Families increasingly ask ChatGPT, Perplexity, or Google’s AI Overviews which care home to shortlist. The assistant names a few providers. Everyone else is not in the room.
So we checked 6,678 UK care homes’ websites. The list came from the public registers of all four regulators: the Care Quality Commission in England, the Care Inspectorate in Scotland, Care Inspectorate Wales, and RQIA in Northern Ireland. We kept only services registered for older people’s care, then scored each website on the five questions a site has to answer to earn that mention.
Where the questions come from
The questions come from the Get Recommended Framework, built and refined over the last two years from how search engines and AI assistants decide what to recommend. The framework came first. We then applied its five website questions, unchanged, to every care home website we could reach. Nothing in the scoring was adjusted after seeing the results.
| Step | The question a family, and an assistant, is asking |
|---|---|
| Buyer clarity | Does the site say, near the top, who this home is for, where it is, and roughly what it costs? |
| Recognized | Does the home appear under its own name with consistent details, so a search engine or assistant can identify it as a real, specific place? |
| Answer-ready | Are the common questions answered in plain sentences that can be quoted, rather than in a brochure PDF or a video? |
| Trust on-page | Is what a family checks actually on the page: inspection rating, staff, real photographs, reviews, registration details? |
| Local decisions | Does the site help someone decide locally: fees, funding, how to arrange a visit, what admission involves, what happens next? |
The framework’s sixth step, knowing where an enquiry came from, is not scored here. It lives in your analytics and your phone system, not on the page, so no outside crawl can see it honestly. Every figure below is out of the five questions above. The full scoring rules set out exactly what earns each mark.
What we found
Three numbers carry the whole study.
| What we found | Share of UK care homes |
|---|---|
| No website we could find, after three separate searches | 1 in 9 (10.9%) |
| A website address that leads nowhere | a further 9.6% |
| Of those with a working site, how much of a family’s questions it answers | about six tenths (62%) |
Then the detail underneath it:
- Roughly 3 in 10 homes with a site do this well, scoring 8 or more out of 10.
- About 1 in 16 do it badly enough that a family would learn almost nothing.
- Only 41% have any page or section about fees or funding.
- Just 4.3%, about one site in 25, show an actual weekly price.
- Only one site in five carries the structured data that tells a machine what kind of organisation this is.
One thing you may have heard is not the problem. Care homes are not being kept out of AI answers by blocking the crawlers. Fewer than 2% of sites do that in robots.txt, and not one large operator does. The door is open. What is missing is anything worth quoting once the assistant walks in.
One care home in five has nothing a family could open
Weighted to the real mix of UK operators, 10.9% of older people’s care homes have no website we could find: none listed with the regulator, and none that came up in three separate searches, by name and town, by name and postcode, and on the operator’s own website where we knew it.
A further 9.6% have a website address that leads nowhere. The domain no longer resolves, the page returns a 404, the site is parked or marked coming soon, or it loads and shows nothing. To a family searching in the evening, the second group is the same as the first. Together they are 20.5% of the sector.
This is a small-operator problem. Among operators with one to five homes, one home in four has nothing a family can open under its own name: 14.5% have no site at all and 11.6% have a dead one. Among groups with more than twenty homes, almost nobody lacks a website, but dead pages still turn up.
| Operator size | Homes checked | No website found | Site is dead | Site we could check |
|---|---|---|---|---|
| Small (1 to 5 homes) | 4,902 | 14.5% | 11.6% | 71.8% |
| Mid (6 to 20) | 1,101 | 6.1% | 5.5% | 82.9% |
| Large (21 to 50) | 264 | 0.8% | 3.4% | 95.1% |
| Very large (51+) | 411 | 0.2% | 6.1% | 93.7% |
| All homes in the sample | 6,678 | 11.7% | 10.0% | 75.9% |
| Weighted to the UK operator mix | 10.9% | 9.6% | 77.3% |
The rows do not add to 100%. The remainder are homes whose site refused our request or timed out (0.9%), and homes whose registered website is a council or NHS page rather than a care home site (1.5%).
Many of these homes are full today. They fill through the council, through a directory listing, through the GP and the church noticeboard. That model has worked for decades. It stops working the day a daughter in Leeds asks an assistant which homes near her mother take people with dementia and accept council funding. The assistant answers from what it can read.
The dead sites deserve a moment, because they are the easiest problem in this report to fix and the most embarrassing to have. Dead pages are now as common as missing sites, and they are the one failure that reaches the largest groups: 6.1% of very large operators’ homes point to a page that no longer exists. Usually it is a home page on the group site taken down when a home closed or changed hands, while the regulator record still points at it. Someone at head office owns that link. Nobody has checked it.
This cuts two ways for you. In most local markets one in five of the homes you compete with cannot be found by the tools families now use, and among small operators it is one in four. That is a head start you did not have to earn. It only counts if your own homes are reachable.
The ones with a site are doing six tenths of the job
Among homes with a working website, the typical site gives a family about six tenths of what they need and leaves the rest out. Not broken. Just incomplete, in ways that send people elsewhere.
| Operator size | Homes scored | Distinct websites | How much of what a family needs |
|---|---|---|---|
| Small (1 to 5 homes) | 3,520 | 2,433 | 60% |
| Mid (6 to 20) | 913 | 232 | 60% |
| Large (21 to 50) | 251 | 28 | 68% |
| Very large (51+) | 385 | 17 | 77% |
| Weighted across the sector | 5,069 | 2,594 | 62% |
Read the third column before the others. Group websites cover many homes, so 385 very large homes sit on 17 websites and 251 large homes on 28. The averages for those two rows describe a few dozen sites, not hundreds, and a single group redesign would move them. The small-operator row is different: 3,520 homes on 2,433 separate websites, which is the real spread of the sector. We treat the gap between small operators and the very largest groups as a finding. We do not read anything into the gap between mid and large, because with 28 websites on one side it is not stable enough to carry a conclusion.
Roughly three homes in ten do this well, scoring 8 or more out of 10. About one in 16 does it badly enough that a family would learn almost nothing. Everyone else, about two thirds, is in a wide, flat middle, and that middle is where you win or lose enquiries without ever knowing it. If you are not sure where your own homes would land, ask an assistant to recommend care homes in your town and see whether you are in the answer. If you are not, it is worth finding out why before your competitors do.
Size helps, but less than the budgets would suggest. A small operator and a mid-size group land in the same place. The very largest national groups do noticeably better than a family running one or two homes, but not dramatically so. That is the whole distance between a national marketing team and an owner who edits the site herself on a Sunday. It should be wider.
Operators of all sizes share the same gap
Across all four sizes of operator, the weakest part of a care home website is the same: helping a family make a practical local decision, meaning fees, funding, how to visit and what admission involves.
| Step | All homes | Small | Mid | Large | Very large |
|---|---|---|---|---|---|
| Buyer clarity | 65% | 70% | 60% | 70% | 80% |
| Recognized | 75% | 75% | 75% | 75% | 75% |
| Answer-ready | 60% | 55% | 60% | 70% | 85% |
| Trust on-page | 70% | 65% | 70% | 80% | 95% |
| Local decisions | 40% | 40% | 40% | 40% | 50% |
Each figure is how much of that question the typical site answers, where 100% means a family gets the full answer without hunting. Look at the bottom row, then read across it. Local decisions is the weakest area in the sector by a wide margin, and the only question where the average site scores under half. The very largest groups, with marketing departments and agency retainers, are only a little better than anyone else on it. Whatever that money buys, it is not buying this.
In plain terms, somewhere between 31% and 43% of care home websites give a family nothing at all to help them decide locally. Before you assume your sites are the exception, open one and check, or have someone who works in AI search check it for you.
You hear this gap on the phone every week. A family wants to know what it costs, whether the council will pay part of it, how to visit this week, and what happens on the day their father moves in. Most sites answer none of that. They show a garden, a smiling resident, and a contact form. The visitor leaves with the questions they arrived with and rings the home down the road. You never knew they were there.
Six in ten say nothing about fees, and one site in 25 shows a price
Four signals were recorded on each site, plus a check for an actual price. Each one is a test you can run on your own sites in under ten minutes.
| What we looked for | All homes | Small | Mid | Large | Very large |
|---|---|---|---|---|---|
| Has some FAQ-style content | 80.7% | 79.4% | 74.9% | 96.4% | 89.6% |
| Has a page or section about fees or funding | 40.9% | 34.3% | 44.7% | 46.2% | 77.1% |
| Shows an actual weekly price | 4.3% | 4.5% | 4.3% | 6.8% | 1.6% |
| Has Organization structured data | 20.5% | 15.6% | 19.5% | 35.1% | 47.5% |
| Blocks at least one AI crawler in robots.txt | 1.5% | 1.4% | 3.3% | 0.0% | 0.0% |
Four sites in five already have something question-shaped on them, so the raw material is usually there. Only 41% have any page or section about fees or funding, and most of those talk about funding routes without giving a number. Just 4.3%, about one in 25, show a weekly pound figure anywhere in the pages we read.
Fees are the first question a family asks. Whether publishing a number wins or loses enquiries is not something this study measured, and operators have reasons for caution, since fees depend on assessed need. But a family who cannot find even a starting figure has to phone to get one, and every extra step loses some of them.
Separately, only one site in five carries the structured data that tells a machine what kind of organisation this is and where it operates. That is a few lines of code your developer can add in an afternoon.
The last row deserves a word, because a lot is written about websites blocking AI crawlers. In care homes it barely happens. Fewer than 2% of sites deny access to any AI crawler in robots.txt, and not one large or very large operator does. Most of the handful that do are small sites blocking every crawler, Google included, which usually means a setting left over from when the site was built. So if your homes are not showing up in AI answers, the crawlers are not being turned away at the door. They are getting in and finding nothing they can use.
What to do about it
Most care home websites look presentable and answer about six tenths of what a family needs. The gap is between looking presentable and being useful to someone who has to decide this month. You can close it faster than your competitors expect, because the missing pieces are cheap and almost nobody has bothered.
Fix the cheap one first. Helping families decide locally is the weakest area in the sector and the least expensive to repair. A fees and funding page is writing, not development. A clear admissions and visiting page is writing. Do both and a typical home goes from the crowded middle to the top quarter of the sector on this measure, ahead of most of the national brands. That is a statement about the website, not about occupancy or care, but it is the website that decides whether a family gets as far as phoning.
Then tag your contact routes. We did not score this, because it cannot be seen from outside. It is worth an afternoon anyway. If you cannot tell whether an enquiry came from Google, a directory, a referral, or an AI assistant, you cannot tell what is working, so next year’s budget goes where last year’s went and the directory that stopped delivering two years ago keeps getting paid.
Do not assume the big groups have this covered. They are ahead, but not by much, and they are winning on polish and trust signals while carrying the same gap as everyone else. A well-run operator of five homes who answers the practical questions in plain English will beat most national chains on the measures that decide whether an assistant recommends them. Scale is not the moat it looks like.
Check whether you can be read at all. Does each home in your portfolio have a live site that loads? Does the address the regulator holds for it still work? Does each home appear under its own name, with the same address and phone number everywhere it is listed? Can someone who has never heard of you find what it costs and how to visit without phoning? One home in five fails the first test. None of this is SEO. It sits underneath SEO.
A ten-minute check on where you stand
Open one of your own sites as if you were a daughter looking for a place for her father. Can you find what it costs, whether council funding is accepted, and how to arrange a visit, without phoning? Then ask an AI assistant to recommend care homes in that town and see whether your home is mentioned. If either test fails, you have found the gap this report is about. Closing it is not complicated, but it does need someone who understands how search engines and assistants decide what to recommend, which is a different skill from building an attractive website. If your agency cannot explain why you are not appearing, find someone who can.
How the study was done
Where the list came from. The public registers of the four UK regulators, filtered to services registered for older people’s care. That left 13,603 homes: 11,723 in England, 709 each in Scotland and Wales, and 462 in Northern Ireland.
The sample. A sample full of big brands would flatter the sector, because big brands are easy to find and easy to crawl. So we set quotas to match the real mix of operator sizes in the register. 6,678 homes were drawn. 5,069 were crawled and scored as care home websites. 780 have no website we could find. 665 have a website address that is dead. 62 refused our request or timed out. 102 have a registered website that is a council or NHS page rather than a care home site; those were scored for completeness but kept out of the website statistics, because a council directory page is not the home’s own website and should not be judged as one.
The re-check, and what we got wrong the first time. The first version of this study, published on 19 September, reported that 25.5% of homes had no website. That figure was wrong, by a lot, and we are correcting it here rather than quietly. It was wrong for two reasons, both ours.
First, 1,195 of the homes marked as having no website had never actually been searched for one; a step in the resolution process had skipped them. Second, the rule that decided whether a search result was the home’s own website was too strict in two ways. It rejected pages that sit inside a group website, so a Bupa or Care UK home whose page lives at bupa.co.uk/care-homes/… was thrown away as if it did not exist. And a filter meant to remove the directory carehome.co.uk also removed any operator whose own domain happened to contain the word, so grangehallcarehome.com and hundreds like it were rejected too. Both faults were in the original September pass as well, which is why the first figure was so far out.
On 24 September we searched again for all 1,388 homes then marked as having no website: by name and town, by name and postcode, and, where another home of the same provider had already given us the provider’s website, directly on that site. Every raw result was kept so the matching could be re-run and checked without new searches. Directory sites, local newspapers and property listings were removed by hand from the candidates. 608 websites were found, crawled and scored, and are included in every figure above. Roughly 40% of them are pages on a group website, the rest single-home domains.
Then we checked what was left. A random 100 of the 780 homes still marked as having no website were reviewed by hand against their saved search results. Five had a website we could see but could not attribute automatically, for example a charity’s services page or an operator brochure in PDF form, and six more were possible matches under a different name. So the true share of homes with no findable website is a little below the 10.9% we report, most likely between 9.5% and 10.5%. We report the measured figure and this margin rather than adjusting the number.
On the same day we went back to all 478 homes whose site had failed to load in the first crawl, and to every scored page that looked like a parked domain, an error page or an empty shell. Sites that only load with JavaScript were rendered in a browser this time. Most of the failures stayed failures: for 256 of the addresses we re-fetched, the domain no longer resolves at all. 113 pages that had been scored near the bottom of the scale moved into the dead-site category, where they belong.
The net effect. No website found fell from 25.5% to 10.9%, the dead-site share rose from 7.2% to 9.6%, and the average mark for a working site is 6.2 out of 10. The 19 September version reported 5.3 on a broader scale that has since been narrowed to the five website questions used here. On those five questions the September figure would have been 5.8, so the like-for-like movement is small, and it comes partly from pages that were never real websites no longer dragging the average down and partly from the newly found group pages, which score above the small-operator average. The local-decisions gap, the fees finding and the crawler-blocking finding did not change. The finding that did change is who lacks a website: it is now clearly a small-operator problem, not a sector-wide one.
Who decided the five questions. They come from the Get Recommended Framework, developed by the author over the last two years from how search engines and AI assistants select what to recommend: they need to identify a specific place, find plain-text answers they can quote, see evidence of trust, and be able to tell a searcher what to do next. The framework was fixed before the crawl began and nothing in it was tuned to the results. It was not derived from interviews with families, social workers, or commissioners, and it does not try to cover everything a family asks. Staffing levels, dementia expertise, food, activities, room availability, end-of-life care, and cultural needs all matter to families and are not scored here.
How the marks work. Each site scored 0, 1, or 2 on each of the five questions. A 0 means the answer is missing. A 1 means it is there if you dig. A 2 means it is obvious. That gives a mark out of 10, which every figure in this report expresses as a percentage of the full mark, so 5 out of 10 shows as 50%. “Does this well” means 8 or more; “badly” means 2 or less. The full scoring rules are published alongside this study so anyone can check a site by hand or re-run the scoring.
Why tracking is not scored. The framework’s sixth step, knowing where an enquiry came from, is not part of any figure here. It depends on analytics, tagged forms and call routing, none of which an outside crawl can see without guessing. An earlier version of this study reported a score for it. That was wrong, the figure has been withdrawn, and the marks above are out of the five questions that can be judged from the page.
Weighting. Small operators drop out of a crawl more often than large ones, so each size group is weighted back to its true share of the register. Small operators are 67.2% of UK care homes, so they carry 67.2% of the headline number.
No website, and a dead one, are different categories. A home with no website cannot be scored, and we did not give those homes a zero. A zero means a website that fails. A dead site is a third category: an address exists in the register or was found by search, but it no longer leads to a working page. The fix is different. A home with no site needs one built. A home with a dead site needs someone to notice.
The unit that varies is the website, not the home. The 5,069 scored homes sit on 2,594 distinct websites: 2,433 for small operators, 232 for mid-size groups, 28 for large groups and 17 for very large groups. Where many homes share one site, they share one score. That is why we compare small operators with the very largest groups and say little about the two groups in between.
Crawling and scoring were kept separate. Each page was fetched once and stored, then scored from storage, so the whole study can be re-scored under revised rules without touching a single website again. For each home we fetched the homepage plus up to four further pages the site pointed to. Sites that returned an empty page to a plain request were rendered in a headless browser before being judged. Scoring is fully deterministic. Scorer version 1.2.
Five pages per home is a limit, and we measured its cost. Of the homes that got no credit for helping a family decide locally, 26% had a link on their homepage to a page whose address mentions fees, funding, admissions or pricing that we did not open. Some of those pages would have earned credit. So the true share of sites that give a family nothing to decide with is somewhere between 31% and 43%. We report the range rather than pick a point inside it.
What the fees signal counts. Any page or section about fees, funding, self-funding, local authority funding, or paying for care. It does not distinguish a starting price from a fee range, a funding explanation, or nursing versus residential rates. The separate check for a weekly figure looks for a pound amount within a few words of the word week in the pages we read.
What the crawler check counts. Each site’s robots.txt is read and checked for whether any of these named crawlers is denied the whole site: GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, anthropic-ai, PerplexityBot, Google-Extended, CCBot, and Bytespider. A site that blocks every crawler with a wildcard rule also counts. Blocking a crawler does not by itself remove a home from AI answers, since assistants also draw on search indexes and directories, but it does stop them quoting the home’s own pages.
An earlier correction. A version of this study reported that a quarter of large operators block AI crawlers. That was wrong. The robots.txt parser in scorer version 1.1 misread files with no blank lines between sections, and flagged sites that only blocked SEO tools such as AhrefsBot. Version 1.2 corrects it. The right figure is under 2% overall and 0% for large and very large operators. This note stays here, alongside the note on the no-website figure above, rather than the numbers being quietly changed.
Verification. Each scored home was checked for whether its own name appears on the pages we read. 93.2% pass. Of the 343 that fail, roughly half sit on a group website listing many homes, where the site is the right operator but the home’s own page was not among the five we read. The rest are a mix of rebrands, homes that changed hands since the register was updated, and some wrong matches. Dropping all 343 moves the headline figure by 0.6 of a percentage point, so they do not drive any result here. The sites found in the re-checks were verified by looking for the distinctive words of the home’s name in the page text or the domain, which is a looser test than the one used for the original set.
How the sample sits against the register. Quotas were set by operator size only. It still matches the register closely by nation: England 85.8% of the sample against 86.2% of the register, Scotland 5.3% against 5.2%, Wales 5.6% against 5.2%, Northern Ireland 3.3% against 3.4%. It was not stratified by region, urban or rural setting, or nursing versus residential provision.
Honest limits
- Operator size comes from regulator records, which lag reality. Homes that changed hands can still show a previous owner, and groups sometimes register under several company names.
- No website means no site in the regulator record and none found by three searches: name and town, name and postcode, and the provider’s own website where known. A home with a site that is listed nowhere and does not rank for its own name is counted as having none. From a family’s point of view that is the right call. It is not strictly the same claim.
- We read web pages, not documents. A fees brochure published as a PDF, which is common, was not opened, so a home that publishes its prices only in a PDF gets no credit for the price. The one-in-25 figure is a statement about page text and will understate the number of homes that publish a price somewhere.
- The scoring rules look for English words. Welsh-language pages, and bilingual pages where the Welsh version is what we happened to read, will score lower than their content deserves. Wales is 5.6% of the sample, so the effect on the headline figures is small. It could be material for an individual Welsh home.
- The crawler check reads robots.txt only. A site can also turn crawlers away at the network edge, through a firewall or a CDN, and we did not test that. The right reading of the 1.5% is that almost no care home site asks AI crawlers to stay away.
- There is a survivorship effect in the scored group. Sites that fail to load, or turn out to be dead, are more often small operators, so the 5,069 scored homes are a slightly healthier set than the sector as a whole.
- This study did not test AI assistants directly. What the assistants actually say when a family asks them for care homes in a given town is the subject of a subsequent study, which will put those answers alongside the website results published here.
- It says nothing about the care inside. A home with an outstanding inspection and a poor website is a good home with a poor website, and nothing here suggests otherwise.
What to do next
If you want a clear view of whether your homes would clear these bars, and which gap to fix first:
- Book a short diagnostic call
- Are you looking for help with your marketing?
- Get the free six-step workbook for care homes
Findable still helps. Recommended is the job now. Most UK care home websites we checked are still easier to browse than to recommend.