RECYCLING · EFFICACY
A source-verified field guide · by material · updated 2026

Does recycling actually work?

Sometimes spectacularly; often barely. The answer swings wildly by material — and the chasing-arrows symbol is the worst guide to which is which. This compiles real-world rates, carbon & energy saved, what each saving is worth against carbon-offset prices, and which products contain which materials. Researched in parallel by four AI models and adversarially fact-checked by two.

~9%
of all plastic ever made has been recycled (Geyer 2017). Only ~10% of that more than once.
~95%
energy saved recycling aluminum vs smelting from ore — the biggest win of any material.
~99%
lead-acid batteries recycled in the US — the highest rate of any consumer product.
43%
US aluminum can rate in 2023 — its lowest in decades, down from ~50% (2018).
01

The materials, ranked

Every common recyclable — its latest US rate, energy/carbon benefit, an honest verdict, and what it's actually found in. Search or filter, and note how the verdict tracks the material, not the recycling symbol.

Worth it Conditional Marginal Not worth it · C1 primary → C5 unknown
02

The payoff, visualized

One picture makes the point: the carbon and energy wins are wildly unequal. Aluminum's per-tonne carbon benefit dwarfs glass by ~30×. Switch the metric, then download or copy the chart to share.

03

What's it worth?

Recycling avoids carbon — so how does it stack up against simply paying to remove the same carbon another way? Pick a material and amount; this converts the avoided CO₂e into dollar-equivalents across four very different offset markets, plus energy & everyday equivalents. (A useful gut-check when comparing recycling to funding a different climate project.)

Offset prices (USD/tonne CO₂e, 2023–2026): Social cost of carbon $190 (US EPA 2023, central, 2% rate — what a tonne's damage is "worth" to society) · Voluntary market avg ~$6.5 (Ecosystem Marketplace 2023; cheap but quality-contested) · EU ETS ~$83 (compliance market, 2025) · Direct air capture ~$1,000 (Climeworks today; retail portfolios $100–500; ~$400–600 target by 2030). Carbon figures use EPA WARM net recycling factors; lead-acid & some plastics lack a published per-tonne factor.

04

The case against the bin

The skeptical findings every research pass independently confirmed.

The resin code isn't a recyclability label

The chasing-arrows "♻ 1–7" is a 1988 plastics-industry resin ID, not a promise of recyclability. EPA, ASTM D7611 and the FTC all say so. #3–#7 are accepted by almost no curbside program.

Plastic recycling is mostly "one and done"

Of all plastic waste through 2015, ~9% was recycled and only ~10% of that more than once — so under 1% of plastic has ever been cycled twice. Recycling, the authors wrote, "delays rather than avoids" disposal.

"Recycling rate" is ambiguous by ~2×

Same material, same year: cardboard is 96.5% (collected) or 53.5% (net); aluminum is 50.4% (cans) or 17.2% (all-Al). Collected ≠ reprocessed ≠ new product.

Even the winners are slipping

US aluminum cans fell to 43% (2023), the lowest in decades. China's 2018 import ban cut plastic exports ~99% and dropped the US plastic rate from 8.7% (2018) to ~5–6% (2021).

Circularity ≠ climate impact

Glass is "infinitely recyclable" but its carbon benefit per tonne is ~30× smaller than aluminum's — you still melt silica at 1,500°C. Infinite loops and big carbon wins are different axes.

The true net carbon is under-counted

Most lifecycle studies stop at the reprocessor, only partly counting curbside trucks and wash water. That understates the burden most for heavy glass and contaminated plastics.

05

Master comparison

All figures sourced; WARM = US EPA's lifecycle model (net recycling factor, per short ton; negative = carbon saved). US rates show the latest available with year.

MaterialRecyclable?US rate (yr)CO₂e saved (WARM)Energy savedVerdict

Plastics energy savings (APR/Franklin LCA) are on a cut-off, total-energy-incl-feedstock basis; process-only figures are lower. Aluminum 95% = smelting basis vs ~80% on EPA WARM's manufacturing basis — a boundary difference, not a contradiction. EPA's national MSW dataset is still 2018 (no newer one exists); material-specific rates use the latest industry/Eurostat data.

06

How this was made

Built to fight the two failure modes of AI research: hallucinated numbers and optimistic framing.

01

Four parallel researchers

The same brief went to four AI models independently — Claude (×2), OpenAI Codex, and xAI Grok — each doing its own web searches. No shared context.

02

Synthesis & diff

Answers were merged into one report, every disagreement flagged, and cross-confirmed figures kept at higher confidence (C1–C5 tiers throughout).

03

Adversarial fact-check

Two different models then tried to break the report — re-verifying each load-bearing number against primary sources (EPA, Geyer 2017, WARM, OECD, NAPCOR, Eurostat).

04

Refresh & iterate

A later pass re-checked currency (2023–2026 figures) and added the offset-cost and product data — all three models agreeing before it shipped.

The fact-check earning its keep

A draft claimed "PP plastic saves only ~8% energy." Both Codex and Grok independently caught it — an inversion of the source's "12% of virgin," i.e. an ~88% saving. A same-model self-review would likely have re-asserted it. The two-model loop killed it.

EPA Facts & Figures (2018)EPA WARM v15EPA Social Cost of Carbon (2023)Geyer et al. 2017OECD / UNEPNAPCOR 2024 · Aluminum Assoc. 2023Eurostat 2023APR/Franklin LCAEcosystem Marketplace · ICAP · Climeworks
ai gen