AI
“Sam Altman said that the water used to produce a single almond grown in California is the same amount of water used by 38,000 ChatGPT queries." (Post image text: "SAM ALTMAN SAYS AN ALMOND PRODUCED IN CALIFORNIA USES MORE WATER THAN 38,000 CHATGPT QUERIES")”
Plain restatementOpenAI CEO Sam Altman publicly stated that the water consumed by roughly 38,000 ChatGPT queries equals the water used to produce one California almond. The post additionally presents that equivalence to its audience as a factual data point about AI water use.
Distortion codes this site does not recognise yet: harness_mismatch, cost_compute_omission. Not collectible until the field guide has an entry.
Sam Altman really did say this. On the September 1, 2026 launch episode of the Sources podcast with Alex Heath, he said that 38,000 ChatGPT queries use about the same water as producing one California almond. What the post leaves out is that Altman immediately added he did not have the exact calculation in front of him and that the number might be wrong. The figure appears to come from dividing a 2019 study's 3.2 gallon almond estimate by a per-query water figure that Altman himself published on his blog in 2025, which OpenAI has never shown the workings for and which has not been peer reviewed. PolitiFact rated the statement Mostly False, finding it pairs a generous almond number with a conservative AI number, and that independent estimates suggest roughly 1,000 to 10,000 prompts per almond rather than 38,000. The two sides are also measured differently: the almond figure counts rainfall and water-quality impacts, while the query figure appears to leave out the water used to generate the electricity. Experts quoted by CalMatters say public data on data center water use is too limited to settle the question either way, which is itself part of the story.
Sam Altman [drifted from the evidence:] said that the water used to produce [drifted from the evidence:] a single almond grown in California [drifted from the evidence:] is the [drifted from the evidence:] same amount of water used by 38,000 ChatGPT queries." (Post [drifted from the evidence:] image text: "SAM ALTMAN SAYS AN ALMOND PRODUCED IN CALIFORNIA USES MORE WATER [drifted from the evidence:] THAN 38,000 CHATGPT QUERIES")
[added by the neutral restatement:] OpenAI CEO Sam Altman [added by the neutral restatement:] publicly stated that the water [added by the neutral restatement:] consumed by roughly 38,000 ChatGPT queries equals the water used to produce [added by the neutral restatement:] one California [added by the neutral restatement:] almond. The post [added by the neutral restatement:] additionally presents that equivalence to its audience as a factual data point about AI water [added by the neutral restatement:] use.
Red-tinted words in the claim drifted from the evidence. Green-tinted words are what a neutral restatement needs.
The trace / claim to source
- Sam Altman did make this comparison. The quotation in the post's image is accurate to the wording every outlet reproduces.
- The venue is real and identifiable: the Sources podcast with Alex Heath, launch episode, September 1, 2026.
- The 38,000 figure is not invented by the post. It traces to a real statement.
- The arithmetic behind 38,000 is reconstructable and internally consistent from two real sources: Altman's own 2025 per-query estimate and a 2019 almond water study.
- Almonds are genuinely among the more water-intensive US crops, and per-query ChatGPT water use is genuinely small in absolute terms on any published estimate.
- The post's caption does add a caveat that this "does not mean AI has no environmental cost."
- Omitted qualifier: Altman said he did not have the exact calculation in front of him and that the number "might be wrong, but it's close." The post presents the figure flat, with no hedge, as a clean quote graphic. A reasonable reader takes 38,000 as a computed result rather than a speaker's off-the-cuff recollection.
- Marketing as evidence: the per-query water figure that generates 38,000 is OpenAI's own unaudited, non-peer-reviewed self-report, published on the CEO's personal blog, offered in an interview segment whose explicit purpose was rebutting criticism of his company. The post's hashtags #TechFacts and #DidYouKnow reframe an interested party's defensive estimate as a neutral factoid.
- Exaggeration: independent estimates cited by PolitiFact put the real ratio at roughly 1,000 to 10,000 prompts per almond. Altman's figure is between about 4 and 38 times higher than that range. The post carries the highest available number with none of that spread.
- Harness mismatch: the two halves of the comparison are not measured the same way. The 3.2 gallon almond figure includes rainfall and water-quality impact, while the per-query figure appears to count data center water only and excludes the water used to generate the electricity. Comparing a full life-cycle agricultural footprint against a partial-boundary compute footprint makes the ratio look far larger than a like-for-like comparison would.
- Cost compute omission: model training water is not in the per-query number, and neither is offsite water for power generation, which the ChatGPT-4o study PolitiFact cites does include and which moves the estimate to 0.6 to 17 milliliters.
- Exaggeration, second instance: the post's headline slide converts Altman's "the same amount of water" into "USES MORE WATER THAN 38,000 CHATGPT QUERIES," and the caption into "can use more water than roughly 38,000." Small in magnitude, but it converts an asserted equality into an asserted inequality favouring the AI side.
- Missing adjudication, no canonical name: by the time the post was published on September 3, CalMatters had already published a fact check the same day, and PolitiFact rated the statement Mostly False the next day. The post presents the claim as a viral curiosity with no indication that the underlying math was contested from the outset.
- I did not retrieve the podcast audio or a full transcript, so the verbatim wording rests on four independent named outlets reproducing it identically rather than on the artifact itself. The identical phrasing across outlets and PolitiFact's direct sourcing to the episode make error here unlikely but not excluded.
- I did not retrieve Altman's 2025 blog post or the 2019 Ecological Indicators paper directly. Both are reported consistently by multiple outlets.
- The true per-query water figure is not knowable from public data. This is not a gap in my search, it is the substantive finding: OpenAI does not publish the derivation, and the UC Riverside expert quoted by CalMatters says the public record is too limited to verify either side.
- Whether Altman intended the 38,000 as onsite cooling water only, or as a total-footprint figure, is not stated by him.
- OpenAI did not respond to PolitiFact's request for further information, so the company has neither defended nor withdrawn the derivation.
Altman did say this. On the September 1, 2026 launch episode of the Sources podcast with Alex Heath, he said, in the wording all outlets reproduce identically: "For every 38,000 ChatGPT queries, that is the same amount of water that is used in the production of a single almond in California." He said it while arguing that AI water-use concern is a "robust meme" that does not hold up to scrutiny, and that a modern very large data center uses water comparable to an office building's sinks and toilets. PolitiFact reports that Altman prefaced the figure by saying he did not have the exact calculation in front of him and that it "might be wrong, but it's close." Where the number appears to come from: 3.2 gallons per almond (the 2019 Ecological Indicators study) divided by 0.000085 gallons per query (Altman's own 2025 blog post) gives 37,647, which rounds to 38,000. The arithmetic is internally consistent, but both inputs are contested. PolitiFact rated Altman's statement Mostly False. Its reasoning: the comparison pairs a low estimate of ChatGPT water use with a high estimate of almond water use. The 3.2 gallon almond figure counts rainfall and water-quality impact, not just irrigation withdrawals; the fresh-water figure is about 6 liters. On the AI side, a study estimating total water including electricity generation put ChatGPT-4o at roughly 0.6 to 17 milliliters per prompt, against Altman's 0.32 milliliters. PolitiFact concludes available estimates imply roughly 1,000 to 10,000 prompts per almond, not 38,000. CalMatters interviewed Shaolei Ren of UC Riverside, who said the underlying public data is too sparse to verify either side, noting that location, outdoor temperature, cooling system, prompt length and reasoning load all change the answer. PolitiFact reports that OpenAI does not publish enough information to verify its own per-query figure and did not respond to its request for comment.
Complete reasoning
The reply is formatted for pasting into the thread where the claim is circulating.
Compact share page: ai.trueseeker.com/s/f72d0d17ae51/XZB1m5tCYQnVPIE2N9Q5RTO5a9y
Ask this case
Answers come only from the case file above; nothing is added.
Did Sam Altman actually say this about almonds and ChatGPT queries?
Yes. On the September 1, 2026 launch episode of the Sources podcast, he said 38,000 ChatGPT queries use about the same water as producing one California almond. The quote in the post matches what outlets reproduce.
So why is the post rated misleading if the quote is accurate?
Altman said right after the figure that he did not have the exact calculation in front of him and it might be wrong. The post drops that hedge and presents 38,000 as a clean, settled fact.
Where does the 38,000 number actually come from?
It comes from dividing a 2019 study's almond water estimate by a per-query water figure Altman published on his own blog in 2025. OpenAI has not shown the underlying workings, and the figure has not been peer reviewed.
Is 38,000 the right ratio?
PolitiFact rated the claim Mostly False, saying independent estimates suggest roughly 1,000 to 10,000 prompts per almond, not 38,000. It also found the two sides are measured differently, since the almond figure includes rainfall and water-quality impacts while the query figure appears to leave out electricity generation water.
Does anyone actually know the true water cost per ChatGPT query?
No. The case file states this is a genuine unresolved point, not just a gap in research. An expert quoted by CalMatters says public data on data center water use is too limited to verify either side, and OpenAI did not respond to requests for more information.