RatingPlate: the water heater site where the label beat the plan

A dead water heater and a phone flashlight on the label

The person here is standing at a water heater that just quit, flashlight on the label. Before they can price a heat pump they need to know what they own: the model, the size, the fuel, how old it is. No brand-neutral reference answered that, even though ENERGY STAR publishes clean public data on every certified unit.

I picked the name because the rating plate is the first step before every other question. When my first plan left that page out, I put it back by hand. Then I built the site around something else anyway: state guides, because the money question looked like the bigger search.

A publish gate, then state guides whose claims expire on their own

A Python build turns 1.3K ENERGY STAR records into pages but publishes only the 42 with complete, sourced specs (35 of them heat pumps), recounted with the site's own gate on Oct 13, 2026. Thin pages would have looked like everyone else's. Every field carries its source and the date it was checked. On the first day I stripped tank sizes from 233 tankless records that have no tank, and specs that belonged to a sibling model.

Then I made the state guides the pillar: electricity rates, rebates and payback for all 51 states, with a transparent score. The rebate claims expire on their own. Once a state's "launched" status is older than 30 days, the page stops calling the program current and says when it was last verified instead.

A claim with a date on it can downgrade itself. A claim without one has to be caught by a person, and that is where sites rot.

The data told me which page the site is

Google never sent a click: 0 across the 6 complete months from April to September 2026. Bing did, in small numbers. In September 2026, Bing clicks were about flat on the month before (15 in August, 14 in September). Bing's AI answers were the surprise. In September 2026, Copilot citations rose 36% on the month before (633 in August, 864 in September).

Copilot citations up 36% in September, a new best month, and the fifth month in a row it grew.

23 Apr start

109 May +374%

162 Jun +49%

392 Jul +142%

633 Aug +61%

HI-SCORE864 Sep +36%

Copilot citations up 36% in September, a new best month, and the fifth month in a row it grew.

Month	Copilot citations	Change from the month before
April	23	first month
May	109	+374%
June	162	+49%
July	392	+142%
August	633	+61%
September	864	+36%

Where those citations went is the story. Of the 2.6K citations Bing attributed to a page in the export window ending Oct 11, 2026, 1498 (58.5%) landed on /rating-plate-guide/. The 51 state guides, the pages I had built the site around, drew 190 (7.4%), spread over 25 states. In July 2026 I rebuilt the guide from sourced data it had never rendered: the fields for all four heater types, a full replacement guide, and a grounded FAQ. I can say the citations grew after that; the export windows overlap too much to say the rebuild caused it.

The rating plate guide

59%

All 51 state guides together

7%

The miss is maintenance. The energy prices on the live pages are from January 2026, and every state's rebate status was last verified in March 2026, so the self-downgrade has fired on all of them. It fired unevenly: the New York page says its program status "may have changed" in one card and calls the program active in its FAQ, because the downgrade never reached that surface.

How I measured it. Bing's AI Performance page export lists every cited page for this site (48 of 48), so the shares are exact for that window; Google Search Console and Bing Webmaster Tools daily totals summed by complete month; the publish gate recounted by running the site's own function over its own records. Shares compare pages; they do not prove what made one page win.

Identify first, then decide, and let a warning fail the build

I would trust the name again. The domain was a bet on the reader's first question, and the data proved it right after I had half talked myself out of it. What changed is the order: identify first, then decide. The guide should hand people to their state's math, not sit under a tools menu.

And a freshness warning nobody acts on is just a promise. Today the stale check prints a warning and softens the copy; it never stops a deploy, which is how January prices are on the page in October 2026. The next version fails the build when prices or rebate statuses go stale, and the downgrade reaches every surface that repeats the claim, the FAQ included.