An Exxon briefing prepared in 1982, reproduced in a later study, included a graph of rising atmospheric carbon dioxide and the warming expected to follow. Decades later, researchers digitized that and other projections, then compared them with what had happened. The curves were sufficiently specific to test.
The projections depended on assumptions about future emissions and the response to them. A conditional estimate could be useful without specifying the exact future. Their task was to state the assumptions and the range, rather than wait until everything could be known.
The controversy over climate denial concerns what happens to that distinction in public. Uncertainty can describe the limits of knowledge, or it can be made to suggest that there is no dependable knowledge at all. Exxon and Mobil's research and public messages used uncertainty differently. A later dispute over scientists' emails tested how clearly the scientific community could explain uncertainties in its own work.
Whether Earth is warming, what is causing it, how much more warming a given change would produce, and which policy to choose are related questions. They do not require the same evidence or permit the same kind of answer.
The greenhouse mechanism is a starting point. Sunlight supplies energy to Earth's surface; the warmed surface emits infrared radiation. Gases including carbon dioxide absorb and emit radiation in parts of that infrared spectrum, slowing the loss of energy to space. Increasing their concentration changes the energy balance. The resulting warming and its feedbacks are explained in NASA's account of the causes of climate change.
Carbon-source evidence addresses a different part of the explanation. NOAA describes several lines of evidence, including the amount of fossil fuel burned, accumulation in the atmosphere and oceans, and changing carbon-isotope ratios. Isotopes are forms of an element with different atomic masses. Fossil fuels contain effectively no carbon-14 because they are so old. Adding their carbon changes the isotopic mixture. Carbon-13 supplies another clue, although both fossil fuels and deforestation can produce a depleted carbon-13 signal.
A rising concentration alone would not identify the source. An isotope ratio is interpreted within the carbon cycle, with other contributions considered. The attribution is stronger because measurements and accounting constrain one another. The conclusion draws on observations and their interpretation, including checks on alternative sources.
The IPCC's 2021 physical-science assessment concluded that human influence had warmed the atmosphere, ocean and land. It estimated human-caused global surface warming from 1850–1900 to 2010–2019 at 1.07°C, with an assessed likely range of 0.8–1.3°C. The baseline and period are essential parts of the estimate. It describes a global change, without assigning a single cause to every recent weather event.
There is more uncertainty about some future responses. Equilibrium climate sensitivity describes the warming after the climate adjusts to a doubling of atmospheric carbon dioxide. The same assessment gave a best estimate of 3°C and a likely range of 2.5–4°C. This is a range for a particular response, not an admission that the greenhouse mechanism might produce no warming. The difference between its lower and upper ends is important for assessing risks and responses. The range is not a forecast for a particular calendar year. A forecast also needs assumptions about how concentrations change and how much of the adjustment has occurred by the date being considered.
The 2023 study by Geoffrey Supran, Stefan Rahmstorf and Naomi Oreskes examined projections reported in company documents and scientific publications between 1977 and 2003. Some came from models run by company scientists; others were third-party work discussed in company records. Their analysis distinguished work done in-house from projections the company scientists had learned from elsewhere.
They identified sixteen temperature projections in twelve documents, extracting values from tables and digitizing old graphs so they could compare the numbers. They then compared projected warming through time with observations. A plotted temperature change needed a common reference point before the comparison meant anything: two curves using different baselines can be separated vertically even when they show the same warming.
They also compared the modeled response to the change in radiative forcing—the change in the energy balance driving warming. That second comparison matters because an old model might respond reasonably to carbon dioxide while its assumed future emissions differed from what occurred. If emissions rose more slowly than a scenario assumed, comparing only temperatures at a given date could make a reasonable climate response look too large. Separating the assumed input from the modeled response helps identify what went wrong, or right.
The reported average warming projection was about 0.20°C per decade. Most projections were consistent with subsequent warming under the study's tests, with performance comparable to contemporary independent research. The usefulness lay in forecasting a trend within an uncertainty range. The study did not find that every projection was accurate.
The result makes an indiscriminate public dismissal of climate models difficult to defend. Scientists working in and around the company had produced useful conditional estimates with the tools available. They had obtained information worth using despite the uncertainties.
Some of this work was published, a fact easily lost in the phrase “Exxon knew.” A scientific paper can be available without reaching the audience of a newspaper advertisement, or with the same message being emphasized in both places.
Supran and Oreskes's 2017 communications study compared internal documents, peer-reviewed research, other publications and paid editorial-style advertisements. Their original sample contained 187 documents across those four categories. The advertisements appeared on the newspaper's opinion pages, where the company had paid for space to present its own message.
The researchers classified what each document acknowledged or doubted. Their method allowed for uncertainty that was reasonable in the period when a document appeared. For the question of whether warming was real and human-caused, they found acknowledgment common in the research and internal categories, and doubt much more common in the sampled advertisements.
Classifying language requires judgment about what a statement communicates in context. The document sample and coding rules therefore need scrutiny too. ExxonMobil rejected the study, saying that its statements had been consistent with its understanding of the science. A methodological critique by Kimberly Neuendorf, retained and paid by the company, questioned sampling, the independence of the coders and the grouping of communications from different corporate entities and periods.
The coding also extended beyond whether warming was real and human-caused. It included judgments about the seriousness of impacts and the difficulty or cost of mitigation. Those categories involve economic and political arguments as well as physical evidence. A result for the whole scheme should not be treated as a simple count of statements denying the greenhouse effect.
The distinction between the companies also matters. Exxon and Mobil were separate before their 1999 merger. An internal Exxon document and a Mobil advertisement are not automatically evidence that the same managers approved contradictory messages. The same researchers addressed that distinction in a 2020 addendum, expanding their advertisement search and reporting that the discrepancy persisted when they examined the companies separately.
The regional-temperature example makes the dispute easier to follow. The 2017 paper discusses a 2000 ExxonMobil advertisement titled “Unsettled Science.” It used a reconstruction of historical sea-surface temperatures in the Sargasso Sea. The authors explain why treating that regional record as a picture of the whole planet would misrepresent what the underlying study measured.
A record of past regional variability can be perfectly real and still fail to support the conclusion suggested by its placement in an argument. Showing that climate changed before substantial industrial emissions does not determine the cause of the present change. The issue is the inference being invited, rather than whether a curve exists somewhere in the scientific literature.
A reader can accept the regional record and still reject the conclusion invited by the advertisement. The record describes one place's past; identifying the causes of recent global warming needs evidence at a different scale. Quoting a genuine result does not remove that gap.
The studies make a substantial case for a discrepancy between scientific work and public messaging. Measuring how much each advertisement changed public opinion is a further task. The content comparison identifies a message; it does not by itself measure its success.
In 2009, emails from the University of East Anglia's Climatic Research Unit became public without authorization. The phrase “hide the decline” was presented as evidence that scientists had concealed falling global temperatures. The email concerned a particular use of proxy data.
Tree rings are proxies: their widths or densities can help reconstruct past conditions after comparison with instrumental measurements. Some series ceased to track instrumental temperatures in the later twentieth century. A decline in such a proxy-derived series is not the same thing as a measured decline in global temperature. It raises a problem about how the proxy relates to temperature.
The 2010 review chaired by Muir Russell examined the conduct behind the controversy. On the specific allegations it examined, the review upheld the scientists' honesty and found no behavior undermining the balance of advice to policymakers. It also criticized a persistent failure of openness.
On the disputed figure, its conclusion was more uncomfortable than a simple vindication. A graphic supplied for the World Meteorological Organization's 1999 report combined proxy and instrumental records and truncated one proxy series after 1960. The review found the presentation misleading because those procedures had not been made sufficiently clear. It did not hold that joining data or ending a reconstruction was inherently improper.
Without an explanation, someone looking at the figure could mistake the whole line for one consistent kind of measurement. They would not see where the proxy ceased to track the instruments. A caption needed to explain both the change of source and the omitted segment, so the reader could judge how the picture had been constructed.
The review also stated that it was not issuing a verdict on the scientific validity of the researchers' work. That would require scientific examination, rather than an inquiry into emails and conduct alone. The report's conclusions need to be read within that remit.
Correcting the allegation about concealed global cooling does not require defending the presentation. A scientific community protects its credibility by explaining and repairing such failures, not by insisting that criticism must disappear because opponents have exaggerated it.
The sensitivity range leaves consequential uncertainty about the response to doubled carbon dioxide. Local impacts and the pace of change create further questions for infrastructure and public spending. Each deserves an answer at the scale where a decision will be made.
An uncertain magnitude can still support action. A town does not need to know the exact height of its next flood before considering whether essential equipment should remain in a vulnerable basement. It needs a defensible account of the risk, the cost of protection and the consequences of being wrong in either direction. Uncertainty changes that comparison; it does not automatically make postponement the neutral option.
Conversely, accepting the physical evidence does not settle the design of a tax, the distribution of costs or the power an institution should receive. A policy can be challenged for burdening people who have few alternatives. That challenge should identify the burden and a better response, rather than depend on claiming that carbon dioxide cannot affect climate.
This is where Agenda 21 & Sustainable Development and The Club of Rome & The Limits to Growth connect to the argument: concern about a proposed governing arrangement is different from evidence about a physical process. The concern may deserve a hearing on its own terms. It cannot establish that the process was invented to justify the arrangement.
In the company scientists' work, uncertainty coexisted with useful projections. A public message suggesting that no sound judgment can be made has to account for that achievement. The remaining uncertainty should help define the decision and its risks, rather than make the knowledge already obtained disappear.