AI Checkers, Image Detectors and Watermarking: What Actually Works in 2026

I keep coming back to four search numbers this week, because together they say more about where AI content actually is than any launch announcement.

"AI checker" gets roughly 1,500,000 searches a month, up 23% year over year. "AI image detector" gets roughly 110,000, up 232% in a year. "AI text watermarking" gets only about 1,000 a month — but that is up 3,233%, from around 10 searches in July 2026. And "AI text cleaner" sits at about 3,600 searches, up 650%.

Those numbers come from risingtrends.co's "Top 45 AI Trends (October 2026)", which I checked on October 28, 2026. I read them as one story with two sides. On one side, a lot of people are trying to work out whether content in front of them is real. On the other, a smaller, fast-growing group of us are cleaning up our own AI-assisted drafts before we publish them. Different jobs. Different tools. And a lot of confusion — and, frankly, a lot of dishonest marketing — in between.

Here is what I have concluded, including the parts that are inconvenient for people selling these tools.

The honest nuance in the checker numbers

The 1.5 million figure is real, but it needs its context: risingtrends also notes that "AI checker" searches are down over the past three months, even though they are up year over year. Text detection is cooling.

Image detection is not. It is still up 22% over the past three months, on top of that 232% yearly rise. That split matches my own experience of the internet right now. A suspicious paragraph starts an argument. A suspicious image can be a scam, a fake product photo, or a fabricated screenshot, and getting it wrong costs something immediately. The checking job is migrating toward images and provenance, and I think text-checker hype will keep settling.

What a text checker actually measures

This is the part I wish every teacher, editor, and manager would read before acting on a score.

A text checker was never in the room when the writing happened. It does not know how the text was produced. It measures style: how predictable the word choices are, how even the sentences feel, how closely the text matches patterns that are statistically common in model output.

Which means it false-positives on exactly the writers who can least afford it. Non-native English writers often write with a narrower, more careful vocabulary and more regular sentence patterns — because they are writing carefully in a second language. Formal writing — legal, academic, technical, corporate — is predictable and even on purpose. Both can look "AI-like" to a pattern-matcher, because the patterns genuinely overlap.

So my rule is absolute: never treat a checker score as a verdict. Not for a student. Not for an employee. Not for a freelancer you are deciding whether to pay. At most, it is a reason to look for real evidence — drafts, version history, the person talking you through their own work.

Image detectors: guesswork is losing to provenance

Image detectors come in two kinds, and confusing them is where most bad decisions start.

Pixel-guessing detectors analyse the image itself and estimate whether a generator made it. That is the image version of a text checker: a similarity judgement, a useful signal, weak proof, and it degrades as generators improve.

Provenance checks ask a better question: does this file carry a record of where it came from? Metadata, content credentials of the C2PA type, or a watermark signal embedded at generation — Google's SynthID, for images and other media, is the best-known built-in mark. A positive provenance signal from a trusted source is much stronger than a pixel guess, because it is a claim attached by the tool that made the file, not an inference from how it looks.

But provenance is fragile, and you should know that before relying on it. Screenshots capture the pixels and leave most metadata behind. Re-encoding and compression — uploading to a social platform, converting formats — can strip metadata and weaken embedded signals. And a missing signal proves nothing: the image might be human-made, or made by a tool that adds no mark, or the mark may simply have been lost in transit.

Watermarking: marks built in, not guessed afterwards

Watermarking is the other side of detection. Instead of guessing after the fact, the generator leaves a mark at the moment of creation.

OpenAI's text approach — called textGrain in coverage of the rollout — statistically alters word choice at generation time. As the model writes, its choices are nudged in a patterned way that reads normally to you and me but forms a detectable signal across enough text. The rollout is for ChatGPT and Codex in the EU from October 2026, under the EU AI Act's marking rules. The surge in searches followed Anthropic and OpenAI both shipping marks, and the coverage also points to a December 2026 grace-period deadline for systems that existed before August 2026.

Two limits matter more than the headline, and the coverage states them plainly:

• Detectors for the mark are initially limited to researchers and approved organisations. You and I generally cannot run a suspect paragraph through an official watermark check today.
• The mark is statistical, not a lock. Short texts carry too little signal. Translations and heavy edits weaken it. Finding nothing on a short, heavily edited passage has not cleared it.

Built-in marks are evidence about origin. After-the-fact detectors are guesses about style. When both exist, I believe the mark. When only the guess exists, I hold it lightly.

Text cleaning — and the line I will not cross

Cleaners are the smallest part of this cluster and the fastest growing, and honestly, I understand the appeal — I use AI help on drafts myself. Text copied out of a chat model carries invisible baggage: zero-width and other invisible characters, smart quotes, em dashes, odd spacing. A cleaner strips and normalises that. For your own draft, that is genuinely useful. Your text should not carry stray invisible characters into your CMS or your newsletter tool.

But here is the line, and I want it in plain words: cleaning your own draft is tidying, not disguise — it does not beat a detector, and I will not sell it as that. A cleaner changes characters and punctuation. A detector measures style patterns in the words themselves. Different layers. Anyone marketing a cleaner as a way to "pass" a check is making a claim the mechanics do not support.

The five-minute test I would run before trusting any checker

Full disclosure on my own process: I planned a hands-on pass with publicly reachable, no-login checkers for this piece, and those checkers could not be run from my research environment. So this is researched analysis, not a test report, and I will update the full guide when my own hands-on pass is done.

In the meantime, here is the test I would genuinely recommend — it takes five minutes and it teaches you more than any marketing page:

1. Take a paragraph you wrote yourself years ago, a fresh AI paragraph, and a heavily edited AI paragraph.
2. Run all three through two different checkers.
3. Compare.

Most people who do this once stop treating scores as verdicts. You will likely watch the human paragraph get flagged, or the edited one sail through, or the two tools flatly contradict each other. On samples where you already know the truth, that instability is the real review.

When I check something suspicious myself, my order is: source first (who published it, where did it first appear, who is accountable for it), provenance second for images, a detector score last and lightly — and for a person's own work, drafts and a conversation beat every tool I know.

What I actually trust

Strongest to weakest: a verified built-in mark or intact provenance credentials; then the source itself; then my own comparison test on samples where I know the truth; and last, by a long way, a single detector score on a single item. That score is a prompt to look closer. It is not proof, and it cannot be made into proof by putting a percentage sign after it.

I wrote the full guide — with the workflow for checking a suspicious image or text step by step, and for cleaning your own draft before publishing — over on ToolsDig: AI Checkers, Image Detectors and Watermarking: What Actually Works in 2026. And if you want the tidying tool itself, the cleaner I mentioned is free, no signup: AI Text Cleaner. Use it to make your drafts clean. Just do not let anyone — including me — tell you it makes them undetectable.