Corrections

What we got wrong, and what it says now.

When a sentence on one of our pages turns out to be wrong, we fix the page and list the fix here: what it said, what it says now, and why.

So far: 186 corrections on 20 pages, the newest made on October 12, 2026.

How this list works

It covers the words on our pages since October 4, 2026: a sentence that was false, or that another of our pages contradicted. Clearer words for something already true are not listed.

183 of the 186 so far were made on the list's first two days, 93 on October 4 and 90 on October 5, during a review of every page.

Fixes to our checking tool or the census data are not listed, nor anything corrected before that day. One earlier correction has its own page: why we took our headline number down.

Each date is the day we made the change, in Eastern time; it reached the site with our next update. An entry goes live with its fix or after it, never before.

What a page says now is its wording today; rewording it later for clarity keeps the entry's date.

Words in square brackets in what a page says now are filled in automatically each time we update the site: a figure from our records, or the words that fit it.

No entry names a business or other group whose website we measured. The same list, as one file: corrections.json.

If you find something we got wrong, write to us and say what and where.

October 12, 2026
It said	It says now	Why
The proof: how the Inspector is tested
planted on purpose in a built page or the files that build it	planted on purpose in a built page or our own files	Some of the planted mistakes sit in our own tool files, which do not build the site, so the old words described only part of where they go.
Insider: Nobody has a census of business websites. We are building one.
Both files include the exact search we send OpenStreetMap	Both files include the search template we sent OpenStreetMap	The files carry the search with the map area left blank, a template rather than the exact search, and the census no longer sends it: it stopped asking the public map server it used.

October 10, 2026
It said	It says now	Why
Insider: Nobody has a census of business websites. We are building one.
the page is parked, hijacked or a placeholder, or the certificate failed our check	A security certificate is the proof of identity a browser checks; our check refusing one does not show the site is gone.	A certificate our check refused does not show the site is gone, broken or taken over. Those addresses are now counted as not settled yet.

October 5, 2026
It said	It says now	Why
The home page
We are counting the websites of working businesses.	We are counting the websites of businesses.	A website that answers does not show that the business behind it is still open, so the word claimed more than we know.
Nobody has a full count of them.	Nobody publishes a full count of them.	We cannot check what everybody has. We can check that nobody publishes such a count.
Our crawler reads each website, measures the same things on every one	Our crawler reads each website the same way, adding a few checks for some kinds of business and for Florida	Some checks run only for some kinds of business, such as a license number for a trade, and some only for Florida sites, so not every site is measured on the same things.
a server	at least between the starts of two visits to the same website	The wait is kept for each website address, not for each server, and many websites can share one server.
It said nothing about this.	requests at most, plus one each time the site forwards us to another address.	Each time a site forwards us to another address, we make one more request, and the line did not count them.
The free inspection
About ten seconds · No email needed	A few seconds · No email needed	Ten seconds overstated how long the machine takes to read a page.
Time to run 10 sec	Time to run a few sec	Ten seconds overstated how long the machine takes to read a page.
We run the checks we run on our own sites.	We run the same checks our census runs.	The words read as if we build websites for clients, and we do not. The free inspection runs the census's own checks.
Reply once and we stop emailing you.	We send one email, your report, and nothing after it.	The form at the top sends one email only, so there was never a stream of email to stop.
It said nothing about this.	Your address and the report go from your browser to Web3Forms, a form service that passes them to us.	The page did not say that an outside form service carries the address and the report to us.
A good result ends with nothing on the list.	A good result has no serious problem to fix.	A site that measures well can still show small findings; what it does not show is a serious one.
one reply gets you off the list	Taken out.	There is no list to get off. The form sends one email, and more only if you tick the follow-up box.
Either way we keep one anonymous line: trade, area, score	It holds the date, the kind of business, the town your site lists, the score and grade, and which checks failed or came back with a warning.	The line we keep also holds the date, the grade and the checks that failed or came back with a warning, which the note left out. Our privacy page already listed them.
never your address unless you tick the box.	It never holds your website address. Tick the box above and we keep the address in a separate record, with that result beside it.	The note read as if ticking the box put your website address into the anonymous line. It never does: the box keeps the address in a separate record. It is your website address the box keeps, not your email address.
The Benchmark
Where 88 real websites landed.	Where [number] graded pages landed. Or, depending on the figures: Where [number] real websites landed.	The count included our reference pages, which are not census businesses. The heading now says pages until the runs count the businesses alone.
sites graded; 14 could not be measured, not guessed at	pages graded, our [two] reference pages among them; [number] could not be measured, not guessed at Or, depending on the figures: websites graded; [number] could not be measured, not guessed at	The count included our reference pages, which are not census businesses. The page now says so, and runs from now on count the businesses alone.
one the industry agrees is excellent, one deliberately near-empty	One is a large company's carefully built home page.	No industry agreement stands behind that page. We chose it, and the page now says plainly what it is.
These are census businesses, picked by a fixed rule and never by score. Each was graded on three different days before	Of these, [number] are census businesses, picked by a fixed rule and never by score; the other [two] are our reference pages. Or, depending on the figures: These are census businesses, picked by a fixed rule and never by score.	The latest run still counted our reference pages among the graded pages, so not all of them were businesses, and the rule that picked the businesses was never shown.
The ten most common failures	Checks asked of only some pages are left out, since a share of every page would understate them.	Some checks are asked of only some pages, so a share of every graded page understated how often they fail. Those checks are no longer ranked here.
The Three Rings
We draw this circle around your address, not ours.	We draw this circle around the center of your town, not around ours.	The circle is placed at the center of a town, never at a street address.
We use the town a firm prints on its own page. Town centers are approximations	We use the town a firm prints on its own page, or the town on its map listing when the page gives none we can read.	When a page prints no town we can read, the ring uses the town on the firm's map listing instead.
If we cannot read a firm's own town off its own website, it gets no local ring.	If we cannot read a firm's town off its own website or its map listing, it gets no local ring.	A firm whose page prints no town still gets a local ring when its map listing names one.
We never ask for or store anyone's location.	We never ask a visitor where they are, and a ring uses only a firm's town.	We do keep the town of each business we read, so saying we store no location was not true.
A percentile drawn from eleven firms moves nine points when a twelfth turns up.	A percentile drawn from eleven firms can move [number] points when one more firm turns up.	Worked out with the formula the rings use, the largest move is smaller than the figure we had typed. It is now worked out when the page is built. A percentile is the share of the group that scored below a firm.
The proof: how the Inspector is tested
The engine has read the same 102 websites	The engine has read the same [number] pages, our [two] reference pages among them, Or, depending on the figures: The engine has read the same [number] websites	The count included our reference pages, which are not census businesses. The page now says so until the runs count the businesses alone.
on your site in about ten seconds	on your site in a few seconds	Ten seconds overstated how long the machine takes to read a page.
There are 44 cases across those 18 pages, because one saved page can be asked more than one question.	Our test suite runs [number] cases on those [number] saved pages and on test pages written into the suite itself.	The count missed a whole folder of saved pages and the cases the suite runs in loops, and some cases run on test pages written into the suite, not on a saved page.
The engine reads all 18 before a change can ship, and a different answer on any of them stops the change.	A wrong answer on any case is a reason to stop, though no automatic step blocks the change.	The saved pages are more than the figure we printed, and the suite is not one of the automatic checks a deploy must pass, so nothing stopped a change on its own. We run it when the engine changes.
If it had warnings, you are reading them.	Any warnings the verdict counts are prompts to look again, not failures; their lines are not shown here.	The verdict counted warnings, and the block showed none of their lines. It now says the warning lines are left out.
not tidied up	Shown are [number] of its passing lines, shortened where they ran long, and its verdict.	The block shows a few passing lines picked from the run and cut short, which is tidying. It now says what is shown.
And nothing about the weights may leave the engine in the report.	And the partial credit a page earned on each check stays inside the engine, while every check in the report must carry its published weight.	The session checks the opposite for the weights: a report that leaves out any check's published weight fails. Only the partial credit a page earned stays inside the engine.
add up the weight of the ones it earned, divide	Each check that applied earns a share of its weight: all of it for a pass, part of it for a warning, none for a fail.	A warning earns part of a check's weight, so adding up the full weights of the checks a page passed would not give its score.
Our own pages get 36 rules before every deploy	On [the day of that run], that run held [number] rule sections	That count was taken from one run weeks earlier, and the program now runs more rule sections than that before a deploy. The count is now dated to the run it came from.
The Inspector page, for server owners
one visit	We wait that long between the starts of two visits to your website, up to [number] seconds.	The wait is kept for each website address, not for each server, and it runs from the start of one visit to the start of the next.
It said nothing about this.	At an address your page forwards us to, we read its robots.txt for the report, not for permission.	We read robots.txt where a page forwards us, to report on it, and the page did not say so.
What the check does not measure
We fetch your page the way an ordinary visitor does, without logging in.	We read only public pages, the ones anyone can open without signing in.	A visitor's browser runs a page's scripts and never reads robots.txt. Our check reads robots.txt and runs no scripts.
Anyone who claims to audit those from outside your accounts has guessed.	Reviews are public, but reading them is a different job, and this tool does not do it.	Anyone can read reviews from outside, so saying only a guess could judge them was not true. Ad and analytics results do live inside your own accounts.
Privacy
Every figure is a count.	Every figure sums up many sites.	A middle score is not a count.
The crawler reads robots.txt first and obeys it.	The crawler first asks the robots.txt at the address it was given, and obeys it.	We obey robots.txt only at the address we were given. Where that site forwards us, we read the new address's robots.txt for the report, not as permission. Changed again on October 5, 2026.
Study copies are kept privately, are never published, and are deleted when the study ends.	Copies from one finished study are still on our computer and are waiting to be deleted.	One finished study's copies have not been deleted yet, so the promise was not kept. The page now says so. The same words were also on 1 more page, each fixed the same way.
We do not need your name or email, and we do not ask for them.	We never ask your name, and the email boxes below are optional.	The inspection page has optional boxes that ask for an email address, so saying we never ask for one was not true. Changed again on October 5, 2026.
Every request goes to truemeasurewebdesign.com and nowhere else.	The one exception on the whole site is asking for your report by email, covered below.	Asking for your report by email sends it from your browser straight to a form service, not to our own server.
Use one and we get your email address and the address you inspected.	It holds your email address and the report: the address you inspected, the score, grade and check counts, the verdict and the failed checks.	The message carries more than the page listed, including the score, the failed checks and the choices made on the form.
which passes it along and does not sell it.	We cannot speak for what it keeps.	The form service is a separate company and we had not read its own policy, so we could not promise what it does with your address.
the way any visitor would	we read the page as its server sends it	A visitor's browser runs the page's scripts, and our check does not.
short-term cache	To stop automated programs flooding the inspection, our web host briefly notes your internet address when you start one, so that address cannot start another at once.	Scripts and short-term cache are insider words, so the sentence now says the same thing in plain words.
first asks	The crawler's first request to a site asks for the robots.txt at the address it was given, and obeys it.	Our do-not-measure list is checked before robots.txt, so robots.txt is the crawler's first request, not the first thing it checks.
boxes below	We never ask your name, and the email boxes on the inspection page are optional.	The email boxes are on the inspection page, not further down this page.
Terms
We fetch public pages the way an ordinary visitor does	We read only public pages, the ones anyone can open without signing in.	A visitor's browser runs a page's scripts and never reads robots.txt. Our check reads robots.txt and runs no scripts.
and about thirty other checks.	and more, from a list of [number] checks.	We typed that figure, and the engine has more checks than it said. The count is now read from the engine each time the page is built.
The Insider, the list of articles
Everything here comes from inspecting real business websites	Most of what is here comes from inspecting real business websites	One guide uses a made-up example, and two are about outside studies, not our inspections.
engine's own	Five small checks, how many measured websites fail each one, and the note our inspection program prints on what to change.	Engine is an insider word. The card now says whose note it is: the one our inspection program prints.
grade split	Charts of how many sites in our count got each grade, and how their scores spread out.	The card used insider words and ended on a word that read like a typo.
Insider: We took our headline number down. Here is why.
Your own report takes ten seconds.	Your own report takes a few seconds.	Ten seconds overstated how long the machine takes to read a page.
Insider: What Google grades your website on
That is a 2x difference from copy complexity alone	So the simpler pages converted about twice as often.	The study compared different pages, so it cannot show the writing alone caused the gap.
One converts at twice the rate of the other.	pages written like the second one converted about twice as often as pages written like the first.	Nobody tested those two sentences. The study measured whole pages written at those reading levels.
(n=1,002 US consumers, MEASURED)	(a survey of 1,002 US consumers)	The figure is what people said in a survey, not something anyone measured. The word MEASURED was one of our own working labels, and it said the opposite of the sentence after it.
Insider: Nobody has a census of business websites. We are building one.
Before it comes back to the same server it waits at least 2 seconds	A new visit to the same server starts at least [number] [seconds] after the last one began	The wait is measured from the start of one visit to the start of the next. Changed again on October 5, 2026.
Ten seconds tells you where you stand.	A few seconds tells you where you stand.	Ten seconds overstated how long the machine takes to read a page. The same line closes every Insider article. The same words were also on 9 more pages, each fixed the same way.
A copy of a front page is kept only for a named study, privately, and deleted when the study ends.	Copies from one finished study are still on our computer and are waiting to be deleted.	One finished study's copies have not been deleted yet, so the promise was not kept. The page now says so.
One inspection makes at most 9 requests to a site, counted from the code	One inspection makes at most [number] requests to a site, plus one more each time the site forwards us to another address.	When a site forwards us to another address, each forward is one more request, and the line did not count them.
same server	A new visit to the same website starts at least [number] [seconds] after the last one began	The wait is kept for each website address, not for each server, and many websites can share one server.
Insider: Where your score lands among the websites we measured
The free inspection runs in about ten seconds	The free inspection runs in a few seconds	Ten seconds overstated how long the machine takes to read a page.
Insider: How to read your inspection report without a web person
The free inspection takes about ten seconds.	The free inspection takes a few seconds.	Ten seconds overstated how long the machine takes to read a page.
It comes in three shapes.	It has one of [four] headlines	The engine has four summary headlines, not three.
weighted by	Some checks count more than others, by our own judgment.	Weighted is a statistics word, so the sentence now says the same thing plainly.
heaviest check	A failed check of the heaviest kind costs [number] times as much as one of the lightest.	The sentence now says plainly how many times as much a failed check of the heaviest kind costs.
findings means	"Several serious findings" means [three] or more of the heaviest checks failed on the page we read	The headline is now set in quotation marks, and how many failed checks it takes is read from the engine.
Insider: Three questions that predict your inspection report
The free inspection takes ten seconds	The free inspection takes a few seconds	Ten seconds overstated how long the machine takes to read a page.
about thirty more things depending on your type of business	In all it has [number] checks, and some run only on some sites	Most of those checks run on every site, not only for some kinds of business, and the count is now read from the engine.
whether your contact form is reachable	whether a visitor can find a way to get in touch	The inspection looks for a way to get in touch, such as a form, an email link or a phone link. It never tests that a form works.
Insider: Five small findings, and how long each one takes to fix
against your live pages in about ten seconds.	against your live pages in a few seconds.	Ten seconds overstated how long the machine takes to read a page. Changed again on October 5, 2026.
None of these need a web designer.	None of these needs a redesign. Whoever can edit your site can make each one.	Several of these fixes mean editing the page's code, which only whoever runs the site can do. The page itself says to hand the notes over.
your live pages	against the page you give it, in a few seconds.	The inspection reads only the page you give it, not all your pages.
Insider: A chamber of commerce website, scored
Screen readers and Google both use the headline to understand what a page is about.	People using screen readers jump between headings, and the top one tells them where they are.	Google says the number of headings on a page does not matter to it. Our own engine stopped saying otherwise in September.
Google uses the headline the same way. It is one of the signals Google uses to understand what the page covers.	Google says the number of headings does not matter to it.	Google says the number of headings on a page does not matter to it.
Every page on the web is supposed to have exactly one.	A page should have one clear top headline for its readers.	No rule says every page must have exactly one. One clear top headline is for readers, not for Google.
does not tell Google or a screen reader	does not tell a screen reader user what the	Google says the number of headings does not matter to it, so only the screen reader half was true.
This is a 10-minute fix.	The engine's note calls it a quick template fix.	We typed that time. The engine's own note gives none.
The file is a 5-minute addition.	The engine's note says the file takes minutes to add.	We typed that time. The engine's own note says only minutes.
They are configuration settings that take under an hour to enable.	The engine's note says they take under an hour for whoever runs the hosting.	The hour is for whoever runs the hosting, and on some plans only the hosting company can make the change.
The missing headline is 10 minutes. The robots file is 5 minutes.	The engine's notes call the missing headline a quick template fix and the robots file minutes of work.	We typed those times. The engine's own notes give no exact minutes.
accounts for 67 percent of total load time on average	On a typical visit, the work done inside the browser, which includes running scripts, took about half the load time	That share was for the slower visits, not the average. On a typical visit the study found about half.
Scripts load from four outside companies	Scripts load from four outside web addresses run by three organizations	Two of the addresses belong to one company. The engine counts web addresses, not companies.
four separate organizations have code running	three outside organizations have code running	Two of the addresses belong to one company. The engine counts web addresses, not companies.
A homepage without a main headline was reviewed and approved at some point. Nobody noticed, or nobody flagged it.	Taken out.	We never saw how the site was reviewed or approved, so we could not say.
That is a review process that did not include a check of this kind.	A check like this one, run before a page goes live, catches it.	We never saw the review process, so we could not say what it included.
national sites	our benchmark list at the time held [number] pages, our [two] reference pages among them, and [number] of them gave a grade.	That list carried our reference pages, which are not national sites, and its figures and middle score counted them.
Insider: Your contact form might be lying to you
But in about ten seconds it will tell you	But in a few seconds it will tell you	Ten seconds overstated how long the machine takes to read a page.
A site with no contact details is at least honest about it. It offers nothing, and the visitor can see that.	A site with no link to reach anyone is at least plain about it. [share] of every hundred of these sites show what looks like a phone number as plain text	About half of these sites print what looks like a phone number, which the visitor can copy by hand.

October 4, 2026
It said	It says now	Why
The home page
The same check reads higher on the benchmark list, which is a different and much smaller set of sites.	On the benchmark list, a different and much smaller group of [number] graded sites, the share without one is [about the same, higher or lower]: [share]. Or, depending on the figures: On the benchmark list, a different and much smaller group of [number] graded pages, our [two] reference pages among them, the share without one is [about the same, higher or lower]: [share].	We typed that word in September and it stopped being true. The comparison is now worked out from both figures each time the page is built.
2 sites are on our do-not-measure list.	[number] [request or requests] so far to put a site on our do-not-measure list.	The sites on the list were ones we added ourselves. No site owner had asked. The line now counts requests.
at least before it visits the same server again	at least from the start of one visit to a server to the start of the next	The wait is between visits, measured from the start of one to the start of the next, and one visit is several requests sent together. Changed again on October 5, 2026.
States with a site counted	States and DC with a site counted	The count includes the District of Columbia, which is not a state.
If robots.txt tells every crawler, or ours by name, to stay out, it stays out.	If the robots.txt at the address we were given tells every crawler, or ours by name, to stay out, it stays out.	We obey robots.txt only at the address we were given. Where that site forwards us, we read the new address's robots.txt for the report, not as permission.
The free inspection
First it asks your robots.txt whether we are welcome.	First it checks our do-not-measure list, then reads your robots.txt, the file where a site says which automated readers are welcome.	The do-not-measure list is checked first, before robots.txt is asked.
Then it reads what kind of business you are, so the checks fit	Then it tries to tell what kind of business you are, so the checks fit.	It cannot always tell what kind of business a site is, and the line said it always could.
Chains are set aside here and counted on the corpus page instead.	The [number] [chain and franchise websites are] left out here; the Three Rings page counts [them].	The page it pointed to did not show how many chains there were.
The Benchmark
and together they match the census mix	they matched the census's mix of trades, in Florida and elsewhere. The census has grown since.	That was true on the day the list was drawn. The census has grown since, and its mix has moved.
The Three Rings
Another 32,243 websites in 51 states are on the list and not yet read.	Another [number] addresses on our list are not counted above. Of those,	Every one of them had been tried. The sentences now say what happened to them.
Ranked, with a percentile	Enough sites for a percentile	Nothing on this site ranks anyone yet.
trades deep enough to rank.	trades with enough sites to compare.	Nothing on this site ranks anyone yet.
Then a firm can be ranked against its own trade	That is what will let a firm be compared with its own trade, not with some general standard nobody set.	Nothing on this site ranks anyone yet.
states, which is how a ring reaches past Florida	states, plus the District of Columbia, which is how a ring reaches past Florida	The count includes the District of Columbia, which is not a state.
Every website sits in all three rings at once.	Every website we can place on a map sits in all three rings at once.	A site we cannot place on a map gets no local ring, as the same page said further down.
The Inspector page, for server owners
which most crawlers do not, including the largest one.	Taken out.	We could not show that comparison, so we took it out.
We wait that long before we return to your server	We wait that long between one visit to your server and the next	The wait is between visits, and one visit is several requests sent together. Changed again on October 5, 2026.
It said nothing about this.	plus any redirects your server sends us through.	Redirects add requests, and the old line did not count them.
It said nothing about this.	We do not check narrower rules, such as one folder, against the small files we fetch with the page.	The page did not say where our reading of robots.txt stops.
What the check does not measure
Every check on this list is something the tool cannot see	Every item on this list is a limit: something the tool cannot see, even where it runs a check close by.	Each item names a check the tool does run, then where that check stops.
The check fetches HTML the way a search engine does.	The check fetches the page's HTML, the code the server sends, and does not run its scripts.	Search engines run a page's scripts, and our check does not.
a phone number, a form, an email link.	a form, an email or phone link, or a link to a contact page.	The list did not match what the check looks for.
Privacy
The only information we hold about you is what you typed on the inspection page and sent us on purpose.	To stop scripts flooding the inspection, our host's short-term cache notes your internet address when you start one, so that address cannot start another at once.	Our host also notes your internet address for a few seconds, and the page did not say so. Changed again on October 5, 2026.
We ask its robots.txt whether we are welcome.	Unless the site is on our do-not-measure list (sites we never start a request to), we read its robots.txt	The do-not-measure list is checked first, before robots.txt is asked.
the names of the checks that passed and failed.	the names of the checks that failed or came back with a warning.	The line keeps failed and warned checks, not passed ones.
a roofing site in Lee County scored 74	a roofing site in Naples, FL scored 74	The line keeps the town a site lists for itself, not a county.
It said nothing about this.	It is anonymous, but not impossible to trace: a trade, a town and a score together could point to the only roofer in a small town.	The page said there was no way back to the person who ran the inspection, and did not say the site itself could be traced.
Terms
The census, and whether your business appears in it	The census, and how your business appears in it	The heading promised an answer the section never gave. It now gives one.
Every term below is one this site already states in plain view.	Every term below, except the law that applies, is one this site already states in plain view.	The law that applies is stated only on this page.
readable text size, and a working contact path.	That covers things like server response time, page size, whether the page tells phones how to fit it, and whether there is a way to get in touch.	No check measures text size, and the contact check finds a way to get in touch without testing that it works.
The published figures are anonymous: a trade, an area, a score.	The published figures are totals and summaries, such as how many sites are in a trade, in a state, or at each grade, and the middle score.	No published figure gives one business's trade, area and score. Every figure sums up many sites.
Whoever inspects that address next sees its last reading beside the new one. It is never published.	We never list or publish it, but anyone who inspects that same address later sees its last reading beside the new one.	Anyone who inspects the same address later sees that reading, not only the next person. The old words could be read as saying no one else sees it.
the census and the free inspection both obey robots.txt before they read anything.	The census and the free inspection both obey the robots.txt at the address they were given before they read anything else.	We obey robots.txt only at the address we were given. Where that site forwards us, we read the new address's robots.txt for the report, not as permission.
No business name, domain, or address is published anywhere.	No published figure names a business, a domain, or an address.	One of our articles names the organization whose website it read, so that sentence was too broad.
The Insider, the list of articles
The five findings most business websites fail, and what each one means	Five small findings, and how long each one takes to fix	No check in that article is failed by most websites, and they are not the five failed most often.
The five checks most measured websites fail, how many of every hundred fail each one, and what a site that passes has that the others do not.	Five small checks, how many measured websites fail each one, and the engine's own note on what to change.	They are not the five checks failed most often, and the notes say what to change, not what a passing site has. Changed again on October 5, 2026.
and a box to see where yours falls.	The grade split and the spread of scores across the websites in our count, drawn.	That article has no box to type a score into. Changed again on October 5, 2026.
explained on a made-up example you can tap through.	The grade, the finding count, and the findings, explained on a made-up example.	Nothing in that article can be tapped through.
Insider: We took our headline number down. Here is why.
It said nothing about this.	It is for those two counties only, and other Florida counties differ.	The report gives that figure to every Florida site, and the page did not say it holds for two counties only.
Insider: What Google grades your website on
You can check your own scores free at chromeuxreport.web.dev	You can check your own scores free with Google's CrUX Vis at cruxvis.withgoogle.com	That address stopped working. Google now points readers to CrUX Vis.
Go to chromeuxreport.web.dev and type your domain.	Go to cruxvis.withgoogle.com and type your site's address.	That address stopped working. Google now points readers to CrUX Vis.
Every claim below links to a named source.	Each outside study below is named, with the size of its sample where it gives one.	Several claims below named no source, and none of them linked to one.
Google measured what made users go back to search results right after clicking a result.	Taken out.	We could not find a source for that account, so we took it out.
Attention spans have shortened since then, which makes the first-ten-seconds finding more urgent, not less.	Taken out.	We had no source for it, so we took it out.
An image in WebP format at a sensible resolution can carry the same visual information at one-fifth the file size.	A photo resized for the web and saved in a modern format such as WebP is often a fifth of the size or smaller, and looks the same.	Most of that saving comes from resizing the photo, not from the format.
will be fully reflected in your score on the following Monday plus 28 days.	A fix you make on a Monday is fully counted four Mondays later.	The old line added up to five weeks, and the window is four.
All major hosting providers include HTTPS certificates free.	Most hosting companies now include the certificate free; ask yours.	We could not show that every major host does.
A site someone looks after will often beat a neglected one by a wide margin.	We expect a site someone looks after to do better than a neglected one	We had no measurement behind it, so it now says what we expect.
and click the Field Data tab.	If neither tool shows a report, your site does not get enough Chrome visits for Google to publish one.	That tab reads the same Chrome report, so a site with no report found nothing there either.
The five findings most business websites fail, and what each one means.	Five small findings, and how long each one takes to fix.	No check in that article is failed by most websites, and they are not the five failed most often.
Insider: Nobody has a census of business websites. We are building one.
A quarter of the addresses never gave us a page	[a share worked out from the count] addresses never gave us a page	A quarter was further from the true share than the paragraph under the heading, which gave the exact figure.
About one in ten earned an A.	The middle one scored [number] out of 100. [share] earned an A.	A plain fraction was more than a point away from the true share. The page now prints the number when that happens.
It reads one page, the front page, in about ten seconds.	It reads one page, the front page, in a few seconds.	Ten seconds overstated how long the machine takes to read a page.
The count reaches 51 states, and it is thin	The count reaches [number] states and the District of Columbia, and it is thin	The count includes the District of Columbia, which is not a state.
sites are on it today.	No one has asked us to add a site yet. Or, depending on the figures: [number] [site is or sites are] on it because someone asked.	The sites on the list were ones we added ourselves. No site owner had asked.
Every number on it is read from our own records when the page is built.	Every count and share on it is read from our own records when the page is built.	Some figures on the page, such as how long a reading takes, were typed by hand.
the way a stranger's first visit receives it	the way the server first sends it to a stranger, before any scripts run.	A visitor's browser runs a page's scripts, and our machine does not.
on your site in about ten seconds	on your site in a few seconds	Ten seconds overstated how long the machine takes to read a page.
Insider: Where your score lands among the websites we measured
Here is the full picture, from every site the engine has measured.	Here is the picture from the [number] sites in our count.	The count leaves out chains, addresses that no longer lead to the business, and sites outside the United States.
Most sites land in the B range (47% of all sites measured)	The largest group lands in the [B] range ([share]% of the sites in our count) Or, depending on the figures: Most sites land in the [B] range ([share]% of the sites in our count)	Fewer than half of the sites graded a B, so most was wrong. The word is now chosen from the share each time the page is built.
an A means nothing was found to charge you with.	The top grade does not mean nothing was found. Of every 100 sites we have measured that grade A, [share] fail at least one check.	Many A sites still fail at least one check.
A score in the 60s is common.	A score in the [80s] is the most common.	Few sites score in the sixties. The eighties are the most common.
The weight on each check reflects how much that problem costs.	Each check carries a weight from [lightest] to [heaviest]: our judgment of what the problem costs a visitor.	The weights are our judgment, not a measured cost.
A page with no phone number cannot be called from a phone.	A phone number that cannot be tapped must be copied by hand.	The check fails a number that cannot be tapped, not a missing number.
Insider: How to read your inspection report without a web person
That is where most sites land.	Of the [number] sites in our count, [number] grade a D.	More sites grade a B than any other letter, and a D is far less common.
The cuts are the same as a school report	The grade bands are:	Our grade lines are not the usual school ones.
A site with no phone number is not asked about its booking system.	A restaurant is not asked about online booking; a salon is.	The booking check depends on the kind of business, not on a phone number.
weighted by how much each one costs.	weighted by how much we judge each one matters.	The weights are our judgment, not a measured cost. Changed again on October 5, 2026.
Seven small failures can hurt less than one large one.	One failure on the heaviest check costs as much as [number] on the lightest.	Under our weights, seven small failures always cost more than one large one. Changed again on October 5, 2026.
It is the engine being honest about what it cannot measure: whether text is too light to read, how the page looks on a real phone, whether your photos are any good.	Near the bottom is a list of checks the engine could not measure on your site, and why.	That list names the checks skipped on your site. The report says nothing about photos.
The report ends with a short summary.	Right under the grade is a short summary.	The summary sits under the grade, not at the end.
Serious findings mean the problems reach past any one page	Several serious findings means three or more of the heaviest checks failed on the page we read	The inspection reads one page, so it cannot say a problem reaches past it. Changed again on October 5, 2026.
Insider: Three questions that predict your inspection report
and each one is something you can see without a tool.	You can see the first two without a tool, and you can make a good guess at the third.	The third question says you cannot check it with your eyes.
A passing site has one that names the business, what it does, and where.	A passing title is [number] to [number] characters and not a placeholder like Home.	The title check passes any title of the right length, whatever it says.
The check looks for that block in the page head.	The check looks for that block in the page's code.	The check reads the whole page, not only its head.
most of your customers find you on a phone.	many customers look a business up on a phone.	We had no source for the word most.
Insider: The blue line people click: what your business looks like in a Google result
The only choice is who holds the pen.	Write it, and Google starts from your words instead of its guess.	Google may rewrite a description anyway, as the paragraph above it said.
The engine has measured 75,010 websites so far.	There are [number] websites in our count so far.	The engine has read more websites than the count holds. The count leaves some out, and the page now says which.
In about ten seconds it reads your live page	In a few seconds it reads your site as it is today	Ten seconds overstated how long the machine takes to read a page.
Insider: Five small findings, and how long each one takes to fix
The five findings most business websites fail, and what each one means	Five small findings, and how long each one takes to fix	No check in this article is failed by most websites, and they are not the five failed most often.
Nobody told the owner, so it stays that way for years.	Often nobody tells the owner, so it can sit there unnoticed.	We have no record that shows how long a problem lasts.
what a passing site has, and how long it thinks that takes.	what to change, and about how long it takes.	The notes say what to change, not what a passing site has.
Insider: A chamber of commerce website, scored
On the benchmark of 41 national sites this engine re-runs every few days, the median is 85.	our benchmark list at the time held [number] national sites, and [number] of them gave a grade.	That list had been replaced, and its middle score came from fewer sites than the line said. Changed again on October 5, 2026.
The results are reproducible by anyone who runs the same check today.	The engine has changed since this run, so the numbers will not match exactly.	The engine has changed since that run, so a new run gives different numbers.
The engine checks 28 things that can be measured from the delivered HTML.	the engine checked [number] things that can be measured from the page as the server sends it.	The engine checks more things now. That number was true on the day of the run.
Lee County is about 20% Hispanic.	About one in four Lee County residents is Hispanic	The census table gives about one in four, and the source the old figure cited could not be found.
Lee County's population is about 20 percent Hispanic	Census Bureau's American Community Survey five-year estimates	The census table gives about one in four, and the source the old figure cited could not be found.
Many of those residents are business owners.	Taken out.	We had no source for it, so we took it out.
The national benchmark median for server response time is around 400 milliseconds.	Taken out.	We had no source for that figure, so we took it out.
It renders correctly on phones.	It carries the instruction phones need to fit the page to the screen.	The check finds that instruction. It does not see how the page looks on a phone.
Insider: Your contact form might be lying to you
Here is the most expensive silent failure we find on small business websites.	Here is a silent failure that no outside inspection can catch, ours included.	Our inspection never sends a message through a form, so it cannot find this failure, and we have never measured what it costs.
Every week that form was broken, someone ready to hire you gave up without a sound.	While that form was broken, anyone who wrote to you through it was never heard.	Nothing we have counts lost messages by the week.
works from the outside, the way a stranger's first visit does.	works from the outside, reading the page the way the server first sends it.	A visitor's browser runs a page's scripts, and our engine does not.