How Much Electricity Does Bot Web Traffic Actually Use? A transparent order-of-magnitude bound (v1)

Public claims about the electricity consumed by automated ("bot") web traffic vary by more than an order
of magnitude, yet are routinely presented as single figures. This paper shows that no single honest figure
is 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 an
order of magnitude by three implicit methodological choices: (1) multiplying a request-share by total
data-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 of
a whole-network one (~31%); and (3) allocating energy by average share rather than the marginal
energy actually saved on a baseload system. Two independent methods  -  a top-down carve and a
bottom-up request count  -  then bracket the answer at order ~10 TWh/yr (plausibly ~1-70 TWh), while
disagreeing on the central value by ~5x  -  a disagreement that is itself the point. The contribution is
methodological: a reproducible "deflation" procedure built entirely from free, primary sources (Cloudflare
Radar, IEA), a derived (not asserted) allocation factor, and an explicit account of which assumptions
move the result most. All inputs, the computation, and the live data pull are included for reproduction.