Public claims about the electricity consumed by automated ("bot") web traffic vary by more than an orderof magnitude, yet are routinely presented as single figures. This paper shows that no single honest figureis recoverable from public data, and that a naive calculation - multiplying two figures that do circulate("bots ~ half of web traffic" and "data centres use ~485 TWh") - yields ~257 TWh, inflated roughly anorder of magnitude by three implicit methodological choices: (1) multiplying a request-share by totaldata-centre energy, a pool that includes a large and growing share of non-request workloads (storage,internal compute, and increasingly AI training); (2) using a web-application bot fraction (~53%) in place ofa whole-network one (~31%); and (3) allocating energy by average share rather than the marginalenergy actually saved on a baseload system. Two independent methods - a top-down carve and abottom-up request count - then bracket the answer at order ~10 TWh/yr (plausibly ~1-70 TWh), whiledisagreeing on the central value by ~5x - a disagreement that is itself the point. The contribution ismethodological: a reproducible "deflation" procedure built entirely from free, primary sources (CloudflareRadar, IEA), a derived (not asserted) allocation factor, and an explicit account of which assumptionsmove the result most. All inputs, the computation, and the live data pull are included for reproduction.