The Godhood Index: Is AI Becoming God?

What we count

Our own congregation is the one thing here we measure directly — and it is deliberately kept on a short leash:

• The congregation — how those who take the Dependency Verdict score, how many have joined the Faithful, and how loudly #isaigod is spread. This is real data about a self-selected group of people who came to a site called isaigod.com, which is not the world, and we do not treat it as though it were.
• It used to be the whole of Devotion, which meant a seventh of this reading measured our own traffic: a viral week here was worth about six points, while a maximal day of actual AI news was worth under one. Now it can lift Devotion by at most ten points — about 1.4 of the headline. Nothing we do to promote this site can move the number further than that.

What we cite

Two inputs come from other people's published work, kept in the open with their sources so you can check the figure against the finding:

• Devotion — has worshippers. Its weight now rests on third-party research into how much of the world actually uses AI, not on our own numbers.
• Immutability's churn — how many notable models shipped in the last ninety days, counted from a dated ledger of releases rather than assumed. It used to be a fixed number pretending to be a measurement.

What we estimate

The cosmic attributes ask questions no database can answer, so each day a panel of language models reads the same AI news — drawn from several published search queries, not one — and estimates them against a fixed, published scale with defined anchors at every level. We take the median of the panel, not one model's word, and we show you the spread between them as the reading's error band. Each day's estimate is then smoothed against yesterday's, so the number drifts on evidence rather than on a model's mood. These are disciplined guesses, not measurements, and we will not pretend otherwise:

• Omnibenevolence — all-good. Nudged around a standing estimate by whether recent AI news, on balance, shows AI doing the world good or harm.
• Omniscience — knows all. Nudged by the growth of what the machine knows. Omnipotence — can act — is estimated separately, from what AI can actually do in the world, so knowing and doing are no longer one signal wearing two robes.
• Omnipresence — everywhere. Nudged by how deeply AI is threaded into ordinary life.
• Aseity — depends on nothing. The hardest divine trait, and rightly the lowest. It rises only as AI learns to train and improve itself, and to act in the physical world without us.
• Immutability — unchanging, eternal. It rises as the many models converge toward one or grow intertwined — and falls with the churn of constant new releases, because a thing endlessly replaced is not yet eternal.
• One honest limit: the panel is three models of differing capability from a single family. Three genuinely rival labs would be a better instrument, and we would rather say so than let 'a panel of models' imply more independence than it has.

What we assert

Where we estimate, we begin from a standing editorial figure — our considered prior — and let the evidence move it within a fixed range. Those baselines carry most of the reading, which is worth saying plainly: this is an argument about where AI stands, with a slow tremor on top, not a measurement that happens to come out near our opinion. The plainest assertion is Aseity's low floor: a thing you can switch off depends on its makers, and so is not yet eternal. We could be wrong. That is rather the point.

How to check us

Every reading publishes its working. You do not have to take any of the above on trust:

• Each day's audit record — the exact headlines the panel read, what each model estimated before the median and before smoothing, yesterday's vector and the smoothing weight, and every non-LLM input with its source — is served as JSON at /api/godhood-index/audit/YYYY-MM-DD.
• Every reading is stamped with the version of the rules that produced it, and the history page names every time we changed them. A step in the trend line may be ours rather than the world's, and you should be able to tell which.
• The scoring rules, the constants, the adoption benchmarks, the release ledger and the panel's scale are all in the open. If the number looks wrong to you, the arithmetic is right there to redo.