About didaiwritethis.ai
didaiwritethis.ai checks text against AI watermark-detection APIs to help you tell whether it was generated by a given AI model. We started with Anthropic's Claude, and plan to add other labs (OpenAI, Google, xAI, and others) as they release public watermark-detection APIs of their own.
Current status
Anthropic has not yet released a public watermark-detection API. Until it does, results on this site are produced by a placeholder that mimics the expected response shape and is clearly labeled as a mock result in the UI. Nothing you submit right now is being checked against a real Anthropic watermark signal.
How it works
- Paste the text you want checked into the box.
- Enter your email when you run the check — we send a one-time sign-in link, no password. Your text is kept as you left it while you click through.
- We send that text to the relevant detector (Anthropic's, once available).
- We show you the verdict in plain language.
What happens to your text
- Your submitted text is sent to the detector for that one check and is not stored by us afterward — not in our database, not in logs.
- We do keep a lightweight history per check: the timestamp, the word count, and the verdict — never the text itself.
- We don't use submitted text to train models, and we don't sell or share it with anyone other than the detection provider needed to run the check.
What a verdict means
A watermark-detection result is a probabilistic signal from the AI lab that issued it, not a certainty. "Watermarked" means the detector found a strong signal consistent with that model's output. "Not watermarked" does not prove a human wrote the text — it only means no watermark signal was found, which can also happen with edited, paraphrased, or otherwise altered AI text. "Inconclusive" means the signal wasn't strong enough either way.
Word limits & pricing
Every account gets 5,000 words of checks, total, for free. Only the text you submit counts against that limit — the verdict we return does not. Once you use your free words, you'll be able to purchase more; pricing for that is still being finalized.
Keeping this free of bots
Anyone can paste text and see how the tool works, but a verified email is required before we run a check. Alongside that we use Cloudflare Turnstile plus per-IP request limits to keep the service usable for real people without runaway API costs from automated abuse.
How this site's writing is produced
The explainer guides on this site were drafted with AI assistance. Given that didaiwritethis.ai exists to help people reason about AI authorship, saying so plainly seemed like the only defensible option — a tool that asks you to care where text came from should be willing to answer that question about itself.
The claims and citations in those guides are meant to be checkable. Where a guide cites a study or an announcement, it links to the source so you can read it rather than take our word for it.
Frequently asked questions
How can you tell if AI wrote a piece of text?
There are two very different approaches. Conventional AI detectors guess from writing style — sentence length, word predictability, how 'average' the phrasing feels — and they are unreliable enough that many institutions have stopped trusting them. Watermark detection is different: some AI models embed a deliberate statistical signal into the text they generate, and a detector checks for that specific signal. When a watermark is present it is strong evidence the text came from that model. This site uses the watermark approach, not stylistic guessing.
What is an AI text watermark?
An AI text watermark is a hidden statistical pattern that a language model embeds in its output as it generates text. It does not change the meaning and is invisible when you read it, but a detector holding the right key can measure whether that pattern is present. It is closer to a signature woven into the text than to a visible label.
Does Claude watermark its output?
Anthropic has not publicly released a watermark detection API for Claude. This site is built ahead of that release: the interface, word tracking and result display all work today, but verdicts are currently generated from placeholder data and are clearly labelled as such. When Anthropic ships a real detection API, we swap it in behind the same interface and results become real.
Are AI detectors accurate?
Most conventional AI detectors are not reliable enough to accuse anyone based on their output. They produce false positives on non-native English writing and on formal or formulaic prose, and they can be defeated by light editing. Watermark detection is more dependable when a watermark exists, but it has a hard limit: it can only find watermarks in text produced by a model that watermarks its output. It cannot detect AI text in general.
Does 'no watermark detected' mean a human wrote it?
No. It only means no watermark signal was found. The text may have come from a model that does not watermark, from a watermarking model with the feature disabled, or from AI text that was paraphrased or edited enough to disturb the signal. A negative result is not proof of human authorship.
Do you store the text I paste?
No. Your text is sent to the detector for that single check and is not written to our database or logs. We keep only the word count, timestamp and verdict for each check, so you can see your history and we can track your free word balance. The text itself is never retained, never used for training, and never sold.
How much does it cost?
Every account gets 5,000 words of checks for free, counted across all your checks. Only the text you submit counts toward that limit — the verdict we return does not. Additional word packs will be available for purchase later; pricing has not been set yet.