About Wibble

What we build

Wibble is an AI humanizer: it rewrites AI-assisted text so it reads like natural human writing and passes AI detectors, without wrecking the meaning, the argument, or the citations. Around that core sit citation tooling, connectors for Claude and ChatGPT, and an API.

Alongside the product we run a public testing lab. The AI humanizer market is saturated with unverifiable numbers — tidy pass rates with no corpus, no detector versions, no dates, no raw outputs. The lab exists to test humanizers and detectors properly and publish the data so anyone can check it.

Who we are — and why we publish anonymously

Wibble is built by a small team that publishes anonymously. That is a deliberate choice, made for two reasons.

First, writing publicly about detector evasion invites harassment. This is a contested topic, and people who attach their names to it become targets. We would rather spend that energy on the work.

Second — and more importantly — our work is designed not to need personal authority. A named expert asking you to trust their judgment is a weaker guarantee than a method you can re-run yourself. Everything we publish stands on reproducible evidence: sources you can follow, tests you can replay, raw data you can check. So we do not invent author names, stock headshots, or padded credentials to look more trustworthy. We think that would make us less trustworthy, not more.

Our core commitment

We position Wibble aggressively — and we bind that positioning to evidence. The house rule is no test, no claim: every tested claim on this site must be backed by published, reproducible data, and claims we have not yet tested are stated as design intent, never as results. The full protocol lives in our public benchmark methodology, which we published before running any benchmark, so the results that follow are accountable to it.

How we hold ourselves to it

Three public policies codify the rule:

• Editorial policy — how content is produced, how competitor facts are sourced, and how we disclose that we review products we compete with.
• Testing methodology — how we test, what we publish alongside results, and what detector scores can and cannot prove.
• Corrections & updates — how errors get fixed, dated, and logged.

Contact

Questions, corrections, or anything else: support@wibbleai.com. We read everything.