Why your eye fails on anything taller than a house
People estimate size well up to roughly the height of a room, badly up to the height of a tree, and barely at all above that. The failure is not a lack of attention. It is that the eye measures angles, and angle stops tracking size the moment distance enters the picture.
Where does the failure start?
It starts at roughly the point where you can no longer reach the top of something. Below about two and a half metres, almost everyone estimates well, because the body itself is the ruler and the comparison is physical rather than visual. You know a door frame is a little above your head. You know a shelf is at your shoulder. Nothing about that involves looking.
Between about two and ten metres, estimates get worse but stay useful. A person can tell a giraffe from a horse without hesitating: a giraffe at 5.3 m against a person at 1.71 m is a ratio of about three, and ratios of three are still within reach. Above ten metres the useful part disappears. Asked how tall a five-storey building is, most people answer somewhere between ten metres and thirty, and the confidence of the answer has no relationship to its accuracy.
Why angle is not size
The eye does not measure size. It measures the angle a thing occupies in the visual field, and that angle depends on distance as much as on size. A thumb held at arm's length covers the Moon. To recover size from angle, the brain has to estimate distance, and distance estimation is itself unreliable past about thirty metres because the cues it relies on, mostly the disparity between two eyes and the apparent motion of nearer objects, stop being informative.
Past that point the brain switches strategy. Instead of computing size from angle and distance, it retrieves a size from memory based on what it thinks the object is. This works beautifully for familiar things and fails completely for unfamiliar ones, which is why an unlabelled photograph of a tall building is almost impossible to scale, and a photograph of the same building with a car parked outside becomes easy.
The Eiffel Tower is 330 m to the tip of its antennas. The Leaning Tower of Pisa is 56.7 m on its high side. Almost nobody guesses that the Eiffel Tower is 3.44 times Elizabeth Tower, which stands at 96 m, even though both are photographed constantly. What is remembered is a ranking, not a ratio.
The compression effect
There is a second failure layered on the first. When people are asked to estimate large sizes, their answers compress: big things are underestimated and small things are overestimated, and the effect grows with distance from familiar scale. The Burj Khalifa at 828 m is routinely guessed at half that. A garden ant at 5.8 mm is routinely guessed at twice.
This is why the Burj Khalifa against the Eiffel Tower surprises people. The real ratio is 2.51, which sounds modest, but nobody has a mental image that contains both at once. Each is stored at its own remembered scale, and the two scales were never calibrated against each other.
Animals are worse than buildings
Buildings at least stand still in a street with cars and people around them. Large animals are usually seen in water, in the air, or in a photograph with nothing familiar in the frame, and the results are correspondingly bad.
A blue whale is 27 m along its length. That is more than a quarter of Elizabeth Tower's height, which very few people believe on first hearing. A sperm whale at 16 m is longer than a modern London double-decker bus at 10.9 m by half again. A killer whale, which most people picture as enormous, is 7.3 m, meaning a male orca is only about forty per cent longer than a giraffe is tall.
The reason these land as surprises is that marine animals are almost always photographed either alone against water, which has no scale, or beside a diver, which fixes the scale but is usually cropped. Take away the diver and the whale could be any size at all.
What actually fixes it
Three things, in order of effectiveness.
A shared baseline. Two objects standing on the same ground line at the same scale is worth more than any number of words. It converts a size question into a length comparison on a flat surface, which is a task the visual system is very good at. This is the entire reason the game works: a silhouette resized against a reference is a problem the eye can attack, where the same question asked in metres is not.
A known object in the frame. The classic trick is a human figure, and it works because human height has a narrow distribution. The 1.71 m figure used here is a global average for adult men, and national averages spread by more than fifteen centimetres around it, but that spread is small enough not to matter when the thing being scaled is twenty times larger.
A ratio stated in figures. Not much bigger, not roughly twice, but the actual number. T. rex at 12 m against an African elephant at 3.2 m is 3.75, and having that number changes what you can do with the comparison. You can chain it: four elephants nose to tail overshoot a T. rex by 0.8 m. That kind of construction is what makes a size stick.
The limits of the fix
None of this makes anyone good at estimating. It makes them good at reasoning about estimates, which is a different skill and a more durable one. The gap shows up clearly at the extremes of the catalogue, where objects like Australia and the Moon sit at scales that have no experiential anchor at all. The mainland span of Australia is 4,035.6 km and the Moon is 3,474.8 km across, so Australia is the larger of the two on those axes, which is true and useless as intuition. You can verify it, chain it, and quote it. You will never see it.
That is the honest limit. Somewhere above the height of a mountain, estimation stops being perception and becomes arithmetic, and the best any drawing can do is remind you which arithmetic to trust.