About the project
About But AI Bad
But AI Bad checks familiar claims about AI against concrete evidence, fair comparisons, and sources readers can inspect.
Evidence before allegiance.
But AI Bad checks familiar claims about artificial intelligence without starting from “AI good” or “AI bad.” The question is narrower: what does the strongest available evidence actually support?
What the project does
Claim pages start with wording people recognize, identify the strongest fair concern behind it, and reach a verdict readers can scan quickly. Topics include jobs, water, energy, copyright, privacy, education, hallucinations, bias, security, and other claims that travel faster than their sources.
How a claim is checked
- Preserve the claim's central idea instead of disproving a technicality.
- Read the underlying sources, not only headlines or search snippets.
- Use concrete facts, dates, units, and same-category comparisons.
- Look for credible evidence both for and against the claim.
- Prefer at least three independent references, with five when the evidence supports it.
- State what remains uncertain and why it matters.
What the verdicts mean
A claim can be confirmed, rejected, misleading, or unproven. “Misleading” means a real concern exists, but the familiar wording points toward the wrong conclusion. “Unproven” means the available evidence cannot yet carry the claim.
Plain language, visible sources
Articles are written for readers who should not need a technical background to follow the evidence. Sources remain linked, major comparisons are shown directly, and material changes are recorded on the claim page.
Editorial position
This is not generic pro-AI marketing or an anti-AI archive. The project criticizes weak claims, not the people repeating them. A well-supported concern gets the same direct treatment as a claim that falls apart.
Automation and review
Automation may help find topics, collect candidate sources, and prepare drafts. Published claims still need source checks, evidence review, and an explicit publication decision. Generated confidence is not a substitute for evidence.
Funding
The site may display advertising to help cover its domain, hosting, research, and automation costs. Advertising does not decide which claims are covered or what verdict they receive.
Corrections matter.
If a source is stale, a number changed, or stronger evidence shifts the conclusion, the page should change too. Send corrections, better sources, or overlooked context through the contact form.