I spent an evening trying to figure out one simple question: Does E20 ethanol-blended petrol actually damage vehicles, or is it mostly noise?

I didn’t just read articles and pick a side. I used Claude as a sparring partner — asking it to defend a claim, then deliberately pushing back on its own answer, forcing it to check its sources, its math, and its assumptions at every step. That back-and-forth is the method I’d actually recommend for reading any contested news story right now — not “trust the AI” or “trust the article,” but treat either one as a first draft you’re obligated to interrogate.

It started with the official line: government trials, ARAI testing, automaker statements — no significant damage, marginal mileage loss, viral claims debunked. Clean and reassuring. So I asked the obvious follow-up: if it’s really not a problem, why doesn’t everyone use it? That pushed past the “myths vs facts” framing into the real economics and infrastructure reasons adoption isn’t universal.

Then came the water number — “10,000 litres of water per litre of ethanol” being debunked with a “3-5 litres” counter-figure. Instead of accepting either headline, I asked for the actual science and pushed directly: does growing sugarcane for fuel really cost thousands of litres of water or not? That forced out the real answer — both numbers were true, just measuring different boundaries. One counted only factory processing, the other counted growing the crop. Selective framing on both sides, not a lie on either.

Next it was the claim that ethanol comes from “surplus” rice and molasses, so it doesn’t compete with food. I asked why it was so confident about “surplus” — and that was enough to expose that it had absorbed the government’s framing uncritically without independently verifying it.

At that point I called it out directly: it had been sympathetic to the government’s side the entire conversation, treating ARAI and manufacturer numbers as neutral when every one of those sources had a direct stake in the program succeeding. It conceded the point immediately.

So I raised the bar: forget the news, reason as an automobile engineer, factor in India’s heat and humidity, and actually estimate the impact instead of quoting someone else’s conclusion. That’s when the real analysis showed up — ethanol absorbs water, and that gets worse with humidity; rubber degrades faster with heat, roughly doubling every 10°C rise. Combined with India’s mostly older vehicle fleet, that’s a genuine physics-based reason to expect more real-world wear than a controlled lab trial would predict — no political framing required, just materials science.

Then I did the mileage math myself: ethanol has roughly a third less energy density than petrol, so a 20% blend should cost 6-7% mileage — not the “3-3.5%” everyone kept quoting. That single push uncovered the trick: the smaller figure wasn’t comparing E20 to pure petrol, it was comparing E20 to the E10 fuel already at the pumps. Same physics, different baseline, very different-sounding number.

I pushed once more — that 3.4% is a best-case lab number, and real losses can’t be that clean once you factor in unreliable blending, transport, and storage across India. The estimate moved again: blending tolerance drift, water contamination risk in storage, and real-world driving cycles all push the honest range closer to 4-8%, not a flat 3.4%.

Last thing I caught: a pattern of reaching for the silliest viral claims — ants, sugarcane juice poured straight into tanks — as the go-to example of “misinformation,” which makes all skepticism look dumb by association. Once I named it, it was acknowledged outright: debunking someone’s weakest argument and treating that as proof the whole category of doubt is baseless is a rhetorical trick.

None of those steps happened on the first pass. Each one happened because I refused to accept the previous answer as final. Every pushback surfaced something that hadn’t been checked on its own: a hidden baseline switch, an unstated conflict of interest, an unexamined assumption, a rhetorical trick.

An AI — or a journalist, or an official spokesperson — will give you a coherent, confident-sounding answer on the first pass. Coherent isn’t the same as complete. The only way I’ve found to close that gap is to keep asking “compared to what baseline?”, “who benefits from this framing?”, and “what would change your answer?” — multiple rounds deep, not just once.

Most contested technical debates — fuel, nutrition, economics, health — run on exactly this pattern. Two sides aren’t usually lying to each other. They’re usually measuring different things and calling it the same number. Treating any single source — human or AI — as an oracle instead of a first draft is how misinformation and “official truth” both slip past unchallenged.

Worth remembering next time a headline says “myth debunked.”