AI Readiness Assessment
https://www.essex.gov.uk
Governancestage 2 · not assessed yet...more
43 judgements need a named person in the organisation to confirm, because they turn on what you know to be true rather than anything measurement can settle: whether a claim is supported, whether a page is still current, which of two versions is the one to quote. In order, they cover: who is answerable for this content; is it still true today; can each claim be backed; does our published content agree; will the right meaning come across; can someone reach it and act on it. An answer recorded against each, with a name and a date, is what turns published content into governed content.
Visibilitystage 3 · not assessed yet...more
Once the foundations are sound and the content is governed, the visibility stage measures how AI systems and agents actually find the estate: which pages they reach, how often, and how the organisation is represented when they answer questions about it, tracked over time. Building visibility on unconfirmed foundations only amplifies whatever is wrong, which is why it comes last.
How much of your content is reachable and readable, rather than blocked, buried or missing. Higher is better.
- No page is reachable by an AI system, so what is said about the organisation is settled by other people's material, with nothing of the organisation's own in it. pages
Where your content could lead an AI system to state something incorrect. Only direct routes to a wrong answer count here.
- With the organisation's own pages unreadable, anything an AI states about Essex is assembled from directories, news, forums and opinion, with no first-party page to check it against or to correct it.whole estate
AI systems were refused or given no answer at the door. The site returned them no readable content.
This is an access finding, and it affects AI systems, not visitors. It does not stop AI systems answering questions about the organisation: they answer them either way. It decides whose material the answer is built from. With the pages closed to them, what is said is put together from directories, news, forums and opinion, and the organisation has no hand in it.
Anyone asking for this site, a person on a phone or an AI system building an answer, is let in, turned away, or left with no answer, in milliseconds, by machinery most organisations never see. An AI system that gets nothing here reads nothing at all: to that system the organisation simply does not answer.
The proof behind this picture
29 live agents were first set up and readied for this assessment. 26 were sent to this door: 4 people and 22 AI systems, in the order shown on the chart; the other 3 stood ready as backup, to be called in depending on how the door answered. Of those sent, 24 made one timed call each, in their own name, asking for the same page. The 2 desktop browsers made a full visit, exactly as a person at a computer would. Nothing here is simulated: these are real calls to the live site, recorded as they happened, and each line below carries the moment it happened. Alongside them, two further readings were taken at the door: the site's security certificate, and its published rules for crawlers.
| Visitor | When | What happened | Answer | Took | Sent | Received | Answered from |
|---|---|---|---|---|---|---|---|
| People | |||||||
| iPhone user | 16:02:07.853 | ✓ served | 200 | 23 ms | 250 B | 51.9 KB | 18.65.194.20 |
| Android user | 16:02:07.853 | ✓ served | 202 | 21 ms | 232 B | 584 B | 18.65.194.20 |
| Safari on a Mac | 16:02:12.171 | ✓ served | 200 | 256 ms | n/a | n/a | n/a |
| Chrome on a PC | 16:02:12.746 | ✓ served | 200 | 3231 ms | n/a | n/a | n/a |
| AI systems | |||||||
| GPTBot · OpenAI | 16:02:07.730 | ✓ served | 200 | 32 ms | 217 B | 51.9 KB | 18.65.194.20 |
| OAI-SearchBot · OpenAI | 16:02:07.730 | ✓ served | 200 | 37 ms | 294 B | 51.9 KB | 18.65.194.20 |
| ChatGPT-User · OpenAI | 16:02:07.730 | ✓ served | 200 | 33 ms | 220 B | 51.9 KB | 18.65.194.20 |
| ClaudeBot · Anthropic | 16:02:07.730 | ✓ served | 200 | 30 ms | 180 B | 51.9 KB | 18.65.194.20 |
| Claude-SearchBot · Anthropic | 16:02:07.730 | ✓ served | 200 | 33 ms | 232 B | 51.9 KB | 18.65.194.20 |
| Claude-User · Anthropic | 16:02:07.730 | ✓ served | 200 | 35 ms | 222 B | 51.9 KB | 18.65.194.20 |
| Googlebot (Gemini) | 16:02:07.761 | ✓ served | 202 | 22 ms | 256 B | 584 B | 18.65.194.20 |
| GoogleOther | 16:02:07.763 | ✓ served | 202 | 29 ms | 221 B | 584 B | 18.65.194.20 |
| Bingbot (Copilot) · Microsoft | 16:02:07.764 | ✓ served | 202 | 30 ms | 255 B | 584 B | 18.65.194.20 |
| PerplexityBot | 16:02:07.765 | ✓ served | 200 | 69 ms | 234 B | 51.9 KB | 18.65.194.20 |
| Perplexity-User | 16:02:07.766 | ✓ served | 200 | 27 ms | 238 B | 51.9 KB | 18.65.194.20 |
| CCBot · Common Crawl | 16:02:07.767 | ✓ served | 200 | 28 ms | 155 B | 51.9 KB | 18.65.194.20 |
| Bytespider · ByteDance | 16:02:07.784 | ✓ served | 200 | 23 ms | 263 B | 51.9 KB | 18.65.194.20 |
| Amazonbot | 16:02:07.793 | ✗ turned away | 403 | 26 ms | 223 B | 1.2 KB | 18.65.194.20 |
| GrokBot · xAI | 16:02:07.793 | ✓ served | 200 | 26 ms | 167 B | 51.9 KB | 18.65.194.20 |
| Meta-ExternalAgent | 16:02:07.794 | ✓ served | 200 | 32 ms | 204 B | 51.9 KB | 18.65.194.20 |
| Meta-WebIndexer | 16:02:07.796 | ✓ served | 200 | 28 ms | 206 B | 51.9 KB | 18.65.194.20 |
| Meta-ExternalFetcher | 16:02:07.807 | ✓ served | 200 | 24 ms | 206 B | 51.9 KB | 18.65.194.20 |
| Applebot | 16:02:07.820 | ✓ served | 200 | 25 ms | 282 B | 51.9 KB | 18.65.194.20 |
| MistralAI-Training | 16:02:07.820 | ✓ served | 200 | 25 ms | 234 B | 51.9 KB | 18.65.194.20 |
| MistralAI-Index | 16:02:07.825 | ✓ served | 200 | 25 ms | 231 B | 51.9 KB | 18.65.194.20 |
| MistralAI-User | 16:02:07.827 | ✓ served | 200 | 25 ms | 230 B | 51.9 KB | 18.65.194.20 |
The desktop browsers are run in full rather than as a single call, so one figure for the information moved does not apply to them; their outcome and timing are shown.
What this shows
All 4 people were served. 21 of the 22 AI systems were served and 1 was turned away. The visitors fall into two groups, and each is judged on its own. 13 of the 13 services that answer people's questions were let in. 8 of the 9 services that gather material for future answers were let in, and Amazon was turned away. The services answering people today are all being served; those turned away gather material for future answers, so the cost falls on what these systems will know of the organisation later rather than on the answers being given now. Where a service arrived carrying a question, the question used was the organisation's own name, “Essex”, because that is what the public asks. The sizes tell the same story: a caller who is served receives the page in full, about 51.9 KB here, while a caller who is turned away receives a short refusal in its place, about 1.2 KB, and nothing more. A door of this kind decides call by call. This record is what happened at this door, at this time, from this address; it can answer differently at another time, which is why the time and the address are part of the record.
How this stands up
These calls can be repeated on request while you watch. A separate checking tool, kept apart from the assessment engine so the two cannot share a mistake, knocks on the same door several times over and records whether the answers hold still; each of its checks is saved with the date, the time and the calling address.
Whether the estate can be reached and read at all: access rules, links, redirects, sitemap, transport.
| Check | Result | Evidence | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Fails · 5 | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| DS23 | fail | the address a site uses to list its own pages answered, but what came back is not a list of pages, so an AI can only find what it reaches by following links, and anything poorly linked is never found | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
What we check The site publishes a sitemap at the standard address and it returns successfully. What good looks like The site publishes a list of its own pages at the usual address. Its absence is a fail: discovery then rests on links alone. Why it matters Without a list of its own pages, the site can only be explored by following links, so any page that is poorly linked is never found and plays no part in what AI systems know about the organisation. What we found on this page the address a site uses to list its own pages answered, but what came back is not a list of pages, so an AI can only find what it reaches by following links, and anything poorly linked is never found The evidence, taken from the page as served The sitemap, as the site answered for it
The address sitemap.xml was asked for. It is the file a site uses to list its own pages.
Why it reads as a failure The result is set by what was measured on this item, quoted above, and not by an opinion of it. It reads as a failure because the standard was not met: the site publishes a list of its own pages at the usual address. What it does to the figure This check counts toward the readiness figure for this item, at a weight of 3 out of a possible 5, because it is one of the checks that decide how cleanly content can be read once reached, so it carries a middle weight in the figure for this item. Why it moves the misinformation reading This failure could contribute to a wrong answer in some cases, so it raises the misinformation reading for this item. Where an item carries more than one such failure, the readings compound and the item moves up the scale. What this result does not say This says what was found on this item at the time of assessment, and nothing beyond it: it is not a statement about the rest of the estate, and it does not judge the quality of the content itself. | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| DS24 | fail | what the address returned is not a list of pages | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
What we check The sitemap that is served actually parses as valid XML in the standard format. What good looks like The served sitemap parses as valid XML in the standard format. Why it matters A sitemap that cannot be read does nothing, and it fails silently, so the site looks covered when it is not. What we found on this page what the address returned is not a list of pages Why it reads as a failure The result is set by what was measured on this item, quoted above, and not by an opinion of it. It reads as a failure because the standard was not met: the served sitemap parses as valid XML in the standard format. What it does to the figure This check counts toward the readiness figure for this item, at a weight of 1 out of a possible 5, because it is one of the supporting checks, so it carries the lightest weight in the figure for this item. Why it does not move the misinformation reading This failure makes content harder to reach or to read, but it is not a direct route to a wrong answer, so it is held out of the misinformation reading rather than counted twice. What this result does not say This says what was found on this item at the time of assessment, and nothing beyond it: it is not a statement about the rest of the estate, and it does not judge the quality of the content itself. | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| MX14 | fail | soft-404: a missing URL returns 202, so error-page content is served as if it were real content | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
What we check Whether a URL that does not exist returns a real 404, or answers 200 with an error page. What good looks like A missing address returns a real 404, not a success answer wrapping an error page. Why it matters An error page answering as success is stored as real content, and the visitor gets no signal the link was dead. What we found on this page soft-404: a missing URL returns 202, so error-page content is served as if it were real content The evidence, taken from the page as served How this site answers for an address that does not exist
An address that cannot exist was asked for. A site should say plainly that it is not there.
Why it reads as a failure The result is set by what was measured on this item, quoted above, and not by an opinion of it. It reads as a failure because the standard was not met: a missing address returns a real 404, not a success answer wrapping an error page. What it does to the figure This check counts toward the readiness figure for this item, at a weight of 1 out of a possible 5, because it is one of the supporting checks, so it carries the lightest weight in the figure for this item. Why it moves the misinformation reading This failure is a real and common route to a wrong answer for the content it affects, so it raises the misinformation reading for this item. Where an item carries more than one such failure, the readings compound and the item moves up the scale. What this result does not say This says what was found on this item at the time of assessment, and nothing beyond it: it is not a statement about the rest of the estate, and it does not judge the quality of the content itself. | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| DS03 | fail | 1 of 22 AI services are turned away before they reach the site, while the other 21 are served in full, so those without your pages answer about you from other people's material instead of your own. | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
What we check The same page loads for a browser but is refused when the request names itself as an AI service. What good looks like The page returns the same response to a named AI service as to a browser, with no edge block. Why it matters AI systems are being turned away before they reach the site, and because the barrier sits outside the organisation's own systems, nobody inside can see it happening. Those systems carry on answering questions about the organisation, from other people's material rather than its own, so the conversation about it goes on without it and nobody has decided that. What we found on this page 1 of 22 AI services are turned away before they reach the site, while the other 21 are served in full, so those without your pages answer about you from other people's material instead of your own. The evidence, taken from the page as served Each crawler's own request to the entry page, and the status it received
Hostwww.essex.gov.uk · resolves to 18.65.194.20, 18.65.194.118, 18.65.194.86, 18.65.194.116 · Amazon CloudFront
Address trailwww.essex.gov.uk › www.essex-cc.bbdclients.net
Checked2026-09-17 16:02 UTC · request sent from 54.171.52.206 (our checker)
Why it reads as a failure The result is set by what was measured on this item, quoted above, and not by an opinion of it. It reads as a failure because the standard was not met: the page returns the same response to a named AI service as to a browser, with no edge block. What it does to the figure This check counts toward the readiness figure for this item, at a weight of 5 out of a possible 5, because it is one of the checks that decide whether content can be reached and read at all, so it carries the heaviest weight in the figure for this item. Why it moves the misinformation reading This failure could contribute to a wrong answer in some cases, so it raises the misinformation reading for this item. Where an item carries more than one such failure, the readings compound and the item moves up the scale. What this result does not say This says what was found on this item at the time of assessment, and nothing beyond it: it is not a statement about the rest of the estate, and it does not judge the quality of the content itself. | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| AJ04 | fail | none of the three pages a reader checks before trusting a site could be found among the addresses this site publishes: its terms, a privacy notice and a route to make contact. There is nothing to confirm who stands behind the content | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
What we check Whether the site publishes and serves its terms, its privacy notice and a route to make contact or complain. What good looks like Terms, a privacy notice and a contact route are published, and each opens when asked for. Why it matters These are the pages looked for before anything on a site is repeated. Where one is absent or does not open, there is nothing to confirm who stands behind the content. What we found on this page none of the three pages a reader checks before trusting a site could be found among the addresses this site publishes: its terms, a privacy notice and a route to make contact. There is nothing to confirm who stands behind the content The evidence, taken from the page as served The pages a reader checks before trusting a site
The addresses the site publishes were read for each of the three, and each one found was opened.
Why it reads as a failure The result is set by what was measured on this item, quoted above, and not by an opinion of it. It reads as a failure because the standard was not met: terms, a privacy notice and a contact route are published, and each opens when asked for. What it does to the figure This check counts toward the readiness figure for this item, at a weight of 3 out of a possible 5, because it is one of the checks that decide how cleanly content can be read once reached, so it carries a middle weight in the figure for this item. Why it does not move the misinformation reading This failure makes content harder to reach or to read, but it is not a direct route to a wrong answer, so it is held out of the misinformation reading rather than counted twice. What this result does not say This says what was found on this item at the time of assessment, and nothing beyond it: it is not a statement about the rest of the estate, and it does not judge the quality of the content itself. | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| Warnings · 1 | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| MX06 | warning | the page loads for a phone but carries no instruction telling a phone how to lay it out; no mobile layout signal | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
What we check The entry page fetched as a mobile device, compared with the desktop version for substance. What good looks like The mobile page carries the same substance as the desktop version. Why it matters The mobile version is the one that gets stored, so anything stripped for mobile is missing from the record. What we found on this page the page loads for a phone but carries no instruction telling a phone how to lay it out; no mobile layout signal | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Passing checks are omitted here; the full record is in the exports.
No pages could be read on this estate, so none were scored. See the access finding above.
No documents could be reached on this estate.
This report is Stage 1, Foundations: each automated test the engine can run without human input, so there is a usable result at the first stage. Governance (owner-confirmed truth) and Visibility (how AI agents find the estate) are later stages and are shown greyed above.
What was assessed. 0 pages were assessed, selected from 1 reviewed using the site's own signals of importance (its navigation, homepage links, most-linked pages, sitemap, and any llms.txt file), and the first 0 PDF documents found were assessed separately. The industry in the header is classified from the site's own content. The traffic classification comes from a recognised top-sites list or a connected traffic provider; it reads NA when the domain is not connected to a traffic source.
Header identity signals. The regulated sector and listed company values are inferred from the same crawled content, and shown here so the header can stay to the answer alone. Regulated sector reads “Not established”: none. Listed company reads “No”: not a company with publicly traded shares (government-restricted domain).
AI readiness is the weighted share of checks that passed, reported separately for pages and for PDF documents because each is reached and corrected differently. Grade A checks decide whether content can be reached and read at all; grade B whether it is understood correctly; grade C support reliability. A measured score is scaled into a ceiling when an access gate binds it (for example the site blocking AI crawlers at its edge): content that cannot be reached is not partially ready. Scores never read exactly 25, 50, 75 or 100, because those are buckets, not measurements.
Likelihood of misinformation is a band, given as a worst case (the highest any single item or estate-wide finding reaches) and a typical position across the pages or documents. Only checks whose failure is a direct route to a wrong answer carry a band; the rest affect readiness only.
| Band | Range | Meaning |
|---|---|---|
| Likely zero | 0-5% | No failing misinformation-bearing checks. |
| Very little | 6-25% | Unlikely to cause wrong information. |
| Likely | 26-50% | Could contribute to wrong information in some cases. |
| Very likely | 51-75% | A real and common route to wrong information. |
| Extremely likely | 76-100% | A frequent and direct cause of wrong information. |
Of the register's 207 checks, 23 ran automatically in this Foundations stage. 43 are questions for the person accountable for the content, asked at the next stage. 114 are built and found nothing of their kind published on this site, so they had nothing to read. One is defined in the register and not yet built. 7 need information supplied by the organisation. Each check that did not run is listed in the exports, with its reason.
About us
Built on evidence, not assumptions.
We have been doing this for 25 years. For the last two years, we have been training AI to understand and map AI Readiness (fundamentals, governance, visibility), against our unique dataset of more than 3.7 trillion data points.
Who’s talking about us
