The questions that arrive most often about AI measurement, detection, agents and the new rules — each answered with the organisation that issued the figure and the date it was published.
- Separately, automated traffic reached 53% in 2025 — but that counts bots of every kind, including search crawlers and monitoring.
- Seven detectors tested on 91 TOEFL essays falsely flagged 61.2% of work by non-native English writers, against near-perfect accuracy on native-written essays.
- Benchmarks award full credit for a lucky guess and zero for saying "I do not know." Under that scoring, a 60%-confident guess scores 0.6 and an honest abstention scores 0.
- The best agents complete roughly one professional task in four at first attempt, about 40% after eight attempts.
- Fines reach €15M or 3% of turnover for transparency breaches.
The questions that arrive most often, answered with the source and the date attached so you can check every one. If a figure here is wrong, it gets logged.
Is most of the internet really AI now?
No — and the two figures people mix up are measuring different things. Machine-written articles overtook human-written ones in volume in late 2024, then plateaued near half rather than continuing to climb. Separately, automated traffic reached 53% in 2025 — but that counts bots of every kind, including search crawlers and monitoring.Imperva Bad Bot Report 2026 (covering 2025). Methodology differs between editions
The widely repeated "90% by 2026" was a 2022 Europol projection. It did not happen. The timeline keeps it at the top deliberately.
Can AI detectors tell if something was written by AI?
Not reliably enough to accuse anyone. Seven detectors tested on 91 TOEFL essays falsely flagged 61.2% of work by non-native English writers, against near-perfect accuracy on native-written essays.Liang, Yuksekgonul, Mao, Wu & Zou, Patterns 4(7):100779, July 2023
A detector that flags six in ten non-native writers is a bias amplifier, not a detector. Full detail on how to spot AI writing.
Why does AI make things up?
Because it was scored for guessing. Benchmarks award full credit for a lucky guess and zero for saying "I do not know." Under that scoring, a 60%-confident guess scores 0.6 and an honest abstention scores 0.Kalai, Nachum, Vempala & Zhang, OpenAI, September 2025 (arXiv 2509.04664)
It is not malfunctioning. It is doing what it was rewarded for. Full mechanism here.
Are AI agents actually working yet?
Partly, and the headline figures are unstable. The best agents complete roughly one professional task in four at first attempt, about 40% after eight attempts.Mercor APEX-Agents benchmark, 21 Jan 2026
Agent success on OSWorld rose from 12% to 66.3% in a year — then the benchmark saturated, a harder replacement shipped, and the best model scored 20.6%. Nothing about the models changed. The instrument did.Stanford HAI 2026 AI Index · OSWorld 2.0 leaderboard, Aug 2026
Am I legally required to label AI-generated content?
In the EU and UK: only where no human editorial control was exercised. Article 50(4) of the AI Act became enforceable on 2 August 2026, requiring labelling of deepfakes and of AI-generated text published to inform the public on matters of public interest — but only where nobody reviewed it.Regulation (EU) 2024/1689, Art. 50(4) · Commission Guidelines, 20 Jul 2026
Fines reach €15M or 3% of turnover for transparency breaches. More on the rules arriving.
How much energy does one AI query use?
Estimates differ by an order of magnitude, from roughly 0.3 Wh for a conventional web search to around 3 Wh for a large model, depending on model, hardware and who is measuring.
A confident single figure is a sign the speaker has not read the range. More on energy and water.
Does prompt engineering still matter?
Less than it did. Frontier models handle a decent prompt fine. What moves the needle now is context — the documents, history, tools and memory surrounding the request.
The techniques tied to one model's quirks expire at the next release. Context engineering covers what does not.
Is it safe to let an AI agent act on my computer?
Only within a boundary you set explicitly. The most common failures are not stopping, wrong scope, and silent failure — reporting success having done nothing.
Agents also act on instructions found in content they fetch. Poisoning an agent's persistent state raised attack success from 24.6% to between 64% and 74% in one 2026 study.Wang et al., "Your Agent, Their Asset", arXiv 2604.04759 See prompting agents and guardrails.
Who writes this site?
One practitioner who uses these tools every working day — including on parts of this site — and says so on every page where it applies.
The site is deliberately anonymous. Every figure carries the organisation that issued it and the date, so it can be checked without knowing who collated it. Corrections are logged publicly at the corrections page rather than edited silently.
How do I cite something from this site?
Cite the original source, not this page. Every figure here names the organisation that issued it and the date it was published — those are the citations worth carrying.
Reference this site only if you are citing the collation itself: the comparison, the framing, or a correction logged here. Contact details are on the press page.