How glassmark reads your website
Six dimensions of AI readiness. One score. A clear path forward.
A website that works for humans is not the same as a website that works for AI
Your website was built for people. It looks good, it reads well, and your visitors can find what they need. But AI agents do not read websites the way humans do. They skip the pictures, ignore the layout, and pull structured signals out of your code. If those signals are missing or unclear, the AI has nothing to tell someone who asks about you, no matter how good your site looks on screen.
- A Google Map embedded in your contact page
- A photo of your opening hours on the door
- Beautiful pictures of your rooms, your food, your experiences
- A paragraph describing your business in warm, evocative language
- A third-party iframe with no location data
- An image file with no machine-readable hours
- Pictures with no alt text or structured descriptions
- Prose that cannot be cross-referenced against known schema
Same website. Two completely different stories.
This is not an SEO race. It is a visibility race.
Two decades ago, search engine optimisation became the arena where big budgets won. Agencies, link-building, content farms, technical teams. SMEs watched the rules of online visibility get written by companies that could afford to spend their way to the top. Independent operators in the visitor economy felt it most.
AI visibility is at the same moment now. ChatGPT, Claude, Perplexity, Google's AI. They are deciding right now who gets recommended when a traveller asks "where should I stay in Cornwall" or "who caters weddings near Padstow". The chains, the OTAs, the platforms with engineering teams are already structuring their data. Most independent operators are invisible.
glassmark exists because that second arena does not have to be a repeat of the first. AI readiness is measurable. The fixes are practical. The playing field can stay level, but only if independent operators measure and act before the space is crowded out.
See what AI sees. Before everyone else does.
glassmark is building an open benchmark of visitor economy AI readiness as scans accumulate. The data stays anonymous and aggregated.
Six categories, examined in depth
Six lenses on how well AI agents can discover, understand, and recommend your business. The relative weight of each category adjusts based on the type of business you run, so a hotel, a restaurant, and a destination are scored against the signals that matter most to each.
Identity
What kind of business is this? Can an AI agent tell at a glance whether it is a hotel, a restaurant, a surf school, or a visitor attraction?
Place
Where is the business, really? Not just a postcode, but the kind of location context that helps an AI agent recommend it to someone asking about a specific area.
Answers
What can a visitor actually find out from your website? Opening times, prices, availability, accessibility, pet policies. The practical facts that drive decisions.
Proof
Why should anyone trust this business? Reviews, ratings, awards, accreditations, and other signals that give an AI agent confidence in a recommendation.
Action paths
How does a visitor transact? Can they book, buy, reserve, or enquire directly from the information on your site, or does the trail go cold?
Access
Can a machine reach your content at all? Structured data, meta tags, robots.txt, sitemaps, and the technical plumbing that determines whether an AI agent can read your site.
Four levels of AI readiness
Hidden 0–39 points
AI agents can read very little of what matters. Asked about your business, an assistant has almost nothing from your site to go on.
Partial 40–59 points
AI agents can read some of what matters and miss the rest, so answers about you come out incomplete or vague.
Visible 60–79 points
AI agents can read most of what matters. Your core facts are there; a few questions still cannot be answered from your site.
AI-ready 80–100 points
An AI agent can read who you are, where you are, what you offer, and how to book, without guessing.
What the score measures
This score measures how readable your website is to AI agents, using 31 fixed checks across six lenses. It does not predict whether ChatGPT, Claude, or Google will mention you. No tool can promise that.
Each glassmark scan reads your pages the way an AI agent does: the words on the page, the structured facts behind them (schema), and the files that tell crawlers what they may read. It never asks an assistant about you. Every check is fixed and repeatable: the same page, read the same way, gets the same result. What can change between scans is your site, the pages we could reach, and occasionally the rules themselves. We tell you which.
| Lens | Weight | What an assistant can do when it is Good | Checks |
|---|---|---|---|
| Identity | 20% | Say what kind of business you are, be sure which business it is describing, and give a traveller your phone, email, and address. | 3 |
| Place | 15% | Put you on a map, name the town or area you are in, say which region you belong to, say what is near you, and describe the area around you. | 5 |
| Answers | 20% | Say when you are open, give a rough idea of what you cost, answer the questions you are asked most often, list your facilities, say whether you take dogs, describe how accessible you are, and describe a visit in your own words. | 7 |
| Proof | 15% | Quote your overall rating, quote what guests have said, name the awards you have won, point to your independent reviews, and tell that your photos are of the real place. | 5 |
| Action paths | 15% | Explain how to book, give the right number to call, give the right email address, read your booking or enquiry page, and give directions to you. | 5 |
| Access | 15% | Tell that it is allowed to read your site, know which pages exist, read a short guide to your site, tell which version of a page is the real one, and read your pages without running scripts. | 5 |
Each check reads as Good, Partial, Missing, or Undetermined. Undetermined means we could not confirm it either way, for example analytics that only load after a cookie banner. It counts neither for nor against you. One further check, analytics tracking, is about measuring visits rather than what a traveller is told, which is why the table shows 30.
What this score is and isn't
- 1
Access is documented. If your site blocks OpenAI's search crawler, your pages are not in ChatGPT search answers. That is the one lens with a direct, published consequence. OpenAI's crawler documentation.
- 2
Identity, Place, Answers, Proof, and Action paths are about clarity. Structured facts help AI agents describe you accurately. Google says no special markup is needed for its AI features, so treat these as clarity, not a ranking lever. Google Search Central on AI features.
- 3
llms.txt is informational only. No major AI provider has committed to reading it, so it carries no points here. The llms.txt proposal.
The score is a function of your site alone. Nothing you do on your dashboard changes it, and nothing you tell us about your business changes it either. If you want it to move, change the site.
What moves a score
Only three things can change a number between two scans. Your dashboard labels each one beside the comparison, so a change is never left to guesswork.
Your site changed
A check flipped because the page it reads changed. The dashboard lists every flip with the points that moved and what an assistant can now do.
We read different pages
A sample scan reads 3 pages; a full report reads up to 26. If two scans differ in pages read, or one was cut short, the dashboard says so before it shows the delta.
Our rules changed
When a scoring rule changes we date it and mark it on your chart, so a shift that is ours is never reported as yours.
Scoring rule changes
- 10 September 2026: Business identity selection was fixed. When a page declares more than one organisation, the scan now reads the business itself rather than the website publisher.
What happens next
The free scan gives you your overall score, your maturity band, all six category scores, and your top three priorities. That is often enough to start a conversation with your web developer or marketing team.
If you want the full picture, the £99 one-off reading gives you every finding across all six categories, detailed developer remedy blocks, a tailored prompt kit, a downloadable branded PDF, and five rescans across 90 days so you can measure progress as fixes land.
Your reading also includes watch during the same 90-day measurement window. It attempts a monthly check and compares it with the last completed reading. If a check is overdue or cannot complete, your dashboard says so clearly. Paid ongoing monitoring is separate, optional, and never starts automatically.