Lacewing Journal
Field notes on detecting machine-made things.
Research-backed writing on AI content detection, humanization, synthetic voice and deliverability. Sources cited, numbers checked, and our own products criticised where they deserve it.
-
Building an AI content QA pipeline that actually catches the right things
Five stages, the thresholds, and what to build versus buy. Detection belongs at exactly one of them. For five of the six risks a content team actually faces, it is the wrong instrument entirely.
-
Does AI-written cold email hurt deliverability? What the evidence actually shows
No mailbox provider has ever documented penalising machine-written copy, and no independent study exists either way. What is documented: authentication, complaint thresholds, list decay, and which inbox is genuinely hard to reach.
-
Does Google penalise AI content? What one million ranking pages actually show
AI-heavy pages rank at roughly the same rate at every position on page one. The policy language, the enforcement record, the data, and the traffic collapse that has nothing to do with who wrote your content.
-
Editing AI drafts properly: eight tells, three passes, forty minutes
The eight characteristic failures of model output, what to do about each, and a three-pass workflow that produces writing worth reading rather than text that passes a check.
-
Where AI detection actually pays for itself, and where it will bury you
Detector accuracy is not what decides whether your flags are trustworthy. Prevalence is. The ten-minute calculation to run before you buy, and six questions no marketing page can answer.
-
Watermarking and provenance: the case for marking AI content at source
Detection is forensics on a scene nobody photographed. SynthID, C2PA and the EU AI Act point at a different answer, and each has a specific failure the marketing does not mention.
-
Voice deepfakes in 2026: what detection can do, and what actually stops the money leaving
Benchmark-beating voice deepfake detectors fall apart on real phone calls. The documented losses, why lab accuracy does not survive a codec, how to read a vendor accuracy claim, and the cheap process controls that stopped two attacks cold.
-
What AI humanizers actually do to your writing
Four distinct techniques sold under one word. What the research says about whether each defeats detection, what each costs you in meaning and terminology, and why the leading independent leaderboard is owned by the tool it ranks first.
-
The false positive problem: who AI detectors wrongly accuse, and why
Detector errors are not random. They concentrate on second-language writers, neurodivergent writers and anyone writing in a constrained register. The evidence, the arithmetic, the court record, and a procedure that holds up.
-
How AI content detectors actually work, and where they break
Perplexity, probability curvature, supervised classifiers, watermarking. What each generation of detector actually measures, what independent testing found, and the theoretical ceiling nobody can engineer around.
-
The 10 best AI detectors in 2026, ranked honestly
Ten detectors ranked on independent test results, real pricing, and where each one falls down. Two of them are ours, and we have said so plainly along with what they get wrong.
-
AI content detectors in 2026: what actually works, and what does not
A short field report on what AI content detection can and cannot do in 2026, written by a team that sells it.