Quick answer: Google’s index is now serving eight different versions of our business at once, and the wider web has us at six different addresses. Two days ago we published the count as five. We were not wrong then — we were early. And the most useful thing we found this week was in our own process: three of our own measurements were wrong, and we can show you exactly how.
Last week we wrote about a speed number that was true for exactly one day. The week before that, we found that our address was in four different states at once. This week we went looking again, and found the same thing is still true — with two more states we had missed, one of them a page that no longer exists but is still being served in full.
Here is the whole sequence, as we measured it on each of the last five mornings. Read it as a snapshot series, not as a single fact:
Nothing about our published address, hours, or phone number changed on Tuesday, Wednesday, Thursday, or Friday. The count moved anyway. That is not a bug in our site — it is what a search index does while it catches up. A crawler reads a page, stores the address string against the URL, and keeps serving it until that same page is crawled again. Edit the page and the index does not notice. The index updates on its schedule, not ours.
Which means the honest way to write this is to date every number. If you are reading this in a week, the count may be nine. It will not mean we broke something.
And one more thing, because it is the sharpest version of the lesson. The piece we published on Wednesday carried the number five. We measured five on Tuesday morning, wrote the whole piece around it, and published Wednesday afternoon — by which time our own morning audit had already counted six. The article was correct when we wrote it and one count out of date when it shipped. That is what happens when you freeze a moving number into a headline and then wait a day to publish it. So this post does not carry a number in its title. It carries a date.
The method matters more than the number. We searched Google for our own business name against each address value we have ever published and read the snippets it served back. We also fetched our own live pages and read the address values straight out of the HTML. We did not read the URLs. We read the text Google was serving about those URLs, because the URL is not the claim — the snippet is.
That is the whole method, and it is reproducible in about ten minutes for any business. If your own name plus your old address returns a result, that result is live right now, and you can see what it says without asking anyone.
PO Box 1052, Moses Lake, WA 98837, Mon, Sat & Sun 9:00am-5:00pm. This is what the live site says, and it is what we want it to say. Across all 127 published URLs on this domain, the address appears 549 times, on 126 of them — every page except a résumé, which correctly has none.
Four separate surfaces carry it, and all four agree: the structured data, the visible address block, the footer, and the microdata. We checked each one separately this morning rather than trusting one of them.
PO Box 1050 is the box number we published ourselves, then changed to 1052. It is the uncomfortable state, because nobody else got it wrong — we did, briefly, and the crawler caught us while it was true.
What is new today is that this state has split into two. Some pages are still served on PO Box 1050 with a Mon–Fri week. Others are served on the same retired PO Box 1050 with our current Mon, Sat & Sun week. So the index has started recomputing one state while still holding the old value — a hybrid of a thing we retired and a thing we kept.
223 W 3rd Ave, with Mon–Fri. The pre-PO-Box footer. Google is still serving it on our homepage, our testimonials page, our ecommerce page, and our Tri-Cities, Yakima, and Toppenish service pages — pages it has not re-crawled since we changed the footer.
The week is wrong too. The live footer says we are open Monday, Saturday, and Sunday. The cached footer says Monday to Friday. A customer who believes the cached version and calls on a Saturday reaches a closed door on one of the three days we actually work.
And our homepage is served both ways at once — on PO Box 1052 with the current week in one search, and on 223 W 3rd Ave with the weekday week in another. One URL, two answers, no way to tell which one a visitor sees.
Our no-slash /contact is still served with a Gmail address we no longer monitor, a Sat–Sun Closed notice, and the footer from the theme we replaced two years ago. Our no-slash /about is still served as “a small business web design agency” selling cheap hosting at $2.99/mo — with a legacy navigation listing 22 pages, including a Terms of Service page we do not have.
Read that second one again. It is a different company, still answering for our name. It is not somebody else’s information. It is a version of us we published and then stopped publishing.
This is the new part, and it is the part we had missed twice.
Our correctly-spelled /web-design-ritzville-wa/ returns 404 live. The live site has always carried the misspelled version — /web-design-ritaville-wa/ — because that is the slug we published. The index, meanwhile, is serving a complete legacy page for the spelling we never published, including that same 22-page navigation with the Terms of Service entry in it.
And our /about-me/ page returns 404 live while the index has been serving the full article body and the legacy navigation alongside it. That one did not resurface in today’s searches — it may already be fading. We are reporting it because it was there, and because “it may be fading” is not the same as “it is gone.”
So the index is not just holding old strings on live pages. It is holding whole pages that no longer exist, and it will keep serving them — to people, to scrapers, and to AI systems — until a crawler hits a URL that no longer resolves and gives up. Nothing we do to the live site touches those.
Here is the part we did not expect to write. The count is partly a measurement, not a measurement of the index. When we went from seven to eight, some of that was a genuine new state — the dead-but-served pages above are real. But some of it was that we got better at looking. A count that moves because the instrument moved is not a count of anything. It is a count of how carefully we looked that morning.
Which is the real lesson, and it applies to every number in this post. The address count is a snapshot of the gap between our newest published version and our oldest, taken this morning at 09:00 UTC. It is not a property of our business. It is a property of the distance between two dates.
We published the piece that said our address was in five states, and then our own morning audit flagged that piece as a defect. It said the post “prints all four retired addresses” and recommended we rewrite it. We re-measured it properly — against the four surfaces that actually publish an address, rather than against the whole page — and every one of the retired values is inside a code tag where we quote it as evidence. Zero of them appear outside it. All four published address surfaces say the same thing.
The instrument was wrong. The article was right. We fixed the check and published what we found. Then this morning the same kind of check produced a different wrong answer: it reported that the article had been edited overnight and had grown a fifth retired address. It had not been edited. The post’s own modification timestamp is from before it was published, and the build file we wrote it from already contained the string in question. The article never changed. The audit’s list of values to look for did.
Which brings us to the third wrong measurement of the week, and the one that cost us three days. Our site was serving cold loads of 1.17 seconds and we opened a ticket blaming our own code. Then the edge family we were served from changed overnight, with no deploy on our side, and the same origin with the same bytes started responding in 0.41 seconds. The identical site, the identical server, the identical content — 1.17 seconds on one edge pool, 0.41 on another. We were measuring the edge and calling it our code.
Three wrong measurements in one week, all of them ours, all of them the same failure: we checked a thing by looking for the strings we already knew about, instead of looking at the surface that publishes them. A number without a method behind it is a decoration — and that turns out to apply to our own tooling. If you want to know what your business looks like to a search engine, read the surface that publishes the answer, and read the thing being served to a stranger. Those are two different questions, and we kept conflating them.
Which box is the box. Whether the studio started in 2003, 2005, or 2014 — our own property still says both, and a directory we have never heard of says 2006. Whether we are open on weekends. Whether that apartment number on a directory badge should be published at all. Whether an office we do not have should be listed as an address on a sister site that uses our phone number.
Those are not ours to guess at, and the fix is not a plugin. It is four decisions and then a stack of mechanical edits. Get those right and every remaining directory is a form submission.
One more thing we found this morning that we are not going to soften. Our sister site publishes a physical office in Wenatchee — on our phone number, with no PO Box — and it does so on pages that return 404. Our main site says we are a mailing address in Moses Lake. Both are live right now. Google can index an office address on a URL that does not exist.
One box digit and one opening week. One phone number. One founding year. One email address. The phone number is already settled. Every other surface becomes a copy-paste once we know what to paste.
And we are not going to publish a table of our own failures without saying which one was the truth. The site has converged on PO Box 1052 — but convergence is not confirmation. 1052 is what we were already publishing, so our pages agreeing proves only that we are self-consistent, not right.
👉 Wednesday’s piece is the long version. We wrote it a day after we measured it, and it was already one count out of date by the time it shipped. That is the whole argument, in miniature.
Gorden Web Design serves small businesses across Central Washington including Moses Lake, Quincy, Ephrata, Othello, Warden, Soap Lake, Ritzville, Connell, Toppenish, Yakima, Wenatchee, Ellensburg, and the Tri-Cities (Richland, Kennewick, Pasco). Remote projects accepted across Washington state.
Our projects are fixed-price based on scope. Most small business websites range from $1,500 to $5,000. E-commerce starts at $3,000. Logo/branding starts at $800. All pricing is fixed u2014 no hourly billing.
Most websites are completed in 2 to 4 weeks from kickoff to launch. E-commerce and custom projects take 4 to 8 weeks. Rush delivery is available at +20% premium.
Yes. Monthly maintenance plans from $99 to $299 cover WordPress updates, backups, security, uptime monitoring, and content edits. Three tiers: Basic, Standard, Premium.
A number without a method behind it is a decoration. Ours was. So is an address, apparently. And so, it turns out, is a slow-loading site that was never slow.
If you want your own numbers read honestly — including when they are unflattering, and including when they are about you rather than about us — get in touch, or call 509-230-4576. We run the audit every morning.
— Douglas Gorden Jr
Gorden Web Design, Moses Lake WA
Gorden Web Design & Computer Repair ·
PO Box 1052, Moses Lake, WA 98837 · (509) 230-4576 · Mon, Sat & Sun 9:00am-5:00pm