google-Extended, the agent that governs generative use, is blocked at 403.
PerplexityBot returns 403 on the homepage, the fees page and the news announcement. Perplexity fetches live, so it can only answer about Clifton from other people’s pages.
Google-Extended returns 403. That is the specific user agent Google honours for Gemini and its generative surfaces.
1 of 4 platforms currently surface Clifton College in relevant AI-generated recommendations.
Access first. After that: pages that state one fact completely, consistent entity signals across Wikipedia and the school directories, and third-party mentions that agree with your own site. Clifton already has the raw material for all three.
We sent 25 requests to https://www.cliftoncollege.com/ from the same machine, on the same connection, within the same minute on 9 September 2026. The only thing that changed between them was the User-Agent string. Fourteen were refused. Eleven were served.
| User agent | What it feeds | Response |
|---|---|---|
GPTBot | ChatGPT model knowledge | 403 |
OAI-SearchBot | ChatGPT Search index and citations | 403 |
ChatGPT-User | Live fetch when a parent asks about a page | 403 |
PerplexityBot | Perplexity answers and sources | 403 |
ClaudeBot / anthropic-ai | Claude | 403 |
Google-Extended | Gemini and Google generative surfaces | 403 |
Applebot, Applebot-Extended | Apple search and Apple Intelligence | 403 |
meta-externalagent | Meta AI | 403 |
Amazonbot, Bytespider, cohere-ai, CCBot | Alexa, Doubao, Cohere, Common Crawl | 403 |
Googlebot, Googlebot-Image | Google Search | 200 |
bingbot, DuckDuckBot | Bing and Copilot’s index, DuckDuckGo | 200 |
AhrefsBot, SemrushBot | Your SEO reporting tools | 200 |
curl, python-requests | Any generic script | 200 |
Twitterbot, facebookexternalhit | Social link previews | 200 |
| Chrome on macOS | A human visitor | 200 |
Generic scripts, scrapers and SEO crawlers all pass. The rule matches AI user agents specifically, and returns
403 with x-vercel-mitigated: deny, which is the signature of a managed AI-bot rule in the Vercel firewall.
It may well have been switched on deliberately. It is worth confirming that whoever switched it on knew it also blocks the
search and citation agents, not just the training crawlers.
curl -I -A "GPTBot/1.2" https://www.cliftoncollege.com/
→ HTTP/2 403
curl -I -A "Mozilla/5.0" https://www.cliftoncollege.com/
→ HTTP/2 200
cliftoncollege.com/llms.txt: 200 to a browser, 200 to Googlebot, 403 to GPTBot, PerplexityBot and ClaudeBot. It begins:
“Clifton College is an independent coeducational boarding and day school in Bristol for pupils aged 3 to 18, with Pre-Prep, Prep, Upper School and Sixth Form.”
That sentence was written for one audience. That audience is the only one that cannot read it.
These are the questions a Bristol parent actually asks. For each one we tested whether the answer engines can reach the Clifton page that would answer it.
The pattern has nothing to do with your content. Every page a parent needs is written, published and returning 200 to a browser. The block sits in front of all of it and applies only to AI user agents. Classic search crawlers pass straight through, which is exactly why your Google rankings look healthy and nothing has flagged this.
This is the rare gap that is one configuration change away from closing. The content exists, the llms.txt is written, the schema is correct. Allow the answer engines through and Clifton goes from unreadable to the best documented independent school in Bristol, ahead of four rivals who have published nothing for AI at all.
Every other independent school in Bristol serves its pages to the answer engines without restriction. Clifton has the strongest domain authority of the group and is the only one they cannot read. Access tested live on 9 September 2026, homepage request, redirects followed.
| Company | DR | ChatGPT GPTBot |
Gemini Google-Extended |
Perplexity PerplexityBot |
Why They Win |
|---|---|---|---|---|---|
| Clifton College You | 42 | Blocked | At Risk | Blocked | Audit target |
| Badminton School | 38 | Open | Open | Open | No llms.txt and no AI strategy we can detect, but every crawler reads the whole site unimpeded. |
| Bristol Grammar School | 36 | Open | Open | Open | 155 ranking keywords against your 95, and nothing between their pages and the answer engines. |
| Queen Elizabeth’s Hospital (QEH) | 29 | Open | Open | Open | A third of your authority, but fully readable, so models can describe QEH first-hand and Clifton only second-hand. |
| Redmaids’ High School | 28 | Open | Open | Open | Frequently listed first in Bristol school round-ups, and every one of those pages can be verified against the school’s own site. |
| Clifton High School | 14 | Not tested | Not tested | Not tested | Their .sch.uk domain did not resolve from our test network, so we have not claimed a result. DR from Ahrefs. |
Badge key: Open (200) Partial Blocked (403) Not tested
Of 2,299 monthly organic visits, 2,086 come from searches that already contain “Clifton”. Those are people who had already heard of you. Discovery searches such as “boarding schools bristol” and “private schools bristol” bring in roughly 148 visits a month between them, and a slice of the remaining non-branded traffic is the uniform shop (“black school pumps”, “navy tights”, “grays hockey bag”) rather than admissions.
Reputation and word of mouth are already doing the heavy lifting. The part of the funnel that finds you when a family has not heard the name yet is the thin part, and AI answers are where that part of the funnel is moving. Right now it is the one channel Clifton is not in, while every rival in Bristol is.
These are the highest-leverage changes Clifton College can make right now to start appearing in AI-generated recommendations within 30–90 days.
These are not the same thing, and the distinction is the whole point. GPTBot and CCBot are bulk training crawlers, and there are reasonable arguments for refusing them. OAI-SearchBot, ChatGPT-User, PerplexityBot, ClaudeBot and Google-Extended are the agents that fetch a page in order to cite it in an answer a parent is reading right now. Blocking those does not protect your content, it removes you from the answer. Allowing them is one rule in the Vercel firewall, and nothing else on the site has to change.
While the block is in place, and for as long as it takes models to re-crawl afterwards, third-party pages are the source of truth about Clifton. Wikipedia, ISC, the Good Schools Guide and the main directory profiles need Dr Chris Stevens as Executive Head, the 3 to 18 age range with Pre-Prep and Prep spelled out, and the CCEG structure including Tockington Manor and ELC Bristol. These are the sentences models are repeating today.
llms.txt is a summary. Models cite pages. Once access is restored, the pages that win admissions answers are the ones that state a fact plainly and completely: fees by year group and boarding type, entry points at 3, 7, 11, 13 and 16, the difference between full, flexi and occasional boarding, EAL and guardianship for international families, and open event dates. Clifton has all of this. Very little of it is written in the form a model can lift into an answer.
While auditing how AI sees Clifton College, we also looked at the website itself (Next.js on Vercel, Storyblok CMS). These specific issues are holding back both your AI visibility and how the site converts, and they are all fixable.
The 403 is returned at the Vercel edge, above the application, with x-vercel-mitigated: deny. It will not show in Search Console, Screaming Frog, Ahrefs or Semrush, because AhrefsBot and SemrushBot are on the allowed side of the rule and get 200. If nobody thinks to send a request as GPTBot, the site looks perfect. Worth asking whoever owns the Vercel project whether this was a deliberate decision or a default that came with the build.
The homepage uses <h1> four times: “Welcome to”, “Clifton College”, an italic pull-quote, and “A love for learning”. Extraction models lean on the H1 to decide what a page is about, and four of them dilutes that signal. One H1 that names the school and what it is, with the rest demoted to H2, is a ten-minute fix in Storyblok.
/official-completion-of-charitable-merger-of-tms-into-the-cceg/ 308-redirects to the same path without the trailing slash, which then returns 404. That announcement is the main public record of the Tockington Manor merger into CCEG, it is still indexed, and it is currently a dead end for anyone following an old link, human or model. Worth a sweep of the pre-migration URLs for others like it.
Web development and WordPress are core strengths at Saigon Digital. Fixing the issues above is part of the work, not a separate project. We handle the AI visibility, the SEO, and the website itself.
This audit shows the problem. Fixing it usually means more than AI alone: the website, the SEO and the AI visibility all work together. We handle all three, then keep it growing.
Clifton College has a clear path to winning AI and search visibility, and a website that supports it. The gap to competitors is real but closeable. We typically start with a fixed-price foundation sprint (AI and GEO, SEO, and the website fixes above), then a simple monthly retainer to keep growing and maintain the site. A 30-minute call is all it takes to map it out.
Nick Rowe · CEO & Co-Founder, Saigon Digital
A fixed-price 3-month sprint: get cited in AI answers, fix the SEO foundations, and reposition and repair the website (development and WordPress included). One scope, one price.
A simple monthly retainer once the foundation is set: continued SEO and AI visibility growth, content, and website maintenance and support. Rolling monthly, no long lock-in.
Every month your competitors build more authority signals, the gap widens. AI models are training on content published now, so delay compounds the problem.