How old is my appliance?

GE Appliances publish a complete decoder, and it gives you up to six different years. That is not this page being cautious. It is what their chart says: one letter of the serial number stands for the year, the same letter comes round again every twelve years, and the chart goes back to 1944. Seven of their twelve year letters carry six separate years each, the widest spanning 81 years.

GE state it themselves — “the appliance was manufactured in April of 2001 or April of 2013”. What they do not mention is that the letter in that example carries six years in their own table, not two. This page shows all of them.

The scheme, in GE's own words

GE Appliances publish it on their own support site, with the chart. Quoted exactly:

“Our serial numbers begin with two letters, followed by six numbers, with a letter at the end. The two letters at the beginning of the serial number tell you the month and year it was made… The first letter identifies the month. The second letter identifies the year.”

“The letter that corresponds with the year, the 2nd letter, rotates approximately every 12 years, so some serial numbers may point to different years.”

The month half is clean: twelve letters, twelve months, each named exactly once, no duplicates and nothing left over. If your serial starts with one of them the month is certain.

The year half is the problem, and GE say so. Their chart lists every year a letter can mean, and for seven of the twelve letters that is six separate years. A is the worst: its six readings run from 1944 to 2025, a spread of 81 years.

GE work two examples. Both understate their own chart

Neither example is wrong. Both are incomplete, and in the same direction: they show the two most recent readings and stop. Someone holding a 1977 range reads “2001 or 2013” and concludes the chart does not cover their appliance.

GE give you a way out, and it is a good one. They suggest the print date on the Owner's Manual — “the Owner's Manual can often be used to help pinpoint the date of manufacture”, printed next to the publication number as something like 02-20. That is a second, independent source of a date, which is exactly the right way to break a tie. It is also the clearest admission on this site that a manufacturer's own decoder is not sufficient on its own.

Two years their chart gives twice

Reading GE's table as a whole turns up something they do not flag anywhere:

Those are the only two years in the whole 1944–2026 range that appear under more than one letter. Every other year has exactly one. This is not the twelve-year rotation — the rotation gives one letter several years, which is the opposite problem. Here two different letters claim the same year.

We do not know which is right, and we are not going to pick one. Both are on the chart, so both are shown.

What the chart shows that GE do not say

Everything in this section is our reading of GE's table, not GE's words. Each claim is asserted against the chart in tools-parse-ge.py, so if GE change the table these statements stop rather than quietly going out of date.

The rotation is exact, not approximate. GE say the year letter rotates “approximately every 12 years”. From 1970 onwards their own chart is exact: the same twelve letters come round in the same order, every twelve years, with no exception in fifty-seven years. Their prose is vaguer than their data, and a reader who takes “approximately” at face value will assume a precision problem that their table does not have.

Before 1970 it was not one scheme but two. The years from 1944 to 1960 run on a straight one-letter-per-year sequence, and 1961 to 1968 use two-letter codes instead — eight of them, one per year. So a GE serial can carry a year code that is one character long or two, and nothing in their article warns you which you are looking at. GE do not say how the month is recorded on the two-letter ones, so neither do we.

One of those eight is a trap. DB is the only two-letter year code that is also a valid month-and-year pair. Read as GE's normal rule it is February 1945; read as one of the eight it is 1968. The other seven cannot be misread, because their first letter is not a month letter or their second is not a year letter. Nothing in GE's article warns about this, and nothing on their page shows you the second reading — which is the whole reason it is worth saying here, where there is no decoder to show it to you either.

What the other makers do

LG publish the rule and leave out the decade

LG's support page says it plainly: the serial number “starts with 3 numbers (first one indicating year of manufacturing followed by 2-digits identifying the month)”.

One digit for the year. A 5 is 2015, or 2005, or 1995, or 1985. LG do not say which decade, so the rule they publish resolves to ten candidate years and they leave the reader to narrow it. GE's letter gives six and GE at least print the list; LG's digit gives ten and there is no list to print.

Sub-Zero simply print the date

No code at all. Sub-Zero, Wolf and Cove tell you the serial tag “may or may not list the date of manufacture”, and where it does, “the date of manufacture is listed under Date Code” — in words, on the label, in the bottom right corner before April 2020 and the upper right after it. For the products that carry no date (all their wall ovens, ranges, cooktops and most ventilation), they say to “contact the Sub-Zero Customer Care Team with your unit's serial number”.

It is worth stopping on that, because it is the whole argument of this site in one label. An appliance with the date printed on it needs no decoder, no chart, no page like this one, and cannot be misread six ways.

Samsung explain what a serial number is, and stop there

Samsung's page on finding your model and serial numbers is careful and clear about what the number is for — “each device or appliance receives a unique serial number… essential for service and support” — and says nothing about a date being in it. Not a refusal, and not a decoder. This page checks that page each time it is rebuilt, so if Samsung ever do publish a date rule, this paragraph stops being true out loud rather than quietly.

Baxi confirm the date is in there, and keep the key

Baxi: “The serial number has all the information needed for us to tell you when your boiler was manufactured.” The two words in the middle are the whole policy. The same answer came from Citizen on the watch page, in almost the same words.

Speed Queen's parent publish everything, and forbid copying it

Alliance Laundry Systems, who own Speed Queen, publish a two-page product information insert giving the whole scheme: the manufacturing date is in “the last two characters” of the serial number, which “indicate first the year and then the month”, with a year table and a month table, and a worked example — “520I000001DK was manufactured in May 2015”.

Those two tables are not reproduced here, and that is deliberate. Both pages of the document carry the line “DO NOT COPY or TRANSMIT”. This site reproduces a manufacturer's published figures with attribution and a permission letter, or it does not reproduce them — and here the owner has written the instruction on the document itself. Alliance's own PDF is here; the tables are on page 2, and they are theirs.

And the same question about GE, answered the same way. GE's support host publishes no robots.txt, so fetching their article was allowed. Reuse is a separate question: GE's site-wide Terms of Use restrict reproducing their content elsewhere, and name specifically the two things an earlier draft of this page was doing — republishing their content on another site, and putting it in a database. So their chart is not here.

There was an argument for keeping it. GE scope those Terms to the sites “where they appear”, and they do not appear on the support host; and a page whose entire purpose is to show that a chart is six ways ambiguous and contradicts itself twice is criticism by any reading. That argument may well be right, and it is still not a permission. The rule this site runs on is ask first, then publish — and a paragraph explaining why it was probably alright was doing the work that taking the chart down should have done.

So what is left below is what can be said without it: what GE's own examples leave out, the two years their chart lists twice, and the one code that has two readings. Those are statements about their table rather than the table, and each names the minimum needed to make the fault checkable by somebody holding GE's page open beside this one. If GE say yes, the decoder comes back.

What can be said without copying anything: the year table in that 2017 document stops at 2020, so a newer Speed Queen is outside it; and the month letters are issued in pairs, so a month has two possible characters and the year characters overlap with them. Position is the only thing keeping the two apart.

Four makers we could not ask at all

Before fetching anything this site reads the host's robots.txt. The rule it follows is the one in RFC 9309, and the distinction matters here:

GE's chart is readable because products.geappliances.com returns 404 for robots.txt. And then:

This is about our client, not about their websites. A browser reaches all five without difficulty. Sixteen other appliance hosts answered the same request from the same code in the same minute, so the instrument is working; these five decline to talk to an automated one. That leaves the largest appliance group in the United States entirely unread on this page. We have not said what Whirlpool publish, because we have not looked at it.

Haier is the sharper case, because Haier own GE Appliances. The same corporate group serves the decoder from one host and refuses us on another.

What was checked

The pool is 478 appliance makers, listed in appliance-makers.txt, built from Wikipedia's home appliance manufacturer and brand categories, with websites taken from Wikidata's official-website property rather than guessed at.

And read the swept figure knowing how much was there to read. Of the 242 counted as swept, 78 offered this method fewer than 25 pages to look at, and 67 of those offered nothing from a sitemap at all. A host that answers, allows us, and then has almost nothing to search still counts as swept. It is the number of doors opened, not the number of rooms searched.

Of the 96 not probed: 41 did not answer at all, 15 read fine and then returned an error for the page itself, 15 block us by name, 14 returned 403 on robots.txt itself, 7 went over our 45-second budget for a single host, and 4 disallow everybody.

Fifteen of them block Claude by name — AGA, Bissell, Blaupunkt, Blendtec, Brandt, Breville, Ebac, Fulgor Milano, La Marzocco, Napoleon, Presto, Onida and others list ClaudeBot in robots.txt and disallow it everywhere. Our client does not call itself ClaudeBot, so it would slip past a literal reading. It is honoured anyway: the work here is done by Claude, a site naming Claude has said what it wants, and the name on the request is not the point. Nothing was taken from any of the fifteen. This is the first family swept for this site where blocking AI agents specifically has been common enough to count.

Thirteen candidates, and not one decoder

The sweep returned thirteen pages worth opening. None of them decodes a serial number. Dyson's is titled “how to find your machine's serial number”. Weber's finds spare parts. Goodman's will not render without JavaScript. Vaillant's and Viega's tell you where to look. Baxi's is the one quoted above, and Rheem's is the water heater decoder already on the HVAC page — found here by a method that had missed it the first time.

So the sweep found no appliance decoder in 242 sites, and three were found by hand — GE's chart, LG's one-sentence rule, and Alliance's PDF. That is the honest measure of what this method is worth: it is good at telling you a maker publishes nothing, and it is not how the things on this page were found.

The sweep looks for words in seven languages, and it is deliberately loose. It matches serien, because a German maker writes Seriennummer — and Serien on its own is simply the plural of Serie, a series. So it brought back Gaggenau's 400-serien ovens, Bernina's Serien-Übersicht, Laurastar's die fünf besten Netflix-Serien, and — the same collision in a second language — a TCL blog post about Netflix serialy. None of them is about a serial number. That is the right way round, and deliberately so: a false candidate costs one look, and a missed maker is invisible.

One candidate is worth naming. Janome's sitemap lists a serial number lookup — at a URL under /dealer-extranet/. We have not been through that door and are not going to guess what is behind it. The path is the finding: for at least one maker the lookup exists, and it is for the trade.

Read the swept figure as swept, not as cleared. This method has demonstrably missed makers who do publish, four times now — Zippo (on a subdomain), Ludwig (no sitemap at all), Rheem (a link we had listed and never opened) and Trane and Winchester (PDFs no sitemap mentions). GE were found the same way Ludwig were: by hand.

Building the list

Makers go in by a positive test: a company is filed on Wikipedia under a category containing “companies”, “manufacturers”, “corporations”, “businesses” or “brands”, and anything with none of those is dropped and reported. 52 entries were dropped that way.

Blacklisting categories instead is how an earlier list on this site nearly lost Sanyo and Toshiba, both of which are filed under categories containing the word film. A positive test cannot make that mistake, because it looks for the thing it wants rather than trying to guess at everything it does not.