← Back to Calculator

How we check our numbers

ElectricityFinder prices plans from each provider’s Electricity Facts Label where one is posted. An AI model reads each label’s pages, and we store the prices, fees, credits and renewable percentage it finds. On November 15, 2026 we read every label a second time, with a different AI model working from the PDF itself, and compared the two. We plan to repeat this check every month.

The provider’s Electricity Facts Label is the official statement of a plan’s terms; check it before you enroll.

Last full check: November 15, 2026. Next planned: mid-December 2026.

Results of the last check

CheckResultDetail
Every plan with a facts label was read a second timedone930 of the 944 plans listed that day had a label (14 had none). All 930 were read again by a second AI model, given each label’s PDF directly.
The labels read were the ones our prices came frompassEach PDF’s fingerprint matched the copy our pipeline used.
Prices and charges1 error, fixedOne plan’s energy charge was stored as 9.725¢ per kWh; its label says 9.9725¢. Corrected on November 15.
Renewable percentage61 too low, fixedWe stored the whole number below a label’s decimal figure (25.5% stored as 25%). 56 were corrected the same day. The other 5 needed a separate fix to how we match a provider’s name, and were corrected in the next morning’s update (November 16).
Differences with no effect on the price we show27A daily fee stored as its monthly total, tiered energy prices, and delivery-charge figures copied from a label with a slipped digit. We read each kind by hand on at least one label.
A label’s renewable figure differs from Power to Choose’s by more than a point30Not an error in our data. We show the label’s figure, with a note giving Power to Choose’s.
One side lists a charge as zero, the other leaves it blank181Not counted as an error.
A label that may show two renewable figures1, not yet checkedThe checker read two renewable percentages on one label for the same product. We haven’t yet read that label by hand.
Where the checker itself was wrong22The model misread 20 labels from one provider: where a label breaks down the sources of a 100% renewable product, it listed those shares as two separate product figures. Our comparison script flagged 2 more by cutting off a decimal. None of these were errors in our data.

That’s 323 of the 930 plans with any difference, and 607 with none.

What was checked

For each plan: the label’s energy charge, base charge, delivery charges, average price at 500, 1,000 and 2,000 kWh, contract length and renewable percentage, each compared with what we store. Bill credits and cancellation fees weren’t compared in this run. Every kind of difference was read by hand on at least one label before we called it an error.

What this check can’t see

  • The path from our data to the page. It compares labels with what we store, not with what a page shows.
  • Delivery-rate changes. The delivery charges we use for pricing come from the utilities’ rate filings with the Public Utility Commission of Texas, not from labels. We track those separately.
  • A label changed since our last look. If a provider reissues a label between our checks, this check doesn’t know.
  • The checker’s own mistakes. It misreads too (22 times last run), so nothing it flags is called an error until a person has read the label.

How the check runs

Our pipeline’s reading comes from one AI model. For this check, each plan’s PDF is sent to a different AI model as a document, and it returns the values it reads. A script compares those with our stored values and sorts each difference into the classes above. The November 15, 2026 run read 930 PDFs and cost about $15.